4

Sensors, Vision and

Signal Conditioning
Sensors in robots are like our nose, ears, mouth, and skin, whereas vision or robot
vision can be thought of as our eyes. Based on the function of human organs, for
example, eyes, skin, etc., terminology like vision, tactile, etc., have cropped up in
robot sensors. Robots, like humans, must gather extensive information about their
environment in order to function effectively. They must pick up an object and know
it has been picked up. As the robot arm moves through the 3-dimensional Cartesian
space, it must avoid obstacles and approach items to be handled at a controlled speed.
Some objects are heavy, others are fragile, and others are too hot to handle. These
characteristics of objects and the environment must be recognized, and fed to the
computer that controls a robot’s movement. For example, to move the end-effector
of a robot, with or without a payload along a desired trajectory and to exert a desired
force on an object, the end-effector and sensors (normally located at the joints and at
the end-effector) work in coordination with the robot controller (a microprocessor, or
a computer or a microcontroller).
Note that in industrial applications of robots, as explained in Chapter 2, sensors
must play important roles. For example, they must at least provide the following
functions:
1. Safe Operation Sensors must protect human workers who work in the vicinity
of the robot or other equipment. For example, one can provide a sensor on the floor
of a work cell where a robot is working so that if anybody steps in the robot’s power
should be switched off.
2. Interlocking This is required to coordinate the sequence of operations on a
component. For example, unless a turning is done on a component, it should not be
transferred to the conveyor.
3. Inspection This is essential for Sensor vs. Transducer
quality control. For example, one can Sensor and transducer are used
use a vision system to measure the interchangeably to denote a sensor-
transducer unit. They are the functional
length of a component to check if it is stages of sensing. First, a measurand
within the acceptable tolerance or not. is felt or sensed before it is transduced
or converted from one type of energy to
4. Part Configuration If a robot is another for various purposes including the
used to weld two parts of a car body, measurement of physical parameters like
the sensors must identify the correct position, velocity, etc.
 Sensors, Vision and Signal Conditioning 77

configurations, i.e., position and orientation of the parts, before the robot starts
welding.
There could be other scenarios like identifying the color code of a particular car
model before painting with that color is done by the robot, etc.

4.1 SENSOR CLASSIFICATION

The major capabilities required by a robot are as follows:
Simple Touch The presence or absence of an object.
Taction or Complex Touch The presence of an object plus some information on
its size and shape.
Simple Force Measured force along a single axis.
Complex Force Measured force along two or more axes.
Proximity Noncontact detection of an object.
Simple Vision Detection of edges, holes, corners, and so on.
Complex Vision Recognition of shapes.
Based on the type of signals a sensor or transducer receives and processes, it can
be classified as analog or digital. In analog sensors, with the variation of input there
is a continuous variation of output, whereas in case of digital sensors, the output is
of digital or discrete nature. For example, potentiometers, tacho-generators located
at the joints and strain-gauge-based sensors located at the end-effector of a robot fall
in the category of analog sensors, whereas encoders, located at the robot’s joints,
are digital sensors. In this book, sensors are, however, classified based on what they
sense, i.e., internal or external state of the robots, etc., as shown Fig. 4.1.

4.2 INTERNAL SENSORS

Internal sensors, as the name suggests, are used to measure internal state of a robot,
i.e., its position, velocity, acceleration, etc., at a particular instant. Based on these
information, control command is decided by the controller. Depending on the
quantities it measures, a sensor is termed as the position, velocity, acceleration, or
force sensor.
4.2.1 Position Sensors
Position sensors measure the position of each joint, i.e., joint angle of a robot. From
these joint angles, one can find the end-effector configuration, namely, its position
and orientation, through forward kinematics which will be taken up in Chapter 6.
Different position sensors are explained next.
1. Encoder The encoder is a digital optical device that converts motion into a
sequence of digital pulses. By counting a single bit or by decoding a set of bits, the
pulses can be converted to relative or absolute measurements. Thus, encoders are of
incremental or absolute type. Further, each type may be again linear and rotary.
 78
Introduction to Robotics

Fig. 4.1 Classification of sensors

 Sensors, Vision and Signal Conditioning 79

Incremental Linear Encoder As shown in Fig. 4.2(a), it has a transparent glass

scale with opaque grating. The thickness of grating lines and the gap between them
is made same, which are in the range of microns. One side of the scale is provided
with a light source and a condenser lens. On the other side there are light-sensitive
cells. The resistance of the cells (photodiodes) decreases whenever a beam of light
falls on them. Thus, a pulse is generated each time a beam of light is intersected by
the opaque line. This pulse is fed to the controller, which updates a counter (a record
of the distance traveled).
Absolute Linear Encoder It is similar in principle as the incremental linear
encoder. The difference is that it gives absolute value of the distance covered at
any time. Thus, the chance of missing the pulses at high speeds is less. The output
is digital in this case. The scale is marked in a sequence of opaque and transparent
strips, as shown in Fig. 4.2(b). In the scale shown, if the opaque block represents
1 (one) and the transparent block as 0 (zero) then the leftmost column will show a
binary number as 00000, i.e., a decimal value of 0, and the rightmost column will
show a binary number 11111, i.e., a decimal value of 61.
Incremental Rotary Encoder It is similar to the linear incremental encoder with a
difference that the gratings are now on a circular disc, as in Fig. 4.2(c). The common
value of the width of transparent spaces is 20 microns. There are two sets of grating
lines on two different circles which detect direction of rotation, and one can also
enhance the accuracy of the sensor. There is another circle, which contains only one
grating mark. It is used for measurement of full circles.
Absolute Rotary Encoder Similar to the absolute linear encoder, the circular disk
is divided into a number of circular strips and each strip has definite arc segments,
as shown in Fig. 4.2(d). This sensor directly gives the digital output (absolute). The
encoder is directly mounted on the motor shaft or with some gearing to enhance the
accuracy of measurement. To avoid noise in this encoder, a gray scale is sometimes
used. A Gray code, unlike binary codes, allows only one of the binary bits in a code
sequence to change between radial lines. It prevents confusing changes in the binary
output of the absolute encoder when the encoder oscillates between points. A sample
Gray code is given in Table 4.1 for some numbers. Note the difference between the
Gray and binary codes. The basic arrangement of the rotary encoder is shown in Fig.
4.2(e).

Table 4.1 Sample Gray codes

Decimal Binary code Gray code

0 0000 0000
1 0001 0001
2 0010 0011
3 0011 0010
4 0100 0110
 80 Introduction to Robotics

Fig. 4.2 Encoders

Example 4.1 Angular Position of an Encoder

If maximum count possible is p pulses and the range of a rotary encoder is ±qm then
the angular position corresponding to a count of n pulses is given as follows:
n
q= q (4.1)
p m

2. Potentiometer A potentiometer, also referred as simply pot, is a variable-

resistance device that expresses linear or angular displacements in terms of voltage,
as shown in Figs. 4.3(a-b), respectively. It consists of a wiper that makes contact with
a resistive element, and as this point of contact moves, the resistance between the
 Sensors, Vision and Signal Conditioning 81

wiper and end leads of the device changes in proportion to the displacement, x and q
for linear and angular potentiometers, respectively.

Fig. 4.3 Potentiometers

3. LVDT The Linear Variable Differential Transformer (LVDT) is one of the most
used displacement transducers, particularly when high accuracy is needed. It generates
an ac signal whose magnitude is related to the displacement of a moving core, as
indicated in Fig. 4.4. The basic concept is that of a ferrous core moving in a magnetic
field, the field being produced in a manner similar to that of a standard transformer.
There is a central core surrounded by two identical secondary coils and a primary
coil, as shown in Fig. 4.4. As the core changes position with respect to the coils, it
changes the magnetic field, and hence the voltage amplitude in the secondary coil
changes as a linear function of the core displacement over a considerable segment. A
Rotary Variable Differential Transformer (RVDT) operates under the same principle
as the LVDT is also available with a range of approximately ±40°.

Fig. 4.4 Principle of LVDT

 82 Introduction to Robotics

4. Synchros and Resolver While encoders give digital output, synchros and
resolvers provide analog signal as their
What are arc-minute
output. They consist of a rotating shaft
and arc-second?
(rotor) and a stationary housing (stator).
Their signals must be converted into the They are the measures of small angles.
one degree = 60 arc-minutes and
digital form through an analog-to-digital
one arc-minute = 60 arc-seconds.
converter before the signal is fed to the
computer.
As illustrated in Fig. 4.5, synchros and resolvers employ single-winding rotors
that revolve inside fixed stators. In a simple synchro, the stator has three windings
oriented 120° apart and electrically connected in a Y-connection. Resolvers differ
from synchros in that their stators have only two windings oriented at 90°. Because
synchros have three stator coils in a 120° orientation, they are more difficult than
resolvers to manufacture and are, therefore, more costly.
Modern resolvers, in contrast, are available in a brushless form that employ a
transformer to couple the rotor signals from the stator to the rotor. The primary
winding of this transformer resides on the stator, and the secondary on the rotor.
Other resolvers use more traditional brushes or slip rings to couple the signal into the
rotor winding. Brushless resolvers are more rugged than synchros because there are
no brushes to break or dislodge, and the life of a brushless resolver is limited only by
its bearings. Most resolvers are specified to work over 2 V to 40 V rms (root mean
square) and at frequencies from 400 Hz to 10 kHz. Angular accuracies range from 5
arc-minutes to 0.5 arc-minutes.

Fig. 4.5 Synchros and resolvers

In operation, synchros and resolvers resemble rotating transformers. The rotor

winding is excited by an ac reference voltage, at frequencies up to a few kHz. The
magnitude of the voltage induced in any stator winding is proportional to the sine
 Sensors, Vision and Signal Conditioning 83

of the angle q between the rotor-coil axis and the stator-coil axis. In the case of a
synchro, the voltage induced across any pair of stator terminals will be the vector
sum of the voltages across the two connected coils. For example, if the rotor of a
synchro is excited with a reference voltage, V sin (w t), across its terminal R1 and R2,
the stator’s terminal will see voltages denoted as V0 in the form:
V0(S1 − S3) = V sin (w t) sin q (4.2a)
V0(S3 − S2) = V sin (w t) sin (q + 120°) (4.2b)
V0(S2 − S1) = V sin (w t) sin (q + 240°) (4.2c)
where S1, S2, etc., denotes the stator terminals. Moreover, V and w are the input
amplitude and frequency, respectively, whereas q is the shaft angle. In the case of a
resolver, with a rotor ac reference voltage of V sin (w t), the stator’s terminal voltages
will be
V0(S1 − S3) = V sin (w t) sin q (4.3a)
V0(S4 − S2) = V sin (w t) sin (q + 90°) = V sin (w t) cos q (4.3b)
As said earlier, the output of these synchros and resolvers must be first digitized.
To do this, analog-to-digital converters are used. These are typically 8-bit or 16-
bit. An 8-bit means that the whole range of analog signals will be converted into a
maximum of 28 = 256 values.
4.2.2 Velocity Sensors
Velocity or speed sensors measure by taking consecutive position measurements at
known time intervals and computing the time rate of change of the position values or
directly finding it based on different principles.
1. All Position Sensors Basically, all position sensors when used with certain
time bounds can give velocity, e.g., the number of pulses given by an incremental
position encoder divided by the time consumed in doing so. But this scheme puts
some computational load on the controller which may be busy in some other
computations.
2. Tachometer Such sensors can directly find the velocity at any instant of time,
and without much of computational load. This measures the speed of rotation of
an element. There are various types of tachometers in use but a simpler design is
based on the Fleming’s rule, which states ‘the voltage produced is proportional to the
rate of flux linkage.’ Here, a conductor (basically a coil) is attached to the rotating
element which rotates in a magnetic field (stator). As the speed of the shaft increases,
the voltage produced at the coil terminals also increases. In other ways, as shown in
Fig. 4.6, one can put a magnet on the rotating shaft and a coil on the stator. The
voltage produced is proportional to the speed of rotation of the shaft. This information
is digitized using an analog-to-digital converter and passed on to the computer.
3. Hall-effect Sensor Another velocity-measuring device is the Hall-effect
sensor, whose principle is described next. If a flat piece of conductor material, called
Hall chip, is attached to a potential difference on its two opposite faces, as indicated
in Fig. 4.7 then the voltage across the perpendicular faces is zero. But if a magnetic
field is imposed at right angles to the conductor, the voltage is generated on the two
 84 Introduction to Robotics

other perpendicular faces. Higher the field value, higher the voltage level. If one
provides a ring magnet, the voltage produced is proportional to the speed of rotation
of the magnet.

Fig. 4.6 Schematic diagram of a tachometer

Fig. 4.7 Principle of Hall-effect sensor

4.2.3 Acceleration Sensors

Similar to measurements of velocity Why acceleration from force
from the information of position sensors, but not from velocity?
one can find the accelerations as the time
rate of change of velocities obtained Finding acceleration from force involves
integration which suppresses any noise in
from velocity sensors or calculated the force signal. However, if the velocity
from the position information. But signal is used to determine acceleration,
this is not an efficient way to calculate differentiation has to be performed which
the acceleration because this will put amplifies the noise in the velocity signal.
a heavy computational load on the Hence, the latter is never advisable.
computer and that can hamper the speed
of operation of the system. Another way to compute the acceleration is to measure
the force which is the result of mass times acceleration. Forces are measured, for
example, using strain gauges for which the formula is
DRAE
F= (4.4)
RG
 Sensors, Vision and Signal Conditioning 85

where F is force, DR is the change in resistance of the strain gauge, A is the cross-
sectional area of the member on which the force being applied, E is the elastic
modulus of the strain-gauge material, R is the original resistance of the gauge, and
G is gauge factor of the strain gauge. Then, the acceleration a is the force divided by
mass of the accelerating object m, i.e.,
F DRAE
a= = (4.5)
m RCm
It is pointed out here that the velocities
What is Gauge Factor?
and accelerations that are measured using
It is a measure of sensitivity for the strain
position sensors require differentiations.
gauges, and defined by
It is generally not desirable, as the 1 DR
noise in the measured data, if any, will G=
e R
be amplified. Alternatively, the use of
where G is the gauge factor, and e is
integrators to obtain the velocity from strain.
the acceleration, and consequently the
position, are recommended. Integrators tend to suppress the noise.
Example 4.2 Change in Resistance
If the gauge factor G = 2, resistance of the unreformed wire R = 100 W, and strain
e = 10–6, then change in resistance is given by
DR = GeR = 2 × 10–6 × 100 = 0.0002 W (4.6)

4.2.4 Force Sensors

A spring balance is an example of a force sensor in which a force, namely, the weight,
is applied to the scale pan that causes displacement, i.e., the spring stretches. The
displacement is then a measure of the force. There exist other types of force sensors,
e.g., strain-gauge based, Hall-effect sensor, etc.
1. Strain-gauge Based The principle of this type of sensors is that the elongation
of a conductor increases its resistance. Typical resistances for strain gauges are 50–
100 ohms. The increase in resistance is due to
∑ Increase in the length of the conductor; and
∑ Decrease in the cross-section area of the conductor.
Strain gauges are made of electrical conductors, usually wire or foil, etched on a
base material, as shown in Fig. 4.8. They are glued on the surfaces where strains are
to be measured, e.g., R1 and R2 of Fig. 4.9(a). The strains cause changes in the
resistances of the strain gauges, which are measured by attaching them to the
Wheatstone bridge circuit as one of the four resistances, R1 . . . R4 of Fig. 4.9(b). The
principle of a Wheatstone bridge is explained in Section
4.5.5. It is a cheap and accurate method of measuring
strain. But care should be taken for the temperature
changes. In order to enhance the output voltage and cancel
away the resistance changes due to the change in
temperature, two strain gauges are used, as shown in Fig.
4.9(a), to measure the force at the end of the cantilever
beam. Fig. 4.8 Strain gauges
 86 Introduction to Robotics

Fig. 4.9 Strain measurement

2. Piezoelectric Based A piezoelectric material exhibits a phenomenon known

as the piezoelectric effect. This effect states that when asymmetrical, elastic crystals
are deformed by a force, an electrical potential will be developed within the distorted
crystal lattice, as illustrated in Fig. 4.10. This effect is reversible. That is, if a potential
is applied between the surfaces of the
crystal, it will change its physical
dimensions. The magnitude and polarity of
the induced charges are proportional to the
magnitude and direction of the applied
force. The piezoelectric materials are
quartz, tourmaline, Rochelle salt, and
others. The range of forces that can be
measured using piezoelectric sensors are
from 1 to 20 kN and at a ratio of 2 × 105.
These sensors can be used to measure an Fig. 4.10 Piezoelectric sensor
instantaneous change in force (dynamic
forces).
3. Current Based Since the torque provided by an electric motor is a function
of the current drawn, its measurement, along with the known motor characteristics,
gives the torque sensing.
Example 4.3 Force and Acceleration using Strain Gauges
If the strain gauge of Example 4.2 is used to measure applied force, F on a member
of cross-sectional area, A = 10 mm2, and Young’s modulus, E = 6.9×10-10 N/m2 (for
aluminium), then using Eq. (4.4)
AE DR (10 ¥ 10 -6 )(6.9 ¥ 1010 )(0.0002)
F= = = 0.69 N (4.7)
GR 2(100)
If the mass of the member m = 3 kg then using equations (4.5) and (4.7)
0.69
a= = 0.21 m/s2 (4.8)
3
 Sensors, Vision and Signal Conditioning 87

4.3 EXTERNAL SENSORS

External sensors are primarily used to learn more about a robot’s environment,
especially the objects being manipulated. External sensors can be divided into the
following categories:
∑ Contact type, and
∑ Noncontact type.
4.3.1 Contact Type
In this section, a contact-type force sensor is explained.
Limit Switch A limit switch is constructed
much as the ordinary light switch used at homes
and offices. It has the same on-off characteristics.
The limit switch usually has a pressure-sensitive
mechanical arm, as shown in Fig. 4.11(a). When
an object applies pressure on the mechanical arm,
the switch is energized. An object might have an
attached magnet that causes a contact to rise and
close when the object passes over the arm. As
shown in Fig. 4.11(b), the pull-up register keeps
the signal at +V until the switch closes, sending Fig. 4.11 Limit switch
the signal to ground. Limit switches can be either
Normally Open (NO) or Normally Closed (NC), and may have multiple-poles. A
normally open switch has continuity when pressure is applied. A single-pole switch
allows one circuit to be opened or closed upon contact, whereas a multi-pole switch
allows multiple switch circuits to be open or closed. Limit switches are mechanical
devices which have problems like
∑ they are subjected to mechanical failure,
∑ their mean time between failures is low compared to noncontact sensors, and
∑ the speed of operation is relatively slow compared to the speed of switching of
photoelectric micro-sensors which is up to 3000 times faster.
Limit switches are used in robots to detect the extreme positions of the motions,
where the link reaching an extreme position switches off the corresponding actuator,
thus, safeguarding any possible damage to the mechanical structure of the robot arm.
4.3.2 Noncontact Type
Here, noncontact-type force sensors and their principles are presented.
1. Proximity Sensor Proximity sensing is the technique of detecting the presence
or absence of an object with an electronic noncontact-type sensor. Proximity sensors
are of two types, inductive and capacitive. Inductive proximity sensors are used in
place of limit switches for noncontact sensing of metallic objects, whereas capacitive
proximity sensors are used on the same basis as inductive proximity sensors.
However, these can also detect nonmetallic objects.
Inductive Proximity Sensor All inductive proximity sensors consist of four basic
elements, namely, the following:
 88 Introduction to Robotics

∑ Sensor coil and ferrite core

∑ Oscillator circuit
∑ Detector circuit
∑ Solid-state output circuit
As shown in Fig. 4.12, the oscillator circuit
generates a radio-frequency electromagnetic field.
The field is centred around the axis of the ferrite
core, which shapes the field and directs it at the
sensor face. When a metal target approaches the
face and enters the field, eddy currents are induced
into the surface of the target. This results in a
loading or damping effect that causes a reduction
in amplitude of the oscillator signal. The detector
circuit detects the change in the oscillator
amplitude. The detector circuit will ‘switch on’ at
specific operating amplitude. This signal ‘turns
on’ the solid-state output circuit. This is often
referred to as damped condition. As the target
leaves the sensing field, the oscillator responds
with an increase in amplitude. As the amplitude
increases above a specific value, it is detected by
the detector circuit, which is ‘switched off’ causing
the output signal to return to the normal or ‘off’
state. The sensing range of an inductive proximity
sensor refers to the distance between the sensor
face and the target. It also indicates the shape of
the sensing field generated through the coil and the
core. There are several mechanical and
environmental factors that affect the sensing range.
The usual range is up to 10–15 mm but some Fig. 4.12 Inductive proximity
sensors have ranges as high as 100 mm. sensor

Capacitive Proximity Sensor A capacitive proximity sensor operates much like

an inductive proximity sensor. However, the means of sensing is considerably
different. Capacitive sensing is based on dielectric capacitance. Capacitance is the
property of insulators to store the charge. A capacitor consists of two plates separated
by an insulator, usually called a dielectric.
When the switch is closed, a charge is stored on
the two plates. The distance between the plates
determines the ability of the capacitor to store
the charge and can be calibrated as a function of
stored charge to determine discrete ON and
OFF switching status. Figure 4.13 illustrates
the principle of a capacitive sensor. One
capacitive plate is part of the switch, the sensor
face is the insulator, and the target is the other Fig. 4.13 Principle of capacitive
plate. Ground is the common path. The sensors
 Sensors, Vision and Signal Conditioning 89

capacitive switch has the same four elements as the inductive sensor, i.e., sensor (the
dielectric media), oscillator circuit, detector circuit, and solid-state output circuit.
The oscillator circuit in a capacitive switch operates like one in an inductive
switch. The oscillator circuit includes capacitance from the external target plate
and the internal plate. In a capacitive sensor, the oscillator starts oscillating when
sufficient feedback capacitance is detected. Major characteristics of the capacitive
proximity sensors are as follows:
∑ They can detect non-metallic targets.
∑ They can detect lightweight or small objects that cannot be detected by
mechanical limit switches.
∑ They provide a high switching rate for rapid response in object counting
applications.
∑ They can detect limit targets through nonmetallic barriers (glass, plastics, etc.).
∑ They have long operational life with a virtually unlimited number of operating
cycles.
∑ The solid-state output provides a bounce-free contact signal.
Capacitive proximity sensors have two major limitations.
∑ The sensors are affected by moisture and humidity, and
∑ They must have extended range for effective sensing.
Capacitive proximity sensors have a greater sensing range than inductive
proximity sensors. Sensing distance for capacitive switches is a matter of plate
area, as coil size is for inductive proximity sensors. Capacitive sensors basically
measure a dielectric gap. Accordingly, it is desirable to be able to compensate for
the target and application conditions with a sensitivity adjustment for the sensing
range. Most capacitive proximity sensors are equipped with a sensitivity adjustment
potentiometer.
2. Semiconductor Displacement Sensor As shown in Fig. 4.14, a
semiconductor displacement sensor uses a semiconductor Light Emitting Diode
(LED) or laser as a light source, and a Position-Sensitive Detector (PSD). The laser
[Courtesy: http://www.sensorcentral.com/displacement/
laser02.php]

Fig. 4.14 Semiconductor-based sensor

 90 Introduction to Robotics

beam is focused on the target by a lens. The target reflects the beam, which is then
focused on to the PSD forming a beam spot. The beam spot moves on the PSD as
the target moves. The displacement of the workpiece can then be determined by
detecting the movement of the beam spot.

4.4 VISION
Vision can be defined as the task of
extracting information about the external Computer Vision vs. Computer Graphics
world from light rays imaged by a camera Computer vision can be thought of as
or an eye. Vision, also referred in the ‘inverse computer graphics.’ Computer
literature as computer vision or machine graphics deals with how to generate images
from a specification of the visual scene (e.g.,
vision or robot vision, is a major subject objects, scene structures, light sources),
of research and many textbooks, e.g., by whereas computer vision inverts this process
Haralick and Shapiro (1992, 1993), and to infer the structure of the world from the
others. A good coverage on the topic has observed image(s).
also appeared in Niku (2001). There are
also dedicated journals, e.g., Computer Vision, Graphics, and Image Processing, and
conferences in the area of robot vision. The area is so vast that it cannot be covered
in one section or chapter of a book. However, an attempt is made here to introduce
the basic concepts and techniques so that one is able to understand the systems and
methodologies used in robot vision. For detailed study and research, other references
in the area should be consulted.
Note in Fig. 4.1 that the vision systems or vision sensors are classified as external
noncontact type. They are used by robots to let them look around and find the parts,
for example, picking and placing them at appropriate locations. Earlier, fixtures
were used with robots for accurate positioning of the parts. Such fixtures are very
expensive. A vision system can provide alternative economic solution. Other tasks
of vision systems used with robots include the following:
1. Inspection Checking for gross surface defects, discovery of flaws in labeling,
verification of the presence of components in assembly, measuring for dimensional
accuracy, checking the presence of holes and other features in a part.
2. Identification Here, the purpose is to recognize and classify an object rather
than to inspect it. Inspection implies that the part must be either accepted or rejected.
3. Visual Servoing and Navigation Control The purpose here is to direct the
actions of the robot based on its visual inputs, for example, to control the trajectory
of the robot’s end-effector toward an object in the workspace. Industrial applications
of visual servoing are part positioning, retrieving parts moving along a conveyor,
seam tracking in continuous arc welding, etc.
All of the above applications someway require the determination of the
configuration of the objects, motion of the objects, reconstruction of the 3D geometry
of the objects from their 2D images for measurements, and building the maps of the
environments for a robot’s navigation. Coverage of vision system is from a few
millimetres to tens of metres with either narrow or wide angles, depending upon the
 Sensors, Vision and Signal Conditioning 91

system needs and design. Figure 4.15 shows a typical visual system connected to an
industrial robot.

[Courtesy: PAR Lab., IIT Delhi]

Fig. 4.15 Hardware components of a vision system

4.4.1 Elements in a Vision Sensor

In vision systems, the principal imaging component is a complete camera including
sensing array, associated electronics, output signal format, and lens, as shown in
Fig. 4.16.

Fig. 4.16 Principal of a camera or vision system [Courtesy: Siciliano, 2010]

The task of the camera as a vision sensor is to measure the intensity of the light
reflected by an object, as indicated in Fig. 4.16, using a photosensitive element
termed pixel (or photosite). A pixel is capable of transforming light energy into
electric energy. The sensors of different types like CCD, CMOS, etc., are available
depending on the physical principle exploited to realize the energy transformation.
Depending on the application, the camera could be RS-170/CCIR, NTSC/PAL (These
are American RS-170 monocolor, European/Indian CCIR monocolor, NTSC color,
PAL color television standard signal produced by the video cameras, respectively)
progressive scan, variable scan, or line scan. Five major system parameters which
govern the choice of camera are field of view, resolution, working distance, depth of
field, and image data acquisition rate. As a rule of thumb, for size measurement, the
sensor should have a number of pixels at least twice the ratio of the largest to smallest
object sizes of interest.
 92 Introduction to Robotics

1. Camera Systems As indicated in Fig. 4.16, a camera is a complex system

comprising of several devices inside it. Other than the photosensitive sensor, there
are shutter, a lens, and analog preprocessing electronics. The lens is responsible
for focusing the light reflected by the object on the plane where the photosensitive
sensors lies, called the image plane. In order to use it to compute the position and/
or orientation of an object, the associated coordinate transformations, etc., need
to be performed using the knowledge of Chapter 5, i.e., Transformation. This is
generally carried out by a software residing inside a personal computer which saves
the images. Specific calculations related to camera calibration, etc., can be found in
Haralick and Shapiro (1993), Siciliano et al. (2011), and others. They are not covered
in this introductory book on robotics.
Note that there are two types of video cameras: analog and digital. Analog
cameras are not in common anymore. However, if it is used, a frame grabber or
video capture card, usually a special analog-to-digital converter adopted for video
signal acquisition in the form of a plug-in board which is installed in the computer,
is often required to interface the camera to a host computer. The frame grabber will
store the image data from the camera on-board, or system memory, and performs
sampling and digitizing of the analog data as necessary. In some cases, the camera
may output digital data, which is compatible with a standard computer. So a separate
frame grabber may not be needed. Vision software is needed to create the program
which processes the image data. When an image has been analyzed, the system must
be able to communicate the result to control the process or to pass information to a
database. This requires a digital input/output interface. The human eye and brain can
identify objects and interpret scenes under a wide variety of conditions. Robot-vision
systems are far less versatile. So the creation of a successful system requires careful
consideration of all elements of the system and precise identification of the goals to
be accomplished, which should be kept as simple as possible.
Vidicon Camera Early vision systems employed vidicon cameras, which were
bulky vacuum tube devices. They are almost extinct today but explained here for the
sake of completeness in the development of video cameras. Vidicons are also more
sensitive to electromagnetic noise interference and require high power. Their chief
advantages are higher resolution and better light sensitivity. Figure 4.17 shows the
schematic diagram of a vidicon camera.
The mosaic reacts to the varying intensity of a light by varying its resistance.
Now, as the electric gun generates and sends a continuous cathode beam to the
mosaic passing though two pairs of orthogonal capacitors (deflectors), the electron
beam gets deflected up or down, and left or right based on the charge on each pair of
capacitors. As the beam scans the image, at each instant, the output is proportional to
the resistance of the mosaic or the light intensity on the mosaic. By reading the output
voltage continuously, an analog representation of the image can be obtained. Please
note that the analog signal of vidicon needs to be converted to digital signal using
analog-to-digital converters (ADC), as mentioned in Section 2.1.2 in order to process
the image further using a PC. The ADC which actually performs the digitization
of the analog signal requires mainly three steps, i.e., sampling, quantization, and
encoding.
 Sensors, Vision and Signal Conditioning 93

Fig. 4.17 Schematic diagram of a vidicon camera [Courtesy: Niku (2001)]

In sampling, a given analog signal is sampled periodically to obtain a series of

discrete-time analog signal, as
illustrated in Fig. 4.18. By setting a
Analog Signal
specified sampling rate, the analog
Voltage

signal can be approximated by the

sampled digital outputs. However,
while reconstructing the original
signal from the sample data, one may Time
Time
end up with a completely different
signal. This loss of information is Sampled Output
Voltage

called aliasing, and it can be a serious Signal

problem. In order to prevent aliasing,
according to the sampling theorem,
the sampling rate must be at least Time
twice the largest frequency in the
original video signal if one wishes to Fig. 4.18 Sampling and digitalization
reconstruct that signal exactly.
In quantization, each sampled discrete time voltage level is assigned to a finite
number of defined amplitude levels. These levels correspond to the Gray scale used
in the system. The predefined amplitude levels are characteristics to a particular ADC
and consist of a set of discrete values of voltage levels. The number of quantization
levels is defined by 2n, where n is the number of bits of the ADC. For example, a 1-bit
ADC will quantize only at two values, whereas with an 8-bit ADC, it is possible to
quantize at 28 = 256 different values. Note that a large number of bits enables a signal
to be represented more precisely. Moreover, sampling and quantization resolutions
are completely independent of each other.
Finally, encoding does the job of converting the amplitude levels that are quantized
into digital codes, i.e., 0 or 1. The ability of the encoding process to distinguish
between various amplitude levels is a function of the spacing of each quantization
level.
 94 Introduction to Robotics

Example 4.4 Quantization

An 8-bit ADC is to be used to represent a range of video signal between 0 and 5 V.
The resolution or quantization spacing is given by
5
= 0.0195 V (4.9)
28
Accordingly, quantization error can be given by,
0.0195
± = 0.00975 V (4.10)
2

Digital Camera A digital camera is based on solid-state technology. The main

part of these cameras is a solid-state silicon wafer image area that has hundreds of
thousands of extremely small photosensitive areas called photsites printed on it. Each
small area of the wafer is a pixel. As the
image is projected onto the image area, at What is a Pixel?
each pixel location of the wafer, a charge A pixel is a short form for picture element.
is developed that is proportional to the It is a single point in a graphic image.
intensity of the light at that location.
Thus, a digital camera is also called a Charged Coupled Device (CCD) camera or
Charge Integrated Device (CID) camera. The collection of charges, as shown in
Fig. 4.19, if read sequentially, would be a representation of the image pixels. The
output is a discrete representation of the image as a voltage sampled in time. Solid-
state cameras are smaller, more rugged, last longer, and have less inherent image
distortion than vidicon cameras. They are also slightly more costly, but prices are
coming down.

Fig. 4.19 Basic principle of a CCD device (Image acquisition)

Both the CCDs and CID chips use large transfer techniques to capture an image.
In a CCD camera, light impinges on the optical equivalent of a Random Access
Memory (RAM) chip. The light is absorbed in a silicon substrate, with charge build-
up proportional to the amount of light reaching the array. Once sufficient amount of
energy has been received to provide a picture, the charges are read out through built-
in control registers. Some CCD chips use an interline charge-transfer technique.
Others use frame-transfer approach, which is more flexible for varying the integration
period.
The CID camera works on a similar principle. A CID chip is a Metal Oxide
Semiconductor (MOS) based device with multiple gates similar to CCDs. The video
signal is the result of a current pulse from a recombination of carriers. CIDs produce
 Sensors, Vision and Signal Conditioning 95

a better image (less distortion) and use a different read-out technique than CCDs
which require a separate scanning address unit. CIDs are, therefore, more expensive
than CCDs. The principle difference between a CCD and a CID camera is the method
of generating the video signal.
2. Lighting Techniques One of the key questions in robot vision is what
determines how bright the image of some surface on the object will be? It involves
radiometry (measurement of the flow and transfer of radiant energy), general
illumination models, and surface having both diffuse and specular reflection
components. Different points on the objects in front of the imaging system will
have different intensity values on the image, depending on the amount of incident
radiance, how they are illuminated, how they reflect light, how the reflected light is
collected by a lens system, and how the sensor camera responds to the incoming light.
Figure 4.20 shows the basic reflection phenomenon. Hence, proper illumination of
the scene is important. It also affects the complexity level of the image-processing
algorithm required. The lighting techniques must avoid reflections and shadow unless
they are designed for the purpose of image processing. The main task of lighting is to
create contrast between the object features to be detected. Typical lighting techniques
are explained below.

Fig. 4.20 Light reflections

Direct Incident Lighting This simple lighting technique can be used for non-
reflective materials which strongly scatter the light due to their matte, porous, fibrous,
non-glossy surface. Ideally, a ring light is chosen for smaller illuminated fields that
can be arranged around the lens. Shadows are avoided to the greatest extent due to the
absolutely vertical illumination. Halogen lamps and large fluorescence illumination
can be used too.
Diffuse Incident Lighting Diffused light is necessary for many applications,
e.g., to test reflective, polished, glossy, or metallic objects. It is particularly difficult
if these surfaces are not glossy, perfectly flat, but individually shaped, wrinkled,
curved, or cylindrical. To create diffused lighting, one may use incident light with
diffusers, coaxial illumination, i.e., light is coupled into the axis of the camera by
means of a beam splitter or half-mirror, or the dome-shaped illumination where light
is diffused by means of a diffused coated dome in which the camera looks through
an opening in the dome onto the workpiece.
 96 Introduction to Robotics

Lateral Lighting Light from the side can be radiated at a relatively wide or narrow
angle. The influence on the camera image can be significant. In an extreme case, the
image information can almost be inverted.
Dark Field Lighting At first sight, images captured using dark field illumination
seem unusual to the viewer. The light shines at a shallow angle. According to the
principle of angle of incidence equals the angle of reflection, all the light is directed
away from the camera. The field of view, therefore, remains dark. Inclined edges,
scratches, imprints, slots, and elevations interfere with the beam of light. At these
anomalies, the light is reflected towards the camera. Hence, these defects appear
bright in the camera image.
Backlighting Transmitted light illumination is the first choice of lighting when it
is necessary to measure parts as accurately as possible. The lighting is arranged on
the opposite side of the camera, the component itself is put in the light beam.
Example 4.5 Pixel Processing
Assume a vidicon tube generates a signal for 512 lines of a faceplate. If the sampling
capability of an ADC is 100 nanoseconds, i.e., the time required to process one pixel,
and 1/30 seconds are required to scan all 512 lines then scanning rate for each line is,
1
= 65.1 microsecond per line (4.11)
(30 ¥ 512)
Hence, the number of pixels which can be processed per line is given below:
65.1 ¥ 10 -6
= 651 pixels per line (4.12)
100 ¥ 10 -9

Example 4.6 Memory Requirement

If a video image requires 256 ¥ 256 pixels, the total number of pixels in an image is,
256 ¥ 256 = 65,536 (4.13a)
Assuming each pixel is digitalized at the rate of 8 bits for 256 shades of gray then
it would require
65,536 ¥ 8 = 524,288 bits or 65,536 bytes (4.13b)
In case the picture is in three colors of red, green, and blue (RGB), and the video
image changes at the rate of 30 images per second then the memory requirement is,
65,536 ¥ 3 ¥ 30 = 5,898,240 bytes per second (4.13c)

4.4.2 Steps in a Vision System

As depicted in Fig. 4.21, vision sensing has two steps, namely, image acquisition and
image processing. They are explained below.

Fig. 4.21 Steps in vision

 Sensors, Vision and Signal Conditioning 97

1. Image Acquisition In image acquisition, an image is acquired from a vidicon

which is digitized or from a digital camera (CCD or CID), as explained in the previous
section. The image is stored in computer memory (also called a frame buffer) in
the format such as TIFF, JPG, Bitmap, etc. The buffer may be a part of the frame
grabber card or in the computer itself. Note that the image acquisition is primarily a
hardware function, however, software can be used to control light intensity, focus,
camera angle, synchronization, field of view, read times, and other functions. Image
acquisition has four principle elements, namely, a light source, either controlled or
ambient, which is explained in Section 4.4.1, a lens that focuses reflected light from
the object on to the image sensor, an image sensor that converts the light image into a
stored electrical image, and the electronics to read the sensed image from the image-
sensing element, and after processing, transmit the image information to a computer
for further processing. A typical acquired image is shown in Fig. 4.22.

Fig. 4.22 Digitized picture

Example 4.7 Image Storage and Retrieval

Note that an image is generally acquired at the rate of 30 frames per second or each
image stored in every 1/30 second. If a vision system has a pixel density of 350 pixels
per line and 280 lines, there will be a total of 350 ¥ 280 = 98,000 pixels. Moreover,
for a 6-bit register to represent various gray levels of a pixel, there is a requirement
of storage of 98,000 ¥ 6 = 588,000 bits of data, which is to be processed in every
1/30 second. For storage, a combination of row and column counters is used in the
frame grabber. To read the information stored in the frame buffer, the data is grabbed
via a signal sent from the computer to the address corresponding to a row-column
combination.

2. Image Processing Image-processing techniques are used to enhance,

improve, or otherwise alter an image and to prepare it for image analysis. Usually,
during image processing, information is not extracted from the image. The intention
is to remove faults, trivial information, or information that may be important, and
to improve the image. Image processing examines the digitized data to locate and
 98 Introduction to Robotics

recognize an object within the image field. It is divided into several sub-processes,
which are discussed below:
Image Data Reduction Here, the objective is to reduce the volume of data.
As a preliminary step in the data analysis, the schemes like digital conversion or
windowing can be applied to reduce the data. While the digital conversion reduces
the number of gray levels used by the vision system, windowing involves using
only a portion of the total image stored in the frame buffer for image processing
and analysis. For example, in windowing, to inspect a circuit board, a rectangular
window is selected to surround the component of interest and only pixels within that
window are analyzed.
Histogram Analysis A histogram is a representation of the total number of pixels
of an image at each gray level. Histogram information is used in a number of different
processes, including thresholding. For example, histogram information can help in
determining a cut-off point when an image is to be transformed into binary values.
Thresholding It is the process of dividing an image into different portions or
levels by picking a certain grayness level as a threshold. Comparing each pixel value
with the threshold, and then assigning the pixel to the different portions or level,
depending on whether the pixel’s grayness level is below the threshold (‘off’ or 0, or
not belonging) or above the threshold (‘on’ or 1, or belonging).
Masking A mask may be used for many different purposes, e.g., filtering operations
and noise reduction, and others. It is possible to create masks that behave like a low-
pass filter such that higher frequencies of an image are attenuated while the lower
frequencies are not changed very much. This is illustrated in Example 4.8. Thereby,
the noise is reduced. Masking an image considers a portion of an imaginary image
shown in Fig. 4.23(a), which has all the pixels at a gray value of 20 except the one at
a gray level of 100. The one with 100 may be considered noise. Applying the 3 ¥ 3
mask shown in Fig. 4.23(b) over the corner of the image yields the following value:
(m1 A + m2 B + m3C + m4 E + m5 F + m6 G + m7 I + m8 J + m9 K )
X= = 29 (4.14)
S
where S ∫ m1 + m2 + � + m9 = 9, and values of A, B, …, and m1, m2, … are shown
in Figs. 4.23 (a) and (b), respectively. As a result of application of the mask, the pixel
with the value of 100 changes to 29, as indicated in Fig. 4.23(c). The large difference

Fig. 4.23 Masking of an image

 Sensors, Vision and Signal Conditioning 99

between the noisy pixel and the surrounding pixels, i.e., 100 vs. 20, becomes much
smaller, namely, 29 vs. 20, thus reducing the noise. With this characteristic, the mask
acts as a low-pass filter. Note that the above reduction of noise has been achieved
using what is referred as neighborhood averaging, which causes the reduction of the
sharpness of the image as well.
Edge Detection Edge detection is a general name for a class of computer programs
and techniques that operate on an image and result in a line drawing of the image.
The lines represent changes in values such as cross section of planes, intersections
of planes, textures, lines, etc. In many edge-detection techniques, the resulting edges
are not continuous. However, in many applications, continuous edges are preferred,
which can be obtained using the Hough transform. It is a technique used to determine
the geometric relationship between different pixels on a line, including the slope
of the line. Consider a straight line in the xy-plane, as shown in Fig. 4.24, which is
expressed as
y = mx + c (4.15)
where m is the slope and c is the intercept. The line represented by Eq. (4.15) can
be transformed into a Hough plane of m – c with x and y as its slope and intercept,
respectively. Thus, a line in the xy-plane with a particular slope and intercept will
transform into a point in the Hough plane. For the Hough or m-c plane, if a line
passes through (m1, c1), as shown in Fig. 4.24(b), it represents a point (x1, y1) in the
x-y plane of Fig. 4.24(a). In that way, all lines through a point in the Hough plane
will transform into a single line in the x-y plane. Alternatively, if a group of points
is collinear, their Hough transform will all intersect. By examining these properties
in a particular case, it can be determined whether a cluster of pixels is on a straight
line or not, whereupon the orientation of an object in a plane can be determined by
calculating the orientation of a particular line in the object.

Fig. 4.24 Hough transform

Segmentation Segmentation is a generic name for a number of different techniques

that divide the image into segments of its constituents. The purpose of segmentation
is to separate the information contained in the image into smaller entities that can
be used for other purposes. Segmentation includes edge detection, as mentioned
100 Introduction to Robotics

above, region growing and splitting, and others. While region growing works based
on the similar attributes, such as gray-level ranges or other similarities, and then try
to relate the regions by their average similarities, region splitting is carried out based
on thresholding in which an image is split into closed areas of neighborhood pixels
by comparing them with thresholding value or range.
Morphology Operations Morphology operations are a family of operations
which are applied on the shape of subjects in an image. They include many different
operations, both for binary and gray images, such as thickening, dilation, erosion,
skeletonization, opening, closing, and filing. These operations are performed on an
image in order to aid in its analysis, as well as to reduce the ‘extra’ information that
may be present in the image. For example, Fig. 4.25(a) shows the object which after
skeletonization is shown in Fig. 4.25(b).

(a) Object (b) After reducing image to lines

Fig. 4.25 Skeletonization

3. Image Analysis Image analysis is a collection of operations and techniques

that are used to extract information from images. Among these are feature extraction;
object recognition; analysis of the position, size, orientation; extraction of depth
information, etc. Some techniques can be used for multiple purposes. For example,
moment equations may be used for object recognition, as well as to calculate the
position and orientation of an object. It is assumed that image processing has already
been performed to the image and available for the image analysis. Some of the image-
analysis techniques are explained below.
Feature Extraction Objects in an image may be recognized by their features
that uniquely characterize them. These include, but are not limited to, gray-level
histograms, morphological features such as perimeter, area, diameter, number of
holes, etc., eccentricity, cord length, and moments. As an example, perimeter of an
object may be found by first applying an edge-detection routine and then counting the
number of pixels on the perimeter. Similarly, the area can be calculated by region-
growing techniques, whereas diameter of a non circular object is obtained by the
maximum distance between any two points on any line that crosses the identified
area of the object. In order to know the thinness of an object, it can be calculated
using either of the two ratios:
(Perimeter)2 Diameter
Thickness = or Thinness = (4.16)
Area Area
 Sensors, Vision and Signal Conditioning 101

Moreover, moment of an image, denoted with np,q, can be calculated as

np,q = Â x p yq (4.17)
x, y
where x and y are the coordinates of each pixel that is turned on within an image, and
p and q are their respective powers, as
illustrated in Fig. 4.26. The coordinates x
and y are measured either from a designated
coordinate frame located at the edge of the
image or are measured from the frame
formed by the first row and column of the
image. As an example, in Fig. 4.26, the first
‘on-pixel’ appears in the second row and
fourth column. As per xy coordinate frame,
it is at (4, 2) location, whereas the x distance
of the pixel with respect to the first column
is 3. For calculation purposes, only one
Fig. 4.26 Moment of an image
coordinate frame should be used.
According to Eq. (4.17), when p = q = 0,
the moment n0,0 can be calculated as
n0,0 = Â1 = Number of pixels, i.e., area (4.18)
x, y

Similarly, one can find n0,1 as the summation of y-coordinates which is same as
the summation of each pixel area multiplied by its distance from the x-axis. Hence, it
is similar to the first moment of the area relative to the x-axis. Then, the location of
the center of the area relative to the x-axis can be calculated immediately as
1 n0,1
y– = Â y= (4.19a)
Area n0,0
Following Eq. (4.19a), one can also find the location of the center of the area
relative to the y-axis as
1 n1,0
x– = Â x= (4.19b)
Area n0,0
In a similar manner, one can also find the second moment of area with respect to
x and y axes by putting p = 2; q = 0, and p = 0; q = 2, respectively. Note here that the
first moments are independent of orientation, whereas the second moments are not.

Example 4.8 Moments Calculation

Referring to Fig. 4.26 and Eqs. (4.17–4.19),
n0,0 = Â 1 = 13 (4.20a)
x, y

n0,1 = Â y = Â x 0 y1 = 1(2) + 5(3) + 5(4) + 1(5) + 1(6) = 48 (4.20b)

n1,0 = Â x = Â x1 y 0 = 2(2) + 2(3) + 5(4) + 2(5) + 1(6) + 1(7) = 53 (4.20c)

102 Introduction to Robotics

n0,2 = Â x 0 y 2 = 1(2)2 + 5(3)2 + 5(4)2 + 1(5)2 + 1(6)2 = 190 (4.20d)

2 0 2 2 2 2 2 2
n2,0 = Â x y = 2(2) + 2(3) + 5(4) + 2(5) + 1(6) + 1(7) = 241 (4.20e)
Using Eqs (4.20a-c), one can easily find the center of the picture as,
n1,0 53 n0,1 48
x= = = 4.1; and y = = = 3.7 (4.21)
n0,0 13 n0,0 13

Object Recognition The next in image analysis is to identify the object that the
image represents based on the extracted features. The recognition algorithm should
be powerful enough to uniquely identify the object. Typical techniques used in the
industries are template matching and structural technique.
In template matching, the features of the object in the image, e.g., its area,
diameter, etc., are compared to the corresponding stored values. These values
constitute the stored template. When a match is found, allowing for certain statistical
variations in the comparison process, the object has been properly classified. For
example, the area of an image given by the moment n0,0 can be used to determine
nature of the object, and to distinguish it from other objects that have different
areas. Similarly, the first and second moments can be used to identify an object’s
position and orientation so that it can be picked up by the robot. Note that, instead of
calculating the moment values which change as the object moves, a look-up table can
be prepared. For example, the moment of inertia of an object varies significantly as
the object rotates about its center. Hence, the values of n2,0 and n0,2 will also change.
Since each orientation creates a unique value, a look-up table that contains these
values can later be used to identify the orientation of the object.
An important aspect here is that similar shapes,
as shown in Fig. 4.27 cannot be distinguished with
their first and second moments, as they are similar
or close to each other. In such a situation, a small
difference between the two objects may be
exaggerated through higher-order moments,
Fig. 4.27 Two similar objects
making identification of the object possible.
Structural techniques of pattern recognition rely on the relationship between
features or edges of an object. For example, if the image of an object can be subdivided
in four straight lines connected at their end points, and the connected lines are at right
angles then the object is a rectangle. This kind of technique is known as syntactic
pattern recognition.
In case depth information of an object is required, it can be obtained either by
using a range finder in conjunction with a vision system, or using a stereo vision. The
details of these systems are beyond the scope of this book. Readers are encouraged to
look at Haralick and Shapiro (1992) and others.
4.4.3 Hierarchy of a Vision System
The collection of processes involved in visual perception are often perceived as a
hierarchy spanning the range from ‘low’ via ‘intermediate’ to ‘high-level’ vision. The
notion of ‘low’ and ‘high’ vision are used routinely, but there is no clear definition of
 Sensors, Vision and Signal Conditioning 103

the distinction between what is considered ‘high’ as opposed to ‘low-level’ vision. As

shown in Fig. 4.28, a vision is classified as ‘low’, ‘intermediate’ or ‘high-level’ vision
based on specific activities during the image-processing stage. They are explained
below.
1. Low-level Vision The sequence of steps from image formation to image
acquisition, etc., described above, along with the extraction of certain physical
properties of the visible environment, such as depth, three-dimensional shape, object
boundaries, or surface-material properties, can be classified as a process of low-level
vision. Activity in the low-level vision is to process images for feature extraction
(edge, corner, or optical flow). Operations carried out on the pixels in the image to
extract the above properties with respect to intensity or depth at each point in the
image. One may, for example, be interested in extracting uniform regions, where
the gradient of the pixels remains constant, or first-order changes in gradient, which
would correspond to straight lines, or second-order changes which could be used to
extract surface properties such as peaks, pits, ridges, etc. A number of characteristics
that are typically associated with low-level vision processes are as follows:
∑ They are spatially uniform and parallel, i.e., with allowance for the decrease in
resolution from the center of the visual field outwards, similar process is applied
simultaneously across the visual field. For example, processing involved in edge
detection, motion, or stereo vision, often proceed in parallel across the visual
field, or a large part of it.
∑ Low-level visual processes are also considered ‘bottom-up’ in nature. This
means that they are determined by the data, i.e., data driven, and are relatively
independent of the task at hand or knowledge associated with specific objects.
As far as the edge detection is concerned, it will be performed in the same
manner for images of different objects, with no regard to whether the task to do
with moving around, looking for a misplaced object, or enjoying the landscape.
2. Intermediate-level Vision In this level, objects are recognized and 3D
scenes are interpreted using the features obtained from the low-level vision. The
intermediate-level processing is fundamentally concerned with grouping entities
together. The simplest case is when one groups pixels into lines. One can then express
the line in a functional form. Similarly, if the output of the low-level information
is a depth map, one may further need to distinguish object boundaries, or other
characteristics. Even in the simple case where one is trying to extract a single sphere,
it is not an easy process to go from a surface-depth representation to a center-and-
radius representation. In contrast to higher-level vision, the process here does not
depend on the knowledge about specific object.
3. High-level Vision High-level vision, which is equivalent to image underst-
anding, is concerned mainly with the interpretation of scene in terms of the objects
in it, and is usually based on knowledge of specific objects and relationships. It is
concerned primarily with the interpretation and use of information in the image
rather than the direct recovery of physical properties.
In high-level vision, interpretation of a scene goes beyond the tasks of line
extraction and grouping. It further requires decisions to be made about types of
boundaries, such as which are occluding, and what information is hidden from the
104 Introduction to Robotics

Fig. 4.28 Hierarchy of a vision system

user. Further grouping is essential at this stage since one may still need to be able
to decide which lines group together to form an object. To do this, it is necessary to
further distinguish lines which are part of the object structure, from those which are
part of a surface texture, or caused by shadows. High-level systems are, therefore,
object oriented, and sometimes called ‘top-down’. High-level visual processes are
applied to a selected portion of the image, rather than uniformly across the entire
image, as done in low- and intermediate-level visions. They almost always require
some form of knowledge about the objects of the scene to be included.
4.4.4 Difficulties in Vision and Remedies
A vision system cannot uniquely represent or process all available data because of
computational problem, memory, and processing-time requirements imposed on the
computer. Therefore, the system must compromise. Other problems include variation
of light, part-size, part placement, and limitations in the dynamic ranges available in
typical vision sensors. A vision system requires specialized hardware and software.
It is possible to purchase just the hardware with little or no vision application
programming. In fact, a few third-party programs are available. A hardware-only
approach is less expensive and can be more flexible for handling usual vision
requirements. But, since this approach requires image processing expertise, it is only
of interest to users who wish to retain the responsibility of image interpretation. It
is usual practice to obtain the hardware and application software together from the
supplier. However, the user might still need custom programming for an application.
 Sensors, Vision and Signal Conditioning 105

Major vision system suppliers specialise in providing software for only a few
application areas.
Every vision system requires a sensor to cover the visual image into an electronic
signal. Several types of video sensors are used, including vidicon cameras, vacuum
tube devices, and solid-state sensors. Many of these vision systems were originally
designed for other applications, such as television so the signal must be processed to
extract the visual image and remove synchronization information before the signal
is sent to the computer for further processing. The computer then treats this digital
signal as the array pixels, and processes this data to extract the desired information.
Image processing can be very time consuming. For a typical sensor of 200,000
or more pixels, a vision system can take many seconds, even minutes, to analyze
the complete scene and determine the action to be taken. The number of bits to be
processed is quite large, for example, a system with 512 × 512 pixels array and an
8-bit intensity per pixel yields over two million of bits to be processed. If continuous
image at a 30 FPS frame rate were being received, data bytes would be received at
an 8 MHz rate. Few computers can accept inputs at these data rates and, in any case,
there would be no time left to process the data. When higher resolution system, color
system, or multiple camera systems are considered, data-handling requirements
become astronomical. Several methods can be used to reduce the amount of data
handled and, therefore, the processing time. They are explained as follows:
(a) One approach is the binary vision, which is used when only black-and-white
information is processed (intensity variations and shades of gray are ignored).
In binary vision, a picture is converted into a binary image by thresholding, as
illustrated in Fig. 4.29. In thresholding, a brightness level is selected. All data
with intensities equal to or higher than this value are considered white, and all
other levels are considered black.
(b) Another method of shortening process time is to control object placement so
that objects of interest cannot overlap in the image. Complicated algorithms
to separate images are then not necessary, and the image-processing time is
reduced.

Fig. 4.29 Image thresholding [Courtesy: Shell and Hall (2000)]

106 Introduction to Robotics

(c) A third approach reduces data handling by processing only a small window of
the actual data; that is, the object is located in a predefined field of view. For
example, if the robot is looking for a mark on the printed circuit board, the
system can be held in such a way that the mark is always in the upper right
corner.
(d) A fourth approach takes a statistical sample of data and makes decisions on this
data sample. Unfortunately, all of these approaches ignore some of the available
data and, in effect, produce a less robust system. Processing time is saved, but
some types of complex objects cannot be recognized.
Note that some of the above steps are actually a part of image processing explained
in Section 4.4.2.

4.5 SIGNAL CONDITIONING

The basic information or data generated by the transducers (or sensors as mentioned in
‘Sensors vs. Transducers’ text box in the beginning) generally requires ‘conditioning’
or ‘processing’ of one sort or another before they are presented to the observer as an
indication or a record, or to be used by a robot controller for further action. Some of
the commonly used devices used for signal conditioning of those obtained from the
sensors explained in this chapter are explained next. The detailed descriptions are
out of the scope of this book. However, they are available in the books by Doebelin
(1990), de Silva (2004), Nakra and Chaudhry (2009), and others.
4.5.1 Amplifiers
Since the electrical signals produced by most transducers of a sensor are at a low
voltage, generally they require amplification before they are suitably recognized
by a monitoring system like a data processor, controller, or data logger. While the
discussion is aimed mainly at users (rather than designers) of amplifiers, the use of
operational amplifiers with the sensors will be explained here.
The operational amplifier (op-amp) is the most widely utilized analog electronic
sub-assembly. It is the basis of instrumentation amplifier, filters, and a myriad of
analog and digital data-processing equipment. An op-amp could be manufactured
in the discrete-element form using, say, ten bipolar junction transistors and as many
discrete resistors or alternatively (and preferably) in the monolithic form as in IC
(Integrated Circuit) chip that may be equivalent to over 100 discrete elements. In any
form, the device has an input impedance Zi, an output impedance Zo and a gain K, as
indicated in Fig. 4.30(a). A common 8-pin dual in-line package (DIP) or V-package
is shown in Fig. 4.30(b). From Fig. 4.30(a), the open-loop condition yields
vo = Kvi (4.22)
where the input voltage vi is the differential input voltage defined as the algebraic
difference between the voltages at the +ve and –ve leads. Thus,
vi = vip – vin (4.23)
The open-loop voltage gain K is typically very high (105 – 109), whereas values of
Zi and Zo are typically 2 MW (could be as high as 10 MW) and 10 W (may reach about
75W), respectively. Since v0 is typically 1–15 V, from Eq. (4.22), it follows that vi
@ 0 since K is very large. Hence, from Eq. (4.23) we have vip @ vin. In other words,
 Sensors, Vision and Signal Conditioning 107

Fig. 4.30 Operational amplifier

the voltages at the two input leads are nearly equal. If a large voltage differential vi
(say, 10 V) at the input then according to Eq. (4.22), the output voltage should be
extremely high. This never happens in practice, because the device saturates quickly
beyond moderate output voltages in the order of 15 V. From Eqs (4.22–4.23), it is
clear that if the –ve input lead is grounded, i.e., vin = 0 then
vo = Kvip (4.24a)
and if the +ve input lead is grounded, i.e., vip = 0 then
vo = –Kvin (4.24b)
This is the reason why vip is termed non-inverting input and vin is termed inverting
input.
4.5.2 Filters
A filter is a device that allows through only
Dynamic System vs. Analog Filter
the desirable part of a signal, rejecting
the unwanted part. Unwanted signals can The way an analog sensor is defined, any
dynamic system can be interpreted as an
seriously degrade the performance of a
analog filter.
robotic system. External disturbances,
error components in excitations, and noise generated internally within system
components and instrumentation are such spurious signals, which may be removed
by a filter. There are four broad categories of filters, namely, low-pass filters, high-
pass filters, band-pass filters, and band-reject (or notch) filters, which are shown in
Fig. 4.31.

Magnitude Magnitude Magnitude Magnitude

1 1 1 1

fc(cut-off) fc(cut-off) fc1 fc 2 fc 1 fc 2

Frequency Frequency Frequency Frequency
(a) Low-pass (b) High-pass (c) Band-pass (d) Band-reject

Fig. 4.31 Filter characteristics

108 Introduction to Robotics

An analog filter contains active components like transistors or op-amps. It is a

physical dynamic system, typically an electric circuit, whose dynamics will determine
which (desired) signal components would be allowed through and which (unwanted)
signal components would be rejected. In a way, output of the dynamic system is the
filtered signal. An analog filter can be represented as a differential equation with
respect to time.
Filtering can be achieved through digital filters as well which employ digital signal
processing. Digital filtering is an algorithm by which a sampled signal (or sequence
of numbers), acting as an input, is transformed to a second sequence of numbers
called the output. It is a discrete-time system and can be represented as a difference
equation. Digital filtering has the usual digital benefits of accuracy, stability, and
adjustability by software (rather than hardware) changes.
4.5.3 Modulators and Demodulators
Signals are sometimes deliberately modified to maintain the accuracy during their
transmission, conditioning, and processing. In modulation, the data signal which is
referred to as modulating signal is used to vary a property, say, amplitude or frequency,
of a carrier signal. Thus, the carrier signal is modulated by the data signal. After
transmitting or conditioning, the data signal is recovered by removing the carrier
signal from the modulated signal. This step is known as demodulation. The carrier
signal can be either sine or square wave, and its frequency should be 5–10 times the
highest frequency of the data signal.
Figure 4.32 shows some typical modulation techniques in which the amplitude
of a high-frequency sinusoidal carrier signal is varied according to the amplitude of
the data signal. Figure 4.32(a) shows the data signal which needs to be transmitted.
The carrier signal’s frequency is kept constant, while amplitude is same as that of the
data signal. This is called Amplitude Modulation (AM), and the resulting amplitude-
modulated signal is shown in Fig. 4.32(b). In Fig. 4.32(c), however, the frequency of

Fig. 4.32 Modulation techniques

 Sensors, Vision and Signal Conditioning 109

the carrier signal is varied in proportion to the amplitude of the data signal
(modulating signal), while keeping the amplitude of the carrier signal constant. This
is called Frequency Modulation (FM). An FM is less susceptible to noise than AM.
In Pulse-Width Modulation (PWM), the carrier signal is a pulse sequence, and its
width is changed in proportion to the amplitude of the data signal, while keeping
the pulse spacing constant. This is shown in Fig. 4.32(d). The PWM signals can be
used directly in controlling a process without having to demodulate them. There
also exists Pulse-Frequency Modulation (PFM), where the carrier signal is a pulse
sequence. Here, the frequency of the pulses is changed in proportion to the value
of the data signal, while keeping the pulse width constant. It has the advantages of
ordinary FM.
Demodulation or detection is the process of extracting the original data signal
from a modulated signal. A simple and straightforward method of demodulation is
by detection of the envelope of the modulated signal.
4.5.4 Analog and Digital Conversions
Most sensors have analog output while much data processing is done using digital
computers. Hence, devices for conversion between these two domains have to be
performed. These can be achieved using an analog-to-digital converter (ADC) and a
digital-to-analog converter (DAC), as mentioned in Chapter 2. Explanation on DAC
is given first as some ADC uses DAC as its component.
1. Digital-to-Analog Converter (DAC) The function of a digital-to-analog-
convertor or DAC is to convert a sequence of digital words stored in a data register,
typically in the straight binary form, into an analog signal. A typical DAC unit is
an active circuit in the integrated circuit (IC) form and consists of a data register
(digital circuits), solid-state switching circuits, resistors, and op-amps powered by an
external power supply. The IC chip that represents the DAC is usually one of many
components mounted on a Printed Circuit Board (PCB), which is the I/O board or
card. This board is plugged into a slot of the PC having DAQ. A typical commercial
DAC is shown in Fig. 2.10.
2. Analog-to-Digital Converter (ADC) An analog-to-digital converter or ADC,
on the other hand, converts an analog signal into the digital form, according to an
appropriate code, before the same is used by a digital processor or a computer. The
process of analog-to-digital conversion is more complex and time consuming than
the digital-to-analog conversion. ADCs are usually more costly, and their conversion
rate is slower than DACs. Several types of ADCs are commercially available.
However, their principle of operation may vary depending on the type. A typical
commercial ADC is shown in Fig. 2.8.
Note that the most fundamental property of any DAC or ADC is their number of
bits for which it is designed, since this is a basic limit on resolution. Units of 8 to
12 bits are most common even though higher bits are also available. Both DAC and
ADC are elements in typical input/output board (or I/O board, or data acquisition
and control card, i.e., DAC or DAQ), which are usually situated on the same digital
interface board.
110 Introduction to Robotics

4.5.5 Bridge Circuits B

Various bridge circuits are employed R2
widely for the measurement of R1
resistance, capacitance, and inductance,
as many transducers convert physical
A C
variables to these quantities. Figure 4.33
shows a purely resistive (Wheatstone) Meter
bridge in its simplest form. The basic
R4
principle of the bridge may be applied R3
in two different ways, namely, the null
method and the deflection method. D
If the resistances are adjusted so
Fig. 4.33 Wheatstone bridge
that the bridge is balanced then there is
no voltage across AC, i.e., vAC = 0. This happens when R1/R4 = R2/R3. Now if one of
the resistors, say, R1, changes its resistance then there will be unbalance in the bridge
and a voltage will appear across vAC causing a meter reading. This meter reading is
an indication of the change in R1 of the transducer element, and can be utilized to
compute the change. This method of measurement is called the deflection method.
In the null method, one of the resistors is adjustable manually. Thus, if R1 changes
causing a meter deflection, R2 can be adjusted manually till its effect just cancels
that of R1 and the bridge is returned to its balanced position. Here, the change in R1
is directly related to the change in R2 required to effect the balance. Note that both
the deflection and null methods require a calibration curve so that one knows the
numerical values of R1 or R2 that has caused the imbalance or balance, respectively.
Note that the measurements of rapid dynamic phenomenon can be done using the
deflection method. Moreover, based on the alternate current (ac) and direct current
(dc) excitations in the bridge, there are ac bridges and dc bridges, respectively.
4.5.6 Signal Analyzer
Modern signal analyzers employ digital techniques of signal analysis to extract useful
information that is carried by the signal. Digital Fourier analysis using Fast Fourier
Transform (FFT) is perhaps the singlemost common procedure that is used in the vast
majority of signal analyzers. Fourier analysis produces the frequency spectrum of a
time signal, which is explained here in brief.
Any periodic signal f(t) can be decomposed into a number of sines and cosines
of different amplitudes, an and bn, and frequencies nw t, for n = 1, 2, …•, which is
expressed as
a0 • •
f(t) = + Â an cos(nw t ) + Â bn sin(nw t ) (4.25)
2 n =1 n =1

If one adds sine and cosine functions together, the original signal can be
reconstructed. Equation (4.25) is called a Fourier series, and the collection of
different frequencies present in the equation is called the frequency spectrum or
frequency content of the signal. Even though the signal is in amplitude-time domain,
the frequency spectrum is in the amplitude-frequency domain. For example, for the
function, f(t) = sin(t) of Fig. 4.34(a), which consists of only one frequency with
 Sensors, Vision and Signal Conditioning 111

constant amplitude, the plotted signal would be represented by a single line at the
given frequency, as shown in Fig. 4.34(b). If the plot with given frequency and
amplitude is represented by the arrow in Fig. 4.34(b), the same sine function can be
reconstructed. The plots in Fig. 4.35 are similar and represent the following:
1
f(t) = Â sin(nt ) (4.26)
n =1,3,...,15 n

where w = 1.
Amplitude Amplitude

1 1

Time 1 Frequency

(a) Signal (b) Frequency domain

Fig. 4.34 Time domain vs. frequency domain

Amplitude Amplitude

Time 1 3 5 7 15 Frequency

(a) Time domain (b) Frequency domain

Fig. 4.35 Complex functions

The frequencies are also plotted in the frequency-amplitude domain. Clearly,

when the number of frequencies contained in f(t) increases, the summation becomes
closer to a square function. Theoretically, to reconstruct a square wave from sine
functions, an infinite number of sines must be added together. In practice, however,
some of the major frequencies within the spectrum will have larger amplitudes.
These major frequencies or harmonics are used in identifying and labeling a signal,
including recognizing shapes, objects, etc.

4.6 SENSOR SELECTION

In using sensors, one must first decide what the sensor is supposed to do and what
result one expects. A sensor detects the quantity to be measured (the measurand). The
transducer converts the detected measurand into a convenient form for subsequent
use, e.g., for control or actuation. The transducer signal may be filtered, amplified and
suitably modified using the devices mentioned in Section 4.5. Selection of suitable
sensors for robotic applications relies heavily of their performance specifications.
112 Introduction to Robotics

Majority of manufacturers provide what are actually static parameters. However,

dynamic parameters are also important. In this section, static characteristics will be
studied, whereas the dynamic characteristics can be described using the transient
response of a system explained in Chapter 10 (Control), and the associated definitions
appear therein, as a sensor is also a dynamic system whose behavior changes with
the input signals. The following definitions will help a user or a designer select
appropriate sensors for a robotic application.
1. Range Range or span is a measure of the difference between the minimum and
maximum values of its input or output (response) so as to maintain a required level of
output accuracy. For example, a strain gauge might be able to measure output values
over the range from 0.1 to 10 Newtons.
2. Sensitivity Sensitivity is defined as the ratio of the change of output to change
in input. As an example, if a movement of 0.025 mm in a linear potentiometer causes
an output voltage by 0.02 volt then the sensitivity is 0.8 volts per mm. It is sometimes
used to indicate the smallest change in input that will be observable as a change in
output. Usually, maximum sensitivity that provides a linear and accurate signal is
desired.
3. Linearity Perfect linearity would allow output versus input to be plotted as a
straight line on a graph paper. Linearity is a measure of the constancy of the ratio of
output to input. In the form of an equation, it is
y = mx (4.27)
where x is input and y is output, and m is a constant. If m is a variable, the relationship
is not linear. For example, m may be a function of x, such as m = a + bx where the
value of b would introduce a nonlinearity. A measure of the nonlinearity could be
given as the value of b.
4. Response Time Response time is the time required for a sensor to respond
completely to a change in input. The response time of a system with sensors is the
combination of the responses of all individual components, including the sensor. An
important aspect in selecting an appropriate sensor is to match its time response to
that of the complete system. Associated definitions like rise time, peak time, settling
time, etc., with regard to the dynamic response of a sensor are given in Chapter 10.
5. Bandwidth It determines the maximum speed or frequency at which an
instrument associated with a sensor or otherwise is capable of operating. High
bandwidth implies faster speed of response. Instrument bandwidth should be several
times greater than the maximum frequency of interest in the input signals.
6. Accuracy Accuracy is a measure of the difference between the measured
and actual values. An accuracy of ±0.025 mm means that under all circumstances
considered, the measured value will be within 0.025 mm of the actual value. In
positioning a robot and its end-effector, verification of this level of accuracy would
require careful measurement of the position of the end-effector with respect to the
base reference location with an overall accuracy of 0.025 mm under all conditions
of temperature, acceleration, velocity, and loading. Precision-measuring equipment,
 Sensors, Vision and Signal Conditioning 113

carefully calibrated against secondary standards, would be necessary to verify this

accuracy. Accuracy describes ‘closeness to true values.’
7. Repeatability and Precision Repeatability is a measure of the difference in
value between two successive measurements under the same conditions, and is a far
less stringent criterion than accuracy. As long as the forces, temperature, and other
parameters have not changed, one would expect the successive values to be the same,
however poor the accuracy is.
An associated definition is precision, which means the ‘closeness of agreement’
between independent measurements of a quantity under the same conditions without
any reference to the true value, as done above. Note that the number of divisions
on the scale of the measuring device generally affects the consistency of repeated
measurement and, therefore, the precision. In a way, precision describes ‘repeatability.’
Figure 4.36 illustrates the difference between accuracy and precision.

Fig. 4.36 Interpretation of accuracy and precision

8. Resolution and Threshold Resolution is a measure of the number of

measurements within a range from minimum to maximum. It is also used to indicate
the value of the smallest increment of value that is observable, whereas threshold is a
particular case of resolution. It is defined as the minimum value of input below which
no output can be detected.
9. Hysteresis It is defined as the change in the input/ Output
output curve when the direction of motion changes, as
indicated in Fig. 4.37. This behavior is common in loose
components such as gears, which have backlash, and in
magnetic devices with ferromagnetic media, and others. Input

10. Type of Output Output can be in the form of a

mechanical movement, an electrical current or voltage, a
pressure, or liquid level, a light intensity, or another form. Fig. 4.37 Hysteresis
To be useful, it must be converted to another form, as in
the LVDT (Linear Variable Differential Transducer) or strain gauges, which are
discussed earlier.
In addition to the above characteristics, the following considerations must also be
made while selecting a sensor.
11. Size and Weight Size and weight are usually important physical characteristics
of sensors. If the sensor is to be mounted on the robot hand or arm, it becomes a part
114 Introduction to Robotics

of the mass that must be accelerated and decelerated by the drive motors of the wrist
and arm. So, it directly affects the performance of the robot. It is a challenge to
sensor designers to reduce size and weight. An early wrist force-torque sensor, for
example, was about 125 mm in diameter but was reduced to about 75 mm in diameter
through careful redesign.
12. Environmental Conditions Power requirement and its easy availability
should be considered. Besides, conditions like chemical reactions including
corrosion, extreme temperatures, light, dirt accumulation, electromagnetic field,
radioactive environments, shock and vibrations, etc., should be taken into account
while selecting a sensor or considering how to shield them.
13. Reliability and Maintainability Reliability is of major importance in all
robot applications. It can be measured in terms of Mean Time To Failure (MTTF) as
the average number of hours between failures that cause some part of the sensor to
become inoperative. In industrial use, the total robot system is expected to be available
as much as 98 or 99% of the working days. Since there are hundreds of components
in a robot system, each one must have a very high reliability. Some otherwise good
sensors cannot stand the daily environmental stress and, therefore, cannot be used
with robots. Part of the requirement for reliability is ease of maintenance. A sensor
that can be easily replaced does not have to be as reliable as one that is hidden in the
depths of the robot. Maintainability is a measure in terms of Mean Time To Repair
(MTTR).
14. Interfacing Interfacing of sensors with signal-conditioning devices and the
controller of the robot is often a determining factor in the usefulness of sensors.
Nonstandard plugs or requirements for nonstandard voltages and currents may make
a sensor too complex and expensive to use. Also, the signals from a sensor must be
compatible with other equipment being used if the system is to work properly.
15. Others Other aspects like initial cost, maintenance cost, cost of disposal and
replacement, reputation of manufacturers, operational simplicity, ease of availability
of the sensors and their spares should be taken into account. In many occasions, these
nontechnical considerations become the ultimate deciding factor in the selection of
sensors for an application.

SUMMARY
Necessity of sensors in robotics, different types of sensors used in robotic applications
and the selection criteria are presented in this chapter. Functional aspects, rather than
their mathematical models, are emphasized so that the reader is able to decide which
sensor to use and when. Different sensors are classified from their functional uses, for
example, position, velocity, acceleration, force, sensors, etc. Special attention has been
paid to vision system, its elements, how it can be processed, etc. Signal-conditioning
units, their characteristics, and when to use what are also explained. Finally guidelines
are provided to select appropriate sensors.
 Sensors, Vision and Signal Conditioning 115

EXERCISES
4.1 Define sensitivity and linearity.
4.2 Distinguish accuracy from repeatability.
4.3 State the physical characteristics in sensor selection.
4.4 What are the essential components of a sensor?
4.5 Why are the terms internal and external used to classify sensors?
4.6 Classify internal sensors.
4.7 Name some velocity sensors?
4.8 Why are position sensors not preferred to obtain velocity and acceleration?
4.9 Is there any advantage of external sensors over internal types?
4.10 Name some contact and noncontact type sensors.
4.11 What are the advantages of capacitive proximity sensors?
4.12 What is a machine vision?
4.13 What are the major components in a vision system?
4.14 What is the sensor component in a vision system?
4.15 What are different types of cameras? Explain the vidicon camera.
4.16 What are the typical difficulties in a vision system?
4.17 What is thresholding in vision?
4.18 Why is signal processing important?
4.19 Name at least three devices used in signal conditioning.
4.20 Define different types of filters.
4.21 What are the characteristics one should check while selecting a sensor?
4.22 What is Hough transform?
4.23 What are the static characteristics a sensor should have?

WEB-BASED EXERCISES
Based on Web search, find the answers for the following questions:
4.24 Name at least five commercial companies who make different types of sensors. List the
sensors against each company’s name.
4.25 What are the typical prices of encoders, tachogenerators, and accelerometers?
4.26 What is the price of a CCD camera?
4.27 Name two camera companies and mention which camera you will select for industrial
robot system involved in accurate pick-and-place operation?
4.28 Name a robot manufacturing company who implements vision system in their robots.
4.29 Give a typical significance of a force/torque sensor.
4.30 What command MATLAB uses for digital filtering and image processing?