[email protected].

com

Machine Vision Camera

INDUSTRIAL APPLICATIONS
BY
JANORIOUS RABEELA

IBM | MAY 2018

Contents
Introduction .............................................................................................................................. 2
Network cameras:.................................................................................................................. 2
Industrial cameras: ................................................................................................................ 2
Area scan cameras ................................................................................................................. 2
Applications for AREA SCAN CAMERAs ............................................................................. 2
LINE SCAN CAMERAs .......................................................................................................... 3
Applications for Line SCAN CAMERAs................................................................................ 3
Camera classifications based on their specifications: .......................................................... 3
Resolution CalculATION ...................................................................................................... 5
List of AREA SCAN Cameras and their Specifications ....................................................... 7
List of LINE SCAN Cameras and their Specifications ........................................................8

PAGE 1
Introduction
This document provides the details of the various industrial cameras available in the market
for machine vision applications. The document covers details on the camera selection
based on the customer requirement, comparison of cameras and few application examples.

Cameras for image processing systems are categorized either as industrial/machine vision
(MV) or network/IP (Internet Protocol) cameras.

NETWORK CAMERAS:

Network cameras, also known as IP (Internet Protocol) cameras, record videos. They are
frequently used in classical surveillance applications. Network cameras acquire the images
and compress them. By connecting to a network, a theoretically unlimited number of users
can also access the camera.

INDUSTRIAL CAMERAS:

Industrial cameras by contrast send the images as uncompressed (‘raw’) data directly to
the processing node or cloud; The Processing node is then responsible for processing the
relatively large volume of data. The benefit of this method is that no image information is
lost.
There are two further subdivisions within the world of industrial cameras: Area scan and
line scan cameras.

AREA SCAN CAMERAS

Area scan cameras are equipped with a rectangular sensor featuring numerous lines of
pixels that are exposed at the same time. The image data is thus recorded in one single step
and is also processed in the same way.

APPLICATIONS FOR AREA SCAN CAMERAS

The following are the list of Machine Vision applications for Area Scan Camera:
• In the automotive industry for inspection of individual parts, in control
engineering and robotics for the positioning of components and code identification,
paint defect inspection etc
• In electronics industry for inspection of components such as circuit boards,
soldering checks and pick & place processes.
• In the pharmaceutical and packaging industries for checks of production steps
and inspection for completeness, compliance with dimensional requirement and
detection of material flaws
• LCD Panel Inspection
• Solar Panel Inspection

PAGE 2
 • Aerial Imaging
• Reconnaissance
•

LINE SCAN CAMERAS

Line scan cameras by contrast use one sensor comprised of just 1, 2 or 3 lines of pixels. The
image data is captured line by line, with the individual lines then reconstructed into an
entire image during the processing stage.
Line scan cameras are used universally when products must be inspected as they pass by on
conveyor belts – at times at extremely fast speeds.

APPLICATIONS FOR LINE SCAN CAMERAS

The following are the list of Machine Vision applications for Line Scan Camera:
• In printing industry checks of printed images in newspaper and magazine printing,
where the printed products are moving at up to 100km/h.
• In the logistics industry Millions of packages run through the sorting facilities at
logistics companies each day. Each individual shipment is identified and sorted at
lightning speeds using line scan cameras to ensure quick and smooth delivery.
• In the food industry, they are used to inspect food products for geometry, damaged
spots and particles. They inspect packaging and check the seals and caps on bottles.

CAMERA CLASSIFICATIONS BASED ON THEIR SPECIFICATIONS:

1. Sensor Types
a. CMOS
b. CCD
2. Shutter Technique
a. Rolling Shutter
b. Global Shutter
3. Frame Rates: This term is synonymous with ‘frames per second’ or ‘fps’; For line
scan cameras, the terms ‘line rate’ or ‘line frequency’ are used. It describes the
number of images that the sensor can capture and transmit per second. The frame
rate varies from 10fps to 340 fps. The frame rate required is based on the Image
processing systems requirement.
4. Resolution: Resolution describes a measurement of the smallest possible distance
between two lines or points such that they can still be perceived as separate from
one another within the image.
Resolution=Object Size/ Size of the detail to be inspected
Example 1 : You’d like to capture a precision image of the eye color of a roughly 2m
tall person standing at a specific point:
Resolution= 2m/1mm = 2000 pxl
pxl in x and y=4 MP

PAGE 3
Eye detail1 mm :
A resolution of 4 megapixels is required to ensure that the 1 mm large detail for the eye
is clearly recognizable.

5. Interfaces: The interface serves as the liaison between the camera and Image
processing node, forwarding image data from the camera sensor to the
components that process the images, i.e. the hardware and software. The choice is
available between a variety of modern, widely available technologies
a. Camera Link
b. Gigabit Ethernet

c. USB3.0
d. CoAxpress Connector: CoaXPress (CXP) is the world’s leading standard for
high-speed imaging in professional and industrial imaging applications such
as machine vision, medical imaging, life sciences, broadcast and defence.

6. Housing Sizes

PAGE 4
 The size of the camera housing is tied directly to the choice of interface. In
applications where cameras are organized next to one another (known as multicamera
setups) to better record the entire width of a material web, each millimeter of space
matters

RESOLUTION CALCULATION

Case 1:
Assume Customer Requirement:

1. Industry: Automotive
2. Use case: Paint Defect detection
3. Smallest defect size to be detected = 0.5 mm
4. Assume Object Size – 1mX1m (if the object size is bigger, we may need multiple
cameras)
5. Camera distance from the object – 2 m
6. The object will be there in front of the Camera Momentarily for a period of 4 – 5
seconds

Hence Minimum Resolution required:

Resolution = Object Size / Size of the object to be detected

Resolution = 1000/0.5 = 2000, required resolution in 1 axis

Spatial resolution required which is the physical measurement a pixel represents 0.125 mm
i.e 4 pixel for the smallest possible element to be detected
Spatial resolution = 8000 in 1 axis

Camera Resolution required is 64 M

If the defect size is to be detected in the above case is 1 mm, then the camera resolution
required will be 16M

Case 2:

1. Industry: Automotive
2. Use case: Paint Defect detection
3. Smallest defect size to be detected = 0.5 mm

PAGE 5
 4. Assume Object Size – 4mX 4m
5. Camera distance from the object – 2 m
6. The object will be there in front of the Camera Momentarily for a period of 4 – 5
seconds

Hence Minimum Resolution required: 16000X16000, we need 256K pixel resolution,

camera of that resolution is not available. In that case we need to use multiple cameras.
We can use two Line camera with resolution of 8K and a distance of 2m from each other.
But the obeject has to be moving for Linescan camera (refer : Table for Line Scan camera)
to provide the required results.

Resolution = Object Size / Size of the object to be detected

Resolution = 1000/0.5 = 2000, required resolution in 1 axis

Spatial resolution required which is the physical measurement a pixel represents 0.125 mm
i.e 4 pixel for the smallest possible element to be detected
Spatial resolution = 8000 in 1 axis

Camera Resolution required is 64 M

If the defect size is to be detected in the above case is 1 mm, then the camera resolution
required will be 16M

Example with Camera Vendor specification for the above requirement, the below is the
feasibility

If the FOV is 4mX4m and the minimum defect size is 0.7mm, you will need extremely high resolution cameras. ~ 20-30
MP.

There are 3 ways to do it.

1) Use a line scan camera. For example racer 8K camera.

https://www.baslerweb.com/en/products/cameras/line-scan-cameras/racer/ral8192-12gm/

In this case, camera will work like a scanner. The object have to move under the camera then it is possible to
scan the image. High precision is required in the setup so that you will get a very good image.

2) Use multiple area scan cameras and do stitching of images. Below is the specification of 12MP camera.
You will need at least 2 cameras to stitch.

https://www.baslerweb.com/en/products/cameras/area-scan-cameras/ace/aca4112-30um/

3) Use a 3D profiler incase the object is smaller in size and with curved surfaces

PAGE 6
 LIST OF AREA SCAN CAMERAS AND THEIR SPECIFICATIONS

Camera Model Manufacturer Type of camera Resolution Frame Interface

and and sensing Rate
Manufacturer
1 Model: Falcon4 Teledyne Dalsa CMOS, Area Scan 10720 x 8064 16 fps CameraLi
. 86M Color nk HS
Manufacturer:
Applications :
Flat Panel
Display
Inspection
Aerial Imaging
Reconnaissance
Security and
Surveillance
3D Metrology
2 Model: Falcon2 Teledyne Dalsa CMOS, Area Scan 4096 x 3072 58 fps Camera
. 12M color Link Mini

3 Model: SP- JAI CMOS , Area Scan 5120 x 3840 30fps CoaXPres
20000-CXP2- s
20M 2-
connector
(CXP2)
4 Model :SP- JAI CMOS, Area Scan 5120 x 3840 30fps Power
20000-PMCL - over
20 M Camera
Link
(PMCL)
Deca
5 SP-12000-CXP4 JAI CMOS Area SCAN 4096 x 3072 189fps CoaXPres
12M, s
4-
connector
(CXP2
6 Model: Basler CMOS Area SCAN 4608 px x 3288 7fps GigE
acA4600-7gc 14 px Vision
MP
Applications:

Machine Vision
Detection
Automotive
Manufacturing
Photo Voltaics

PAGE 7
7 Model: Basler CMOS Area SCAN 4608 px x 3288 10fps USB3.0
acA4600-10UC px
14 MP
Basler

8 Model : VC VieWorks CMOS Area Scan Upto 10000 × Upto 337 CoAx
Series 7096 fps Express
and
Camera
Link

LIST OF LINE SCAN CAMERAS AND THEIR SPECIFICATIONS

Camera Model Manufacturer Type of camera Resolution Line Interface

and and sensing Rate
Manufacturer
1 Model: PX-HC- Teledyne Dalsa CMOS,Line Scan 16352 x 12 40 CameraLin
. 16k04T k HS
2 Model: PX-HC- Teledyne Dalsa CMOS, Line Scan 16352 x 12 70 Camera
. 16k07T Link Mini
3 PX-HC-08k07T Teledyne Dalsa CMOS , Line Scan 8192 x 12 70 CoaXPress
2-
connector
(CXP2)
4 Sweep SW- JAI CMOS, Line Scan 8192 100 Camera
8000M-PMCL Link
(MonoChrome) (CL) Deca

5 SW-4000TL- JAI CMOS Line Scan 4096 66 Camera

PMCL Link
(CL) Deca
6 Model: Basler CMOS line SCAN 8192 80 GiGE
raL8192-12gm - Vision
Basler racer

PAGE 8
7 raL12288-8gm - Basler CMOS Line SCAN 12288 8 GiGE
Basler racer Vision

8 raL12288-66km Basler CMOS Line Scan 12288 66 Camera

- Basler racer Link

9 VT Series – M72 VieWorks CMOS Line SCAN 17824 x 256 250 CoaXPress
and
CameraLin
k
1 VT Series – M95 VieWorks CMOS Line SCAN 23360 x 256 140 CoaXPress
0

3D Laser Profilers for Defect detection:

Precision 3D surface profiling can be essential for determining a surface suitability for its
intended purpose. Components failures can often be traced back to a precision surface
that was not manufactured to specification.
3D surface profiling is a general term which encompasses more specific types of
measurements such as surface shape, surface finish, surface profile roughness and surface
defects in a nm precision for smaller parts.

Recording 3D scenes requires camera systems that can sense the depth of information of
each corresponding pixel of an image, in addition to texture information (e.g. RGB). Those
cameras are also known as range or depth cameras. Capturing the third dimension can be
done in many ways, and each of the machine vision technologies available has its pros and
cons. The following gives a basic idea of different 3d imaging systems available.

PAGE 9
There are often many possible ways to solve a specific vision task. In some cases, the
choice of either 2D or 3D vision is obvious, but in other cases both technologies could
work though each provides certain benefits. It is important to understand these benefits
and how they apply to a given application in order to provide a reliable machine vision
solution. In general, 3D is best suited not only for analyzing volume, shape or 3D position
of objects, but also for detecting parts and defects that are low contrast but have a
detectable height difference. The third-dimension is mainly used for measuring,
inspecting and positioning, but there are also cases where 3D is used to read imprinted
code or text when contrast information is missing.

Capturing the third dimension can be done in many different ways, and each of the
machine vision technologies available has its pros and cons. Three-dimensional imaging
can be broken into two main categories: passive and active. From there it can be broken
into much more specific techniques. Passive techniques include depth from focus, light
field, and stereo. The main active techniques are based on time-of-flight, structured light,
and interferometry.
Three-dimensional imaging can be further broken down into how the image is actually
acquired, including snapshot and scanning methods.
The advantages and disadvantages of the four main 3-d imaging system is discussed
below:

PAGE 10
Choosing which type of 3D vision system to use is highly dependent on the parts that need
to be measured. While laser range finders using time of flight methods can be used to
locate distant objects, stereo imaging systems may be better suited to imaging high-
contrast objects. Where such objects are highly specular, it may be more useful to employ
projected texture techniques.

Laser Triangulation Method for 3D Laser Profilers:

3D Laser Profilers use laser triangulation method to detect depth of image. Laser
triangulation uses a 2D camera and a laser light source. The laser projects a line onto the
target zone, to be captured using a 2D camera. The lines bend as they touch the contours
of the object, so the distance between the object and the laser light source is calculated
based on the position coordinates of the lines in multiple photos. The profiles are put
together to create a 3D image while the object is moving. Since the height profile
acquisition requires object movement, the method is a scanning technology.

PAGE 11
Field of Application:
Because structured light is incorporated in the method, there are rarely issues with tricky
surfaces or low ambient light. Laser triangulation delivers highly precise data even for low
contrast objects. There is one problem with laser triangulation, however: it is relatively
slow and struggles to handle the ever-increasing speeds of modern production
environments. The scanning process requires that the object being measured comes to a
stop so that changes to the laser line can be recorded. Typical applications include
log/board/veneer wood inspection, electrical components/ solder paste inspection, and
food and packaging inspection applications.

Advantages:

• No need for ambient light

• High detail resolution and accuracy
• Relatively short measurement range
• “Micrometer to mm” resolution scalability
• No additional scanning needed for moving object.

Limitations:

• Occlusion (shadow effects)

• Laser speckles
• Not suitable for large outdoor applications (~ > 1 m FOV)
• Not snapshot

PAGE 12
Why are 3D laser Profilers needed? 3D Laser Profiling vs. 2D Machine Vision:

Despite the proven advantages of 3D technology for inline quality inspection applications,
many organizations still rely solely on 2D machine vision for their quality control
processes.

While useful in a select number of scenarios, 2D vision is limited in its ability to achieve
100% quality control—which is why it is so important for organizations to invest in a
smart 3D solution. To illustrate this point we should compare the basic capabilities of a 3D
laser profiler to a 2D vision sensor.

3D Shape Measurement:

A commonly overlooked problem is that

2D sensors do not support measurement
related to 3D shape. As a result, they aren’t
able to measure critical features such as
object flatness, surface angles, or part
volumes. The inability to measure 3D shape
is a major disadvantage of 2D considering
most organizations need to inspect every
aspect of the target.

On the other hand, a 3D laser profiler

produces 3D shape information in the form
of profiles or surfaces. These profiles
provide critical information for determining if a part meets key tolerances for surface
quality, assembly, fit and finish.

Apparent Size:

Because 2D sensors cannot detect

depth, they are highly dependent on the
distance from the camera to the target.
This means 2D sensors have to be
precisely fixtured at a fixed distance
from the target, and require the use of
scale-invariant feature detection or
large tele-centric optics (that must
match the size of the sensor’s FOV) to
counteract motion effects along the
optical axis.

PAGE 13
It’s a different story with 3D. The depth measurement information provided by 3D laser
profilers eliminates errors due to object movement. This means objects can move
anywhere within the sensor’s measurement range and still yield accurate results—
effectively eliminating object fixturing requirements and improving overall measurement
reliability.

Contrast-Based Measurements:

2D sensors measure an object’s contrast

(edge data), which means 2D relies on
lighting and color/greyscale variation to
detect features. This is especially
problematic when inspecting low contrast
objects where key features are the same
color as the background. For example, 2D
fails to measure a black object on a black
background or distinguish part features
without specific lighting to expose the
presence and definition of an edge.

Unlike 2D intensity imaging 3D is contrast invariant. This means shape is measured

regardless of surface color, which makes 3D ideal for measuring low contrast objects. And
with 3D you don’t have to worry about ambient lighting or shadows affecting your scan
results.

Application examples include:

• Scanning random parts • Scanning packaging with changing

images/photos/text
• Scanning for a range of product colors
• Isolating a scan of an object from a busy
background

Robotic Inspection Applications:

PAGE 14
 Industrial robots work in a three-
dimensional world. Unfortunately,
2D technology is unable to provide
the necessary depth and spatial
information (in 6 degrees of
freedom) for the growing number
of vision-guided robotics (VGR)
systems used for automated quality
control.

A 3D laser profiler (as well as a

snapshot sensor) provides vision to
a robot, allowing it to sense
variations in its physical environment and adapt accordingly. This increases the robot’s
flexibility, utility and speed in essential applications such as pick-and-place. On top of
vision-guidance, laser profilers also provide built-in scan, measurement, and control
features required for a wide range of flexible robotic inspection applications.

Combining 2D and 3D:

Laser profilers combine 3D and 2D capability for total quality inspection. In addition to 3D
shape measurements, the intensity of the projected laser or LED light is used to create a
2D image of the surface of a part. This information can be used to extract surface markings
like bar codes and printed text.

3D height map of a PCB captures shape data. Accompanying 2D intensity image captures surface
markings such as barcodes and printed text.

PAGE 15
 LIST OF LASER PROFILE CAMERAS AND THEIR SPECIFICATIONS:

Camera Model Manufacturer Type of camera and Resolution and Scan Interface
and sensing Range Rate
Manufacturer
1. Model: DS1300 Cognex 3D Laser Measurement range Up-to 10 Gigabit
Applications : Displacement Sensor: 725 mm KHZ Ethernet
Measure laser profiler Resolution X (mm): interface
Heights, Verify 0.101–0.457 Integrate
Volumes, Laser Class: 2M Resolution Z (mm): d link
Determine 0.016–0.265
Presence/
absence of
components,
Identify surface
defects, Optical
character
recognition.
2. Model: DS1050 Cognex 3D Laser Measurement range Up-to 10 Gigabit
Applications : Displacement Sensor: 76 mm KHZ Ethernet
Measure laser profiler Resolution X (mm): interface
Heights, Verify 0.059–0.090 Integrate
Volumes, Laser Class: 2M Resolution Z (mm): d link
Determine 0.004–0.014
Presence/
absence of
components,
Identify surface
defects, Optical
character
recognition

3 Model: DS925B Cognex 3D Laser Measurement range Up-to 1.2 Gigabit

Applications: Displacement Sensor: 25 mm KHZ Ethernet
Measuring laser profiler Resolution X (mm): interface
component 0.0183-0.0227
height, tilt and Laser Class: 2M Resolution Z (mm):
spacing, 0.002
Verifying depth,
Identifying
surface defects,
Inspecting
quality and
quantity of glue.

PAGE 16
4 Model : Gocator LMI technologies 3d laser Point Profilers Measurement range Up-to 32 Gigabit
1390 2000 mm KHZ Ethernet
Applications: Laser Class: 3B Resolution X (mm):
surface Not Applicable
roughness, obje Resolution Z (mm):
ct thickness. .0250

5 LMI technologies 3d laser Point Profilers Up-to32 Gigabit

Model : Gocator Measurement range KHZ Ethernet
1320 Laser Class: 3R 20 mm
Applications: Resolution X (mm):
surface Not Applicable
roughness, obje Resolution Z (mm):
ct thickness. .0004-.0004

6 Model: Gocator LMI technologies 3d laser Point Profilers Measurement range Up-to Gigabit
2380 6 mm 5KHZ Ethernet
Applications: Laser Class: 3R Resolution X (mm):
SCAN 0.0058 – 0.0062
MICRON- Resolution Z (mm):
LEVEL DETAILs 0.0011

7 Model: Gocator LMI technologies 3d laser Point Profilers Measurement range Up-to Gigabit
2410 800 mm 5KHZ Ethernet
Applications: Laser Class: 3R Resolution X (mm):
SCAN 0.375 - 1.100
MICRON- Resolution Z (mm):
LEVEL DETAILs 0.092 - 0.488

8 Model: Gocator LMI technologies 3d laser Point Profilers Measurement range Up-to Gigabit
2880 800 mm 2.5KHZ Ethernet
Applications: Laser Class: 3B Resolution X (mm):
scan micron- 0.375 - 1.100
level detail with Resolution Z (mm):
dual camera 0.092 - 0.488
scan

PAGE 17
Machine Vision Lighting:

1. Introduction

The one aspect of vision system design and implementation that consistently causes more
delays, cost overruns, and general consternation than other considerations, is lighting.
The key aspects of solutioning for machine vision lighting are:
Understanding of lighting types and application advantages and disadvantages, vision
camera and sensor quantum efficiency and spectral range, illumination techniques and
their application fields relative to surface flatness and surface reflectivity.
Familiarity with the major four factors of vision illumination: geometry, pattern or
structure, wavelength, and filters.
Detailed analysis of the immediate inspection environment (physical constraints and
requirements).
Appropriate lighting needs to deliver three acceptance criteria :
Maximize the contrast on those features of interest
Minimize the contrast elsewhere
Provide for a measure of robustness.

2. Vision Illumination Sources and Spectral Content

Fluorescent, quartz halogen, and LED are the most widely used lighting types in machine
vision, particularly for small- to medium-scale inspection stations. Metal halide, xenon,
and high-pressure sodium are more typically used in large-scale applications or in areas
requiring a very bright source. Metal halide, also known as mercury, is often used in
microscopy because it has many discrete wavelength peaks, which complements the use of
filters for fluorescence studies. A xenon source is useful for applications requiring a very
bright strobe light. Figure below shows the advantages and disadvantages of fluorescent,
quartz halogen, and LED lighting types and relevant selection criteria, as applied to
machine vision. For example, whereas LED lighting has a longer life expectancy, quartz
halogen lighting may be the choice for a particular inspection because it offers greater
intensity.

PAGE 18
In recent years, LED technology has improved in stability, intensity, and cost-
effectiveness; however, it is still not as cost-effective for large area lighting, particularly
compared with fluorescent sources. However, if application flexibility, output stability,
and longevity are important parameters, then LED lighting might be more appropriate. In
those applications requiring high light intensity, such as high-speed inspections, it may be
useful to match the source’s spectral output with the spectral sensitivity of the particular
vision camera. For example, CMOS sensor-based cameras are more IR sensitive than their
charge-coupled device (CCD) counterparts, imparting a significant sensitivity advantage
in light-starved inspection settings when using IR LED.

3. The Cornerstones of Vision Illumination:

1. Geometry—The 3-D spatial relationship among sample, light and camera

2. Structure or Pattern—The shape of the light projected onto the sample
3. Wavelength or Color—How the light is differentially reflected or absorbed
by the sample and its immediate background
4. Filters—Differentially blocking and passing wavelengths and/or light
directions

Contrast of the object in focus can be manipulated using the above four actors. Effecting
contrast changes through geometry involves moving the sample, light, and/or camera
positions until a suitable configuration can be found. For example, a coaxial ring light (one
mounted around the camera) may generate hotspot glare on a semi-reflective barcode
surface, but by simply moving the light off-axis, the hotspot glare is also moved out of the
camera’s view. Contrast changes through structure or the shape of the light projected on
the sample is generally lighting technique–specific. Contrast changes through color
lighting are related to differential color absorbance versus reflectance.

4. Considerations for an Optimal Lighting Solution:

With respect to the lighting environment, there are two aspects to evaluate when
determining the optimal lighting solution:

Immediate Inspection Environment: Fully understanding the immediate inspection area’s

physical requirements and limitations, in a 3D space, is critical.
Ambient Light Contribution: The presence of ambient light input can have a tremendous
impact on the quality and consistency of inspections, particularly when using a
multispectral source such as white light. There are three active methods for dealing with
ambient light: (1) high-power strobing with short duration pulses, (2) physical enclosures,
and (3) pass filters.
Sample/Light Interactions: How a sample’s surface interacts with task-specific and
ambient light is related to many factors, including the gross surface shape, geometry, and
reflectivity as well as its composition, topography, and color.
Color Analysis: Materials reflect and/or absorb various wavelengths of light differentially,
an effect that is valid for both black and white and color imaging space. Like colors reflect
and surfaces are brightened; conversely, opposing colors absorb and surfaces are
darkened.

PAGE 19
5. Illumination Techniques

Illumination techniques comprise back lighting, diffuse (also known as full bright field)
lighting, bright field (actually partial bright field or directional) lighting, and dark field
lighting.

The application of some techniques requires a specific light and geometry, or relative
placement of the camera, sample, and light—others do not. For example, a standard
bright field bar light may also be used in dark-field mode; whereas a diffuse light is used
exclusively as such.

Back Lighting:
Back lighting generates instant contrast as it creates dark
silhouettes against a bright background. The most common
uses are for detecting the presence/absence of holes and gaps,
part placing or orientating, or measuring objects. Often it is
useful to use a monochrome light, such as red, green, or blue,
with light control polarization if precise (subpixel) edge
detection becomes necessary.

Diffuse (Full Bright Field) Lighting:

Diffuse, or full bright field lighting, is most commonly used on
shiny specular or mixed reflectivity samples. There are three
primary types of diffuse lighting, with hemispherical
dome/cylinder or on-axis being the most common.
Diffuse dome lights are effective at lighting curved, specular
surfaces, commonly found in the automotive industry, for
example. On-axis lights work in a similar fashion for flat
samples and are particularly effective at enhancing
differentially angled, textured, or topographic features on relatively flat objects. To be
effective, diffuse lights, particularly dome varieties, require proximity to the sample.

Partial Bright Field or Directional Lighting:

Partial bright field lighting is the most commonly used vision
lighting technique, and is the most familiar lighting used every
day, including sunlight. This type of lighting is distinguished
from full bright field in that it is directional, typically from a
point source and, because of its directional nature, it is a good
choice for generating contrast and enhancing topographic detail.

Dark Field Lighting:

Dark field lighting is similar to the principle car headlights work on.

PAGE 20
6. Application Fields:

Lighting Technique Application Fields: Surface Shape Versus Surface Reflectivity Detail

Modular Ring-light from

TPL Vision

PAGE 21
Lenses for Camera

How to select a lens for an industrial camera:

Usually an industrial camera does not have a lens. Instead, they come with a C-mount or
CS-mount. As these mounts are standardized, there is a wide range of lenses available for
industrial cameras. But choosing a lens for a particular solution takes a few steps of
calculation and selection based on the use case and ambience.

Step 1: Determining the size of the object:

The goal of this step is to represent an object
on the image sensor of an industrial
camera.

Step 2: Determining the size of the sensor:

Format: 1/4" 1/3" 1/2" 2/3"
1"
Height: 2,4 3,6 4,8 6,6
9,6
Width: 3,2 4,8 6,4 8,8 12,8
From resolution of the camera calculation,
we need to determine the sensor size and
format. The common formats and the sensor
sizes for them are given above.

Step 3: Is it in the macro range?

If the object size is smaller than or equal to
the size of the sensor, it is almost impossible
to take pictures using a normal C-mount
lens. In such cases, a macro lens or even a
microscope is required.

Step 4: Determining the working distance:

Determining the working distance is easy: At
least, if the object is flat and aligned in
parallel to the image sensor. In this case, the
working distance is the distance between the
object and the leading edge of the lens.

Step 5: Portrait or landscape?

If the aspect ratio of the object and the
sensor do not fit, the object long side must
be mapped to sensor’s longer side to
maximize sensor area usage.

PAGE 22
Step 6: Calculating the focal length:
The Focal length is the central parameter of a lens. We can calculate it using the following
formula:

The chosen focal length from the available

list of lenses should be smaller than the
calculated focal length.

Step 7: Is the focal length less than 5 mm or greater than 75 mm?

In these cases, special lenses are required. If the focal length is less than 5 mm, we need a
wide-angle lens. If it is greater than 75 mm, we need a telephoto lens.

Step 8: Does the format fit?

The format of a lens has to be always larger than that of the sensor.
Focal length MOD
Product code Format
[mm] [m]

H0514-MP 1/2" 5 0,1

M0814-MP 1/2" 8 0,1

H1214-M(KP) 1/2" 12 0,25

C1614-M(KP) 2/3" 16 0,25

C2514-M(KP) 2/3" 25 0,25

C3516-M(KP) 2/3" 35 0,35

C5028-M(KP) 2/3" 50 0,90

C7528-M(KP) 2/3" 75 0,70

Step 9: Adapting the MOD:

MOD stands for Minimal Object Distance. A lens can focus between the MOD and
infinity. In this case, the minimal working distance can be decreased by placing extension
rings between the lens and the industrial camera.

Basler, Zeiss, Tamron, Thorlabs, Aegis, Kowa are some of the leading lens vendors for
industrial cameras and machine vision.

PAGE 23
Special types of Lenses:

Vario lens
Vario lenses allow the manual variation of their focal length. Colloquially, this is called "to
zoom". But actually, vario lenses are not zoom lenses. If their focal length is varied,the
focus also has to be readjusted. The advantage of vario lenses: small, lightweight and low-
priced.
It is recommended to use vario lenses, if the machine vision application requires a short
focal length (wide angle).
Beyond their variable focal length, vario lenses are selected according to the rules that
apply to "normal" C-mount lenses.

Zoom lens
Zoom lenses are advanced vario lenses. We can vary their focal length (colloquially, "to
zoom" the lens) and they remain focused. In the case of manual zoom lenses this is
ensured opto-mechanically. Therefore, they are quite big, weighty and expensive.
In the case of industrial USB cameras, FireWire cameras or GigE cameras with integrated
lenses, the readjustment of the focus is performed by software.
Important: Besides their size and weight, manual zoom lenses have another
disadvantage. Their minimum object distance is usually one meter or even more. In
contrast to fixed focus C-mount lenses, the minimum object distance cannot be reduced
by extension rings: The lens would lose its capability to readjust the focus after it has been
zoomed.

Auto-focus lens
In the vast majority of machine vision applications, the basic optical parameters object
size and working distance are fixed. In such cases, C-mount lenses with a fixed focal
length and a manually adjustable focus are used.
If, however, the object size and/or the working distance are variable, the lens' focus has to
be adapted quickly to these variations. Ideally, the application software performs this
adaption based on an auto-focus algorithm. Hence, the focus mechanism of the lens must
have a motor so that it can be adjusted.

Auto-iris lens
One of the most important requirements for the realization of a simple machine vision
system is stable illumination. If this cannot be guaranteed, an auto-iris lens may be
helpful. Like the human eye, it adapts itself to environmental illumination, by opening
and closing its iris.
Outdoor surveillance cameras are fitted with auto-iris lenses by default. These lenses are
controlled directly by the surveillance camera. In machine vision applications, this direct
coupling is seldom acceptable. Thus, in the context of industrial cameras, the application
software should be responsible for the auto-iris control.

PAGE 24
 LIST OF AUTOFOCUS AND ZOOM CAMERAS AND THEIR SPECIFICATIONS

Camera Model Manufacturer Type of camera Resolution Frame Interface

and and sensing Rate
Manufacturer
1 Model: DFK The Imaging Source CMOS, Area Scan 2,592×1,944 (5 15 fps Gigabit
. Z12GP031 Motor zoom: 4.8 MP). Ethernet
mm (wide) to 57.6
mm (telephoto) Format: 1/2.5”
Focus: automatic
and manual
Iris: automatic
and manual
Lens: integrated
2 Model: DFK The Imaging Source CCD, Area Scan 1,280×960 (1.2 30 fps Gigabit
. Z12G445 Motor zoom: 4.8 MP) Ethernet
mm (wide) to 57.6
mm (telephoto) Format: 1/3”
Focus: automatic
and manual
Iris: automatic
and manual
Lens: integrated

GPU - graphics processing unit

A GPU, or graphics processing unit, is used primarily for 3D applications. It is a single-
chip processor that creates lighting effects and transforms objects every time a 3D scene is
redrawn. These are mathematically-intensive tasks, which otherwise, would put quite a
strain on the CPU.

One of the common use-cases of GPU is Video editing; while some powerful CPUs can
handle basic video editing, if you're working with large amounts of high-resolution files -
particularly 4K or 360-degree video - a high-end GPU is a must-have in order to transcode
the files at a reasonable speed.

PAGE 25
What to look for in a GPU?
There are main characteristics of a GPU related to DL are:

• Memory bandwidth — as discussed above, the ability of the GPU to handle large
amount of data. The most important performance metric.
• Processing power —indicates how fast your GPU can crunch data. We will
compute this as the number of CUDA cores multiplied by the clock speed of each
core.
• Video RAM size — the amount of data you can have on the video card at once. If
you are going to work with Computer Vision models, you want this to be as large
as affordable. Especially, if you want to do some CV Kaggle competitions. Amount
of VRAM is not so crucial for Natural Language Processing (NLP) and working
with categorical data.

IMAGE PROCESSING ON GPU :

Image processing algorithms usually consume a lot of computing resources. In many

cases the continuously growing performance of CPUs found in powerful PCs is sufficient
to handle such tasks within the specified time. However, leading vendors of image
processing hardware and software are constantly on the search for faster ways of
improving speed beyond that possible on the PC’s CPU.

Typical methods of increasing speed in image processing include the distribution of the
computing tasks between multiple multi core processors, or also the use of specialized
FPGAs. Each of these technologies has its own advantages and disadvantages, but all
have one aspect in common in that they generally do not use the fastest available
processor in the system which is optimized for imaging algorithms namely the processor
on the graphics cards, also known as the GPU (Graphical Processing Unit).

These "racers" among the processors have an incredible development history. The
evolution has been principally driven by the gaming industry, where the requirements
demanded of the graphical representation of game scenes and animations have greatly
increased. Sales of Millions of games consoles have contributed to the demand, resulting
in large numbers of GPUs and corresponding profits to further boost the development of
graphics components. Other industrial sectors are now reaping the benefits, including
image processing.

Graphics processors outperform other imaging acceleration methods in many technical

aspects, even compared with the fastest available FPGAs (see Table). For example, they
are clocked at rates 10 to 20 times faster than that of typical FPGAs, so that in

PAGE 26
combination with larger memory options can achieve data throughput rates of up to 500
times greater than those of standard FPGAs.

However, these increased speeds are not fully available to image processing users – the
outsorcing of the algorithms to the GPU causes a delay in the data flow, from image
capture to data processing. Regardless of this effect, various analyses of intense
computing operations indicate a rise in performance by a factor of 2 to 10 when using a
GPU in place of a CPU, while the CPU can then be used at for other tasks
simultaneously.

There are two principal reasons as to why the GPU technology has only recently become
available for image processing. On the one hand, until recently graphics cards had
different processors for different tasks. The situation has now changed with the latest
graphics processors, such as the GeForce 8800 from Nvidia or equivalents from
companies such as ATI. Some of the 681 million transistors on the GeForce 8800
processor can be dynamically allocated for operations such as geometry or pixel
computation. On the other hand, the PCI link allows fast data transfer between host and
VGA card.

"The architecture of a graphics chip is always very complex," explains Martin Kersting,
Head of Development at STEMMER IMAGING. "However, the DirectX-API and the
High Level Shader Language (HLSL) compiler from Microsoft together with a handful of
functions in our Common Vision Blox software library enable image processing software
developers to transfer images between the host and GPU, and therefore use all processors
in the system in an optimum way."

As already mentioned, data transfers between the VGA card and GPU cause a certain
delay between capturing the image and processing the data on the graphics card. Kersting

PAGE 27
describes the advantage of the technology as follows: "Skillful use of GPU image
processing can mean that special hardware is not even needed in applications with
extremely high data throughputs."

To bring the benefits of GPU image processing to developers in the most effective way,
the developers at STEMMER IMAGING – the image processing experts based in
Puchheim, Germany – have now integrated the functionality in the Common Vision Blox
(CVB) software library from the company. "To do so, we added several new functions to
CVB that can be called from a CVB application that can be accessed without any
additional GPU programming experience," states Kersting.

These functions currently implement tasks such as image filtering, point operations
between two images, parallel processing of four monochrome images, transformations
from RGB to HSI and from Bayer to RGB formats, so-called flat field corrections,
rotation and scaling of images. To optimize the image data transfer between main
memory and GPU it is also possible to combine several algorithms within the graphics
card by using the open programming possibilities of the HLSL language.

Kersting and his team have carried out multiple tests on possible increases in speed with
the new technology. For example, images captured with a monochrome CCIR camera
such as the JAI A11 using a PC-based system with an Nvidia 8800 graphics card were
upscaled to 2K x 2K pixels and displayed on a PC monitor. At the same time, the
graphics processor computed a 3x3 Sobel filter at a rate of 30 pictures/second. In a direct
comparison between a 2.4 GHz Intel Core 2 Duo processor and the Nvidia 8800, both
units computed a 5x5 filter.

The result impressively demonstrate the possibilities of the cooperation between

Common Vision Blox and the new GPU technology: The Nvidia 8800 was about five
times faster than the CPU (see Fig. 2).

PAGE 28
 Figure 2: In a direct comparison between a 2.4 GHz Intel Core 2 Duo processor and an
Nvidia 8800 graphics card, the GPU completed the set image processing task about 5
times faster than the CPU

GPU Solutions for fast image processing

GPU computing technologies and algorithms are proven to deliver impressive

performance in demosaicing, image resize and compression, image filtering, high-
speed imaging, encoding and decoding tasks. We develop a range of solutions to
integrate with your existing system or to customize to the specific needs upon
request. Check out this list for available solutions that we will be happy to tailor to
fit your specifications.
We can offer custom solutions for ultra fast compression applications in various
fields:
• High speed imaging
• Fast resize and recompression for photo and web applications
• Mobile imaging
• Medical imaging

Fast compression and decompression solutions on GPU

• Fast JPEG codec for lossy image compression and decompression
• Fast MJPEG video codec
• Realtime JPEG compression for high speed camera
• Fast Demosaic JPEG (combined solution for demosaicing and jpeg compression on
GPU)
• Fast decompress-resize-compress solution for web applications
• JPEG compression for raw images on CUDA
• J2K encoding and decoding on GPU
• Custom solutions for fast compression and decompression
• Current projects in the field of ultra fast compression

Fast image processing solutions on GPU

• Fast image demosaicing (Bayer CFA) on CUDA
• Fast 2D DWT on CUDA
• Image trigger for high speed imaging
• Image processing for high speed and high resolution color cameras
• Fast image rotate, flip/flop, crop and resize
• Histogram on CUDA
• Video Wall
• Fast web resize
• CUDA scaling and tiling
• Fast Gaussian blur on GPU
• Realtime image and video processing on Tegra K1, Tegra X1 and Tegra X2

PAGE 29
• 3D LUT Transform on GPU
• Color correction with ICC profiles on GPU
• Adaptive denoising on CUDA
• RAW/DNG realtime image processing on CUDA
• Defringe on CUDA
• Custom solutions for fast image processing

CAMERA SOLUTIONS ON GPU

• Realtime image processing on GPU for XIMEA xiQ (USB-3.0), XIMEA xiD (USB-3.0)
and XIMEA xiB (PCIe) cameras
• Raw Bayer Codec
• GPU Direct technology for camera applications on Linux
• GPU image processing for high speed cameras

GPUs Comparison

Here is performance comparison between all cards. Check the individual card
profiles below. Notably, the performance of Titan XP and GTX 1080 Ti is very close
despite the huge price gap between them.

PAGE 30
 The price comparison reveals that GTX 1080 Ti, GTX 1070 and GTX 1060 have great
value for the compute performance they provide. All the cards are in the same league
value-wise, except Titan XP.

List of GPU’s for Image Processing

GPU MODEL MANUFACTURER SPECIFICATIONS

Quadro GV100 NVIDIA GPU Memory – 32Gb HBM2

Memory interface – 4096 - bit

Memory bandwidth – upto 870 Gb

NVIDIA Cuda Cores – 5120

NVIDIA Tensor cores – 640

Display Resolution - 4x 4096x2160 @ 120 Hz,4x 5120x2880 @ 60

Hz 2x 7680x4320 @ 60 Hz

PAGE 31
NVIDIA p600 NVIDIA GPU Memory – 24Gb GDD5RX

Memory interface – 384 - bit

Memory bandwidth – upto 432 Gb/s

NVIDIA Cuda Cores – 3840

Display Resolution - 2560x1600 @ 60 Hz

7680x4320 @ 30 Hz

NVIDIA NVIDIA GPU Memory – 16 Gb HBM2

QUADRO GP100
Memory interface – 4096 - bit

Memory bandwidth – upto 717 Gb/s

NVIDIA Cuda Cores – 3584

Display Resolution - 4x 4096x2160 @ 120Hz

4x 5120x2880 @ 60Hz

MALI-G72 ARM Frequency -

Memory system – built in MMU to support virtual memory

Multi Core scaling – 1 to 32

Adaptive scalable texture compression – LDR and

HDR,supports both 2D and 3D images

Bus Interface -AMBA®4

ACE & ACE-LITE(supports wide range of APIs and IP
peripherals)

MALI- G31 ARM Frequency - 650 hz

PAGE 32
 Pixel/texturing throughput – 1.3 Gpix/s

Memory system – built in MMU to support virtual memory

Multi Core scaling – unipixel core or dual pixel core

Adaptive scalable texture compression – LDR and

HDR,supports both 2D and 3D images

Bus Interface -AMBA®4

and ACE-LITE(supports wide range of APIs and IP
peripherals)

NVIDIA GPU Memory – 16Gb GDDR5X

QUADRO P5000
Memory interface – 256 - bit

Memory bandwidth – upto 288 Gb/s

NVIDIA Cuda Cores – 2560

Display Resolution - 2560x1600 @ 60 Hz

7680x4320 @ 30 Hz

PAGE 33