1

Image Stabilization Technology Overview

David Sachs, Steven Nasiri, Daniel Goehl
InvenSense, Inc.
3150A Coronado Drive, Santa Clara, CA 95054
(408) 988-7339
www.InvenSense.com
Abstract
As Digital Still Cameras (DSC) become smaller, cheaper and higher in resolution,
photographs are increasingly prone to blurring from shaky hands. Optical image
stabilization (OIS) is an effective solution that addresses the quality of images,
and is an
idea that has been around for at least 30 years. It has only recently made its way
into the
low-cost consumer camera market, and will soon be migrating to the higher end
camera
phones. This paper provides an overview of common design practices and
considerations
for optical image stabilization and how silicon-based MEMS dual-axis gyroscopes
with
their size, cost and performance advantages are enabling this vital function for
image
capturing devices.
Introduction
Image stabilization first hit the consumer market in 1995, with the introduction
of a
stabilized zoom lens by Canon. Canon had published this idea in the United States
as
early as 1976 in a patent entitled “Image stabilizing optical system having a
variable
prism”. Since this introduction, image stabilization has appeared in an increasing
number
of products including digital single lens reflex cameras (DSLR), video cameras and
binoculars. However, optical image stabilization in the mass market “point-and-
shoot”
camera is only a recent phenomenon, with Panasonic releasing the first of these
cameras
in 2004. Since this introduction, integration of image stabilization has been
brisk, with
several cameras launched in 2005 and marketing promotions ranging from prime-time
television commercials to exclusive sections in big-box electronics stores devoted
to
displaying cameras with this feature. Now every DSC maker either has several
models
with IS already in the market, or is planning to have fairly soon.
Market
Optical image stabilization (OIS) is quickly on its way to becoming a standard
feature.
The Gartner Group estimates that 96 million digital cameras will be sold worldwide
in
2006. InvenSense predicts that 30 percent of these will have OIS. By 2009, the
expected
penetration will be close to 90 percent of the 100 million unit plus digital
camera market.
In the high resolution camera phone market (3 megapixels and greater), there is an
even
greater need for optical image stabilization. Adoption is already underway, and
the
number of cameraphones will quickly surpass the total number of digital cameras
over
the next several years. InfoTrends anticipates that nearly 45 million handsets
with 3
2
megapixel cameras and higher will be shipped in 2006 and over 330 million handsets
by
2009. Initially, less than 5 percent will have image stabilization, but as the
size and cost
for key components of the optical image stabilization system decreases, experts
believe
that the penetration will rapidly rise to over 50% of the market.
Needed Solution
As DSC’s shrink in size and weight, migrate to higher megapixel sensors, and
incorporate 3x- and -4x optical zooms, engineers struggle to provide better
quality
images in lower lighting conditions without a color-distorting flash. Virtually
all trends in
photography have pushed cameras toward longer exposure times and higher
resolutions
which have largely resulted in image quality problems from shaky hands.
The human body is always in motion. No photographer is capable of holding a camera
with outstretched hands without moving it to some degree; this movement is a
biological
phenomenon that cannot be easily defeated by training. Human hands shake with a
frequency of 10 to 20 Hz and, when outstretched, will also drift at lower
frequencies. In
photography, this jittering produces an effect often referred to as “camera
shake”, and can
cause blurring in photographs. Modern consumer cameras tend to amplify this effect
by
being small and lightweight, and by replacing the traditional viewfinder with an
LCD
screen, encourages users to hold their
cameras with extended arms. In digital
photography, in which image sensors are
arrays of pixels, this effect can be ignored
if it produces a blur over one pixel or less.
The amount of blur that will be produced
by hand jitter for a certain picture depends
on the exposure time and the angular
field-of-view corresponding to one pixel.
If an image sensor is exposed to light only
for a short amount of time, the camera
may not move enough to blur the image;
however, zoom, higher pixel densities and
smaller movements will produce
significant blur.
Figure 1. Anti-Shake demonstration by Konica Minolta
Figure 2. Micron's CMOS pixel array
3
The exposure time required for enough light to be absorbed in an image sensor
depends
on the amount of light present during the exposure, the aperture through which the
light
must travel, and the design of the sensor itself. A camera system designed for an
action
shot will have film that exposes very quickly, and a wide aperture for letting in
as much
light as possible, allowing the shutter to be open for as little as 1ms. In
contrast, a 3
megapixel camera module designed for a cell phone will have a small aperture, as
the
phone itself must be small. Sensors with higher pixel densities must have smaller
pixels,
which will absorb light more slowly for a given aperture. In addition, most of
these
sensors are based on CMOS technology, in which transistors are integrated in the
sensor
along side the pixel; the part of the pixel that absorbs light is even smaller, to
make room
for this additional circuitry. This is usually expressed with a “fill factor”, the
ratio of the
actual light-absorbing area to the area occupied by the entire pixel. In Figure 2,
only the
photodiodes actually absorb light. Ultimately, when such a system is used in an
indoor
location with a typical lighting condition, the exposure time can easily be over
100ms.
Figure 3 shows two seconds of hand jitter as recorded by an IDG-1000, InvenSense’s
dual-axis gyroscope. It also shows 100ms of this data that has been isolated and
integrated, giving a typical angular path of a camera during an exposure.
Figure 3. Left: Yaw Hand Jitter data, 1s. Right: Pitch and Yaw integrated data,
100ms.
Typical low-cost camera systems have a field of view of 35º to 45º along the
horizontal.
For a 640x480 VGA camera with a field of view of 45º, each pixel corresponds to a
horizontal angle of 0.07º. The derived camera path shown in Figure 3 has a total
horizontal drift of 0.08º, producing a negligible blur of 1.1 pixels in such a VGA
image.
However, as an example, if the pixel density is increased to 2 megapixels
(1600x1200),
the horizontal blur becomes 2.9 pixels. Such a blur will begin to be noticeable,
though the
impact it has on a viewer can, to some degree, be reduced by artificial image
sharpening
algorithms. Pixel densities of 3 and 6 megapixels simply allow greater resolution
of this
blur, revealing it to 3.6 and 5 pixels respectively. In fact, it is clear that,
for such slow
4
shutter speeds, increasing resolution beyond 2 megapixels accomplishes nothing
more
than enlarging the blur caused by hand jitter. Note that these numbers were
derived
without any optical zoom, which would amplify the movement of the camera, and
produce even more blurring. Photography in conditions with low lighting would
extend
the necessary exposure time, introducing additional blur.
Optical image stabilization reduces any hand jitter that is within a given range,
usually
±1º, certainly enough to defeat the hand movement shown in Figure 3. Figure 4
shows an
experiment involving several images that were acquired with and without image
stabilization. The improvement in image quality is impressive.
Figure 4. Minolta A1, 51mm lens, 1/15th second shutter speed. Top: OIS off.
Bottom: OIS On.
As consumers demand higher resolution cameras with optical zoom, packed into
smaller
and lighter DSC’s and cell phones, the blurring due to hand jitter will become
even more
noticeable and stabilization systems will become standard features.
Photography Terminology
Aperture sizes are measured in “f-numbers”,
which refer to the ratio of the lens focal
length to the diameter of the hole, as both
influence the amount of light per area that
will be incident on the film or image sensor.
The film speed is measured using a scale
known as “ISO”, defined by the International
Organization for Standardization, though the
application of this metric to digital sensors is
sometimes vague. Since the blur generated
depends on a multitude of factors, a metric
called a “stop” is often cited in photography
to refer to changes in aperture, shutter speed, or film speed. For example,
increasing the
Aperture Shutter
Speed
Film Speed
f-32 1/8 ISO-25
f-22 1/15 ISO-50
f-16 1/30 ISO-100
f-11 1/60 ISO-200
f-8 1/125 ISO-400
f-5.6 1/250 ISO-800
f-4 1/500 ISO-1600
f-2.8 1/1000 ISO-3200
Table 1. Stops in photography.
5
shutter speed from 1/30th of a second to 1/60th of a second is referred to as
increasing it
“one stop”, and will reduce the final exposure of the image; but decreasing the
aperture
size from f-11 to f-16 will allow half as much light in, having a similar effect,
and is
described as reducing the aperture one stop. If the film used has a speed of ISO-
200, the
film speed can be reduced by one stop by replacing it with a slower film with a
speed of
ISO-100. Some commonly used stops are shown in Table 1.
Evaluating Blur
With this metric, a simple evaluation of an OIS system can be done that takes into
account alterations of any of these parameters, and differences between cameras. A
good
OIS system is usually said to enable 2 to 3 stops of exposure. This means that if
the
exposure time is increased until blur starts to appear, and then an OIS system is
activated,
the blur will disappear; and the exposure time can now be increased an additional
2 to 3
times before blur stars to reappear. This is a relative measurement, as the actual
allowed
exposure time will depend on the other parameters.
Optical Image Stabilization
Blur due to hand jitter is reduced by mechanically stabilizing the camera. A two
axis
gyroscope is used to measure the movement of the camera, and a microcontroller
directs
that signal to small linear motors that move the image sensor, compensating for
the
camera motion. Other designs move a lens somewhere in the optical chain within the
camera. A typical high-level block diagram is shown in Figure 5.
Figure 5. OIS High level block diagram.
With either method, the result is that the body of the camera may shake, but light
strikes
the pixels of the image sensor as though the camera were not shaking.
6
Sensor Requirements
For an OIS system to function
properly, the sensors, actuators, and
electronics must be carefully chosen.
A newcomer to this field may
immediately wonder why gyroscopes
are used in image stabilization, rather
than other sensors, such as
accelerometers. A gyroscope
measures the rotation about an axis, where rotations about X, Y, and Z axes for a
given
object are referred to as roll, pitch, and yaw, depicted in Figure 6. Gyroscopes
are perfect
for the job because they measure rotation, which dominates linear movement in its
effect
on picture taking, at relatively fast rates, ~100 to 150 Hz,
A camera creates an angular representation of the world, mapping everything within
its
angular field of view onto an image sensor. In such optical systems, the linear
motion
becomes negligible for far away objects, but rotation must always be taken into
account.
The diminishing effect of translation on the angular field of view of a scene can
be
visualized with the diagram in Figure 7.
Figure 7. Left: Impact of Translation on FOV. Right: Impact of Rotation on FOV.
Note that when the camera’s frame of reference moves linearly, as in the diagram
on the
left, objects at A1, A2, and A3 find themselves at different angles within the
camera’s
field of view. This effect is known as parallax. Objects that are further away
will appear
to move less within the angular field of view of the camera. In the diagram on the
right,
in which the camera has rotated, all three objects have the same angular
displacement
regardless of their distance from the camera. The impact of linear movement
decreases,
but the impact of rotation does not. As an example, a 0.08º rotation of the camera
will
cause objects in its field of view to be displaced by 0.08º. To achieve this same
displacement of an object 10 meters away with a translation of the camera, the
camera
A1
A2
A3
y y`
x x`
θ1
θ2
θ3
y
y`
x
x`
A1
A2
A3
θ1, θ2, θ3
Figure 6. Roll, Pitch, and Yaw, from NASA.
7
would have to move 10m•tan(0.08º), or about 1.4cm. This would require an
unnaturally
excessive hand jitter.
Linear jitter is only damaging in the field of macrophotography, which involves
close-up
photographs, typically of insects and plants; here, the objects may not be far
away
relative to the magnitude of the linear jitter. Effects due to translation can be
reduced
when using lenses designed for macrophotography, which have large magnification
factors that allow the photographer to be further from the object being captured.
Note that
in practice, there are no consumer cameras that compensate for any motion other
than
pitch and yaw. Blurring due to roll and linear movements is small enough that it
can be
ignored.
While accelerometers can sometimes be used to measure rotation by detecting the
acceleration due to gravity, the resulting output is noisy and coupled with linear
movement in ways that are impossible to separate without gyroscopes. In addition,
accelerometers would not be able to sense yaw. For these reason, the gyroscope has
always been fundamental to image stabilization systems.
Gyroscopes
Gyroscopes are employed in IS systems to sense
pitch and yaw with low noise and high sensitivity in
order to resolve the small movements associated
with hand jitter. Typically, these systems require a
full-scale range of +/-30 degrees per second, with at
least 10-bit resolution.
Almost all reported micromachined gyroscopes use
vibrating mechanical elements (proof-mass) to
sense rotation. They have no rotating parts that
require bearings, and hence they can be easily
miniaturized and batch fabricated using micromachining techniques. All vibratory
gyroscopes are based on the transfer of energy between two vibration modes of a
structure caused by Coriolis acceleration. Coriolis acceleration, named after the
French
scientist and engineer G. G. de Coriolis (1792–1843), is an apparent acceleration
that
arises in a rotating reference frame and is proportional to the rate of rotation,
Fig. 8.
Vibratory gyroscopes were demonstrated in the early
1980’s. An example of this type of devices is the
quartz tuning fork like the Quartz Rate Sensor by
Systron Donner. Although quartz vibratory
gyroscopes can yield very high quality factors at
atmospheric pressure with improved level of
performance, due to use of quartz as the primary
material, their batch processing is not compatible
with IC fabrication technology. In the late 1980’s, after successful demonstration
of
Figure 8. Coriolis Acceleration
Figure 9. Older gyroscope technology
8
batch-fabricated silicon accelerometers, some efforts were initiated to replace
quartz with
silicon in micromachined vibratory gyroscopes. Charles Stark Draper Laboratory
demonstrated one of the first batch fabricated silicon micromachined rate
gyroscopes in
1991.
In older image stabilization systems, two single axis piezoelectric or quartz
gyroscopes
were used with many external components to amplify and filter the signals to
achieve the
desired full-scale range. In order for image stabilization to become a standard
feature in
digital cameras and camera phones, it will be necessary for gyroscopes to be
smaller and
less costly, measure pitch and yaw simultaneously, meet more stringent shock and
vibration standards as well as achieve a lower overall system cost with more
integrated
features.
Although many companies have attempted to
commercialize MEMS gyroscopes, only a
few, namely Bosch and ADI, have succeeded
in bringing single axis solutions to market.
These gyroscopes serve high-end automotive
applications, such as vehicle stability control
and roll-over protection. With gyroscope
prices well in excess of $10 per axis and with
relatively large package sizes, these solutions
are not fit for very cost sensitive consumer
applications. InvenSense is the first and only
company to bring to market an integrated
dual-axis gyroscope using advanced MEMS
technology. InvenSense’s gyroscope is also
the first solution that provides integrated
amplifiers with programmable gain settings
and integrated reset switches that are convenient for IS applications.
One of the key advantages of InvenSense’s
solution is its patented out-of-plane,
resonating structures that enable in-plane
sensing and the integration two axes as
shown in Figure 9. Another key advantage
is its ability to achieve low cost. Patented
wafer-level packaging technology allows
for fully functional device testing at the
wafer level. This is done by combining
MEMS and ASIC wafers using a
proprietary and patented wafer bonding
technology. The wafer bonding process
provides electrical connections between the
CMOS and MEMS wafers, which also
creates a hermetic seal between the two wafers. Integrating the finished MEMS
structures
Figure 11. InvenSense resonating structure
Figure 10. InvenSense gyroscope circuitry
9
with a standard off-the-shelf CMOS ASIC avoids many of the pitfalls associated
with
trying to integrate the MEMS and CMOS processes, as with a surface micromachining
approach.
InvenSense’s IDG-1000 is the first and only integrated MEMS
dual-axis gyroscope currently in high volume production. Most
other gyro manufacturers addressing IS use Piezoelectric quartz
or ceramic technologies that are discrete solutions with packagelevel
integration of the sensor and electronics. These solutions
tend to be bigger in size, more costly to assemble and require
vacuum-sealed ceramic packaging. MEMS technology is a very
mature technology used in numerous sensors and other high
volume consumer applications, such as accelerometers and inkjet
printers.
Actuator Requirements
Actuators for OIS systems must be small,
low-power, and accurate for tiny
movements. The range of movement
required by an OIS actuator depends on
the optics of the system, but the desired
outcome is an ability to compensate for
±1º of rotation. The most common
actuator is the voice coil, an
electromagnetic linear motor, used to drive
the lens in Figure 13. In combination with
strong permanent magnets, two coils are
used to drive a platform both vertically
and horizontally. As this system inherently
creates a strong magnetic field, Hall effect
sensors can be used to track the position
of the platform by measuring this field.
Motors and position sensors can be tightly integrated in a small package, and work
together to precisely control the position of the platform.
Another implementation is shown in Figure 14,
using piezoelectric actuators to shift the image
sensor. These actuators are relatively new to this
field, but can be used with lower power, and at
lower frequencies than voice coils. Newer
systems are including other types of composite
shape-changing materials.
In either case, the actuators should
Figure 13. Nikon, shifting a lens.
Figure 14. Konica Minolta, shifting the image sensor.
InvenSense’s gyroscope
Figure 12
10
be designed to be as linear as possible, with as little cross-axis correlation as
possible.
This means that the displacement resulting from a signal delivered to the X-axis
actuator
should be proportional to that signal, and should not depend on the state of the
Y-axis
actuator. This design often requires two stacked platforms, an X-axis platform,
and a Yaxis
platform riding on the X-axis platform. This ensures that the axes can truly be
controlled separately.
Hall Sensors
When using feedback with Hall sensors, it is important to place them carefully to
prevent
cross-axis errors, so that the X-axis Hall sensor only measures movement along
that axis.
Ideally, the entire system should also be designed so that the magnetic field
changes in a
linear fashion in the vicinity of these sensors, as this will simplify the design
of the
feedback control system. However, a small cross-axis error may be calibrated out
and
accounted for in software. As Hall sensors tend to vary greatly over temperature,
the
required signal conditioning and circuitry must be flexible. Figure 15 shows an
example
taken from an OIS patent (U.S. 6,631,042):
Figure 15. Canon, Hall sensor conditioning.
These sensors require two contacts for generating a known current, and two more
contacts for measuring the deflection of that current due to a magnetic field.
Note that a
current source is used to minimize the temperature dependency, and the sensor is
sampled
using circuitry with a controllable amplification and offset, allowing calibration
to be
done on the fly. For example, by periodically driving the actuators to their
mechanical
limits and measuring the respective Hall sensor outputs, the four corners of the
OIS
system can be found, and these can be used as calibration factors. An alternative
method
is to use two complementary Halls sensors per axis, one on each side of the
system, such
that one always has a decreasing value when the other has an increasing value. The
sum,
which should be constant for a given temperature, can be used for temperature
compensation, while the difference is used to compute the position of the
actuator.
Finally, using integrated Hall sensors with internal amplification greatly
simplifies this
11
design. In practice, however, most OIS systems use raw Hall sensors with more
complex
external circuitry.
Control Algorithm Requirements
Gyroscopes output angular velocity, which must be integrated in order to provide
the
absolute angle through which a camera rotates during an exposure. It must also be
scaled
in order to convert the measured angular camera movement into a directed linear
movement of an image sensor or lens that will cancel (stabilize) the effect of the
hand
jitter effect on the image. This scale factor will depend on the fixed optical
properties of
the individual camera, but it must also be updated to account for different zoom
factors.
A zoom of 2X amplifies any movement of the camera, and the final instructions sent
to
the motors must similarly be amplified.
However, it is first necessary to filter out intentional hand movements from this
integration, which is typically accomplished with a high-pass filter. For example,
if a
photographer begins framing a shot, and the gyroscope begin integrating from zero,
and
then the photographer proceeds to rotate the camera 90º in order to frame a
different shot,
the integrated data will track this 90º rotation, and a system without a high-pass
filter
would attempt to drive the optics of the camera to compensate, pointing at the
original
shot. This would be mechanically impossible, as well as undesirable. Pre-filtering
the
gyroscope data allows the OIS system to focus on removing the unintentional
highfrequency
hand jitter, without interfering with the photographer’s ability to pan and frame
shots.
A simple implementation of this system is shown in Figure 16.
Figure 16. Gyroscope signal processing.
Note that each block has a coefficient that may be updated in real-time by a
microcontroller. The scale factor may only need to be updated while the
photographer is
altering the zoom, but the other factors might be changed continuously, depending
on the
calculated exposure time and the movements of the camera.
The time constants used in the filter depend on the context of the shot. During a
long
exposure time, lower frequency arm movements become more prominent, and a short
time constant intended to isolate a 10-20Hz vibration may not be sufficient. A
time
constant around 10s will allow the OIS system to compensate for any movements made
12
during a period of a few seconds, without the data being distorted by artifacts
produced
by the filter. However, a long time constant interferes with the photographer’s
intentional
movements. For example, when attempting to take a picture of a moving car, in
which the
car is to be sharply displayed against a blurry background, a photographer must
rotate the
camera to follow the car. An OIS system with a long time constant will
inappropriately
attempt to compensate for this movement. For such pictures taken with panning, a
short
time constant on the order of 0.1s to 1s is preferable. A panning shot may be set
as a
mode within the camera menu system, but should ideally be detected from the gyro
data
itself, giving a user greater freedom to move the camera and shoot at will.
Long time constants can become particularly annoying if the OIS system is designed
to
be active before the shot is taken. While low-power implementations will avoid
using the
OIS system until the moment the shutter opens, it may be desirable to use the OIS
during
the framing of a shot, to prevent the jitter from distracting the photographer. In
addition,
while some autofocus systems use ultrasonic ranging to measure distances to
objects,
others depend on the images acquired by the image sensor, incrementally adjusting
the
focus until the image acquired is sharp enough, as measured by an image processing
algorithm. Such systems will not function with the OIS deactivated, as none of the
images will be sharp, especially under 3X zoom or higher conditions.
The Feedback Control Loop
When using Hall sensors as feedback to position the actuator, a feedback loop must
be
designed. This is usually done in firmware in camera systems, but has been
implemented
as an analog circuit in some binocular systems. The classic PID control algorithm
can be
used, in which the actuator is driven by the Hall sensor signal by summing
proportional,
derivative, and integral components.
Figure 17. PID Block Diagram
In Figure 17, a set-point is derived from the gyroscope data as in Figure 16, and
compared to a measured process variable obtained from the Hall sensors. The
difference
between them is passed through three blocks independently, a proportional block,
an
integral block, and a derivative block. The values are summed, and used to drive
the
motor.
Proportional
Derivative
Integral
Process
Variable
Set Point + Motor
- +
13
The proportional term is the meat of the system, driving the actuator until the
set-point
and the process variable are equal. The derivative term is important for stability
with
systems that don’t have high levels of internal friction, such as the voice coils.
In
contrast, the piezo-electric actuators have a certain amount of friction built in
for aiding
with low frequency compensation; control algorithms for the piezo systems will be
quite
different. The integral term is usually important for systems in which the
accuracy of the
final position is important. OIS designers may decide to omit it, as stabilization
systems
only require accuracy in relative positioning. That is, if there is a small,
constant offset in
desired position of the motor and the true position of the motor, this will not
influence the
blur in the final image. However, due to lens aberrations, there may be image
quality
compromises in some OIS systems that can be reduced if the motors stay as close as
possible to the centers of their ranges. In such systems, it may be desirable to
remove
such small offsets, and an integral term should be used. These offsets from the
center will
also decrease the range of movement in some direction, and may reduce the cameras
ability to stabilize when movement is required in that direction.
As the hand jitter being detected is around 10-20Hz, OIS systems must iterate at
least ten
times faster in order to resolve this. However, it may be necessary to run the
actuator
control loop even faster in order to make up for non-linearities in the actuators
or Hall
sensors, as non-linear systems can be treated as linear if they are driven at a
high enough
rate. In practice, the control loop will usually be run from 500Hz to 4kHz, though
the
gyroscope need not be sampled as often.
Note that not all designers use a PID system. In some systems a lead-compensation
control algorithm is used, and in others a PDD system is used, using differences
in
position, velocity, and acceleration.
Microcontroller Requirements
In some systems, the OIS system is implemented as part of an interrupt service
routine on
the main processor; in others, the system is given its own microcontroller. This
controller
need not be excessively fancy, but must be able to run a feedback control
algorithm
somewhere in the 500Hz to 4kHz range. The math used can be 16-bit fixed point, and
can
be implemented on an 8-bit or 16-bit processor relatively easily. When using
filters with
long time constants, some 32-bit math may be necessary to handle the large numbers
that
will accumulate.
Note that the just as the gyroscope must be accurate, the measurement of time in
this
system must also be accurate, due to the integration step in the signal
processing. For
example, the integration of the gyroscope data can be implemented in software as:
α’ = α+ωΔt
The new calculated position is α’, α is the previous position, ω is the gyroscope
angular
velocity output, and Δt is the time since the last update. Any error in the
gyroscope data
will be contained in ω, but any error in the measurement of time will be contained
in Δt.
14
Either way, the product of ωΔt will contain an error that will propagate. For this
reason, it
is important that the chosen microcontroller either have an external clock signal
or crystal
to provide precision timing, or have an internal oscillator that is at least as
accurate per
unit time as the gyroscope is per unit angular velocity. Ideally, it should be
accurate
enough such that any errors in time have a negligible effect on ωΔt when compared
with
the noise in the gyroscope.
Driver Requirements
Piezo and voice coil actuators require different driving signals. Piezo systems
are
typically low current but high voltage, and must be driven with high voltage
transistors or
opto-isolators, which allow low voltage inputs to control high voltage outputs.
Voice
coils, in contrast, require low voltages but high currents, and will need a driver
with an
H-bridge output stage. Low power can be obtained by using a PWM driver, as the
inductance of the coils prevents excessive current from flowing when the driving
signal is
oscillating at a high frequency. This frequency should be well above 100kHz,
ideally
closer to 1MHz.
15
Electronic Image Stabilization
In video systems, two problems arise due to hand jitter. As each frame is recorded
by an
image sensor, just as it would be in a digital camera, the individual frames may
be blurry;
this makes high resolution video problematic. For video cameras with high pixel
densities, optical image stabilization is required for achieving the target
resolution.
However, video cameras of any resolution also face a slightly different problem.
Even if
the individual frames are not blurred from the hand movement, this shaking will
manifest
itself as a frame-to-frame jitter over time. This effect is familiar to anyone who
has
attempted to sit through a home video made on a handheld camcorder without a
stabilization system. As the problem is not the blur within the individual frames,
but the
shift from one frame to the next, it may not be necessary to adjust the optics of
the system
to solve the problem.
Instead, a system can be used that is known as electronic image stabilization
(EIS).
Gyroscopes are still used to detect the hand jitter, but instead of altering the
direction of
the light as it approaches the image sensor, the image is simply shifted in
software by a
certain number of pixels.
Figure 18. Above: jittery path of an object in video. Below: smoothed path with
EIS.
16
In Figure 18, the top image shows a series of jittery frames, and the path of an
object
within those frames. The bottom image shows the output of an EIS system in which
the
frames have been shifted and cropped, and the resulting smoothed path of an object
in the
video. This has an advantage over optical image stabilization in that there are no
moving
parts, and the power dissipation can be much lower. However, if the resolution of
the
camera or hand jitter becomes high enough to cause a noticeable blur within each
frame,
simply shifting the frames in software will not solve the problem; optical image
stabilization must be used. In practice, some video cameras use a hybrid system
combining optical image stabilization and electronic image stabilization, allowing
intelligent compromises to be made on the fly, based on resolution, zoom, and the
amount of hand jitter present.
Note that EIS serves to line up a set of frames, and does not prevent a single
frame from
being blurred. It is a system for video, and is not helpful in preventing blurry
photographs
from being taken. However, as digital still cameras increasing support video,
designers
may find it useful. For example, EIS can be used to stabilize the video presented
on the
LCD screen, as a user frames a shot. Upon taking a picture, the camera can briefly
activate the OIS system for acquiring a high quality image.
Conclusion
Gyroscope-based optical and electronic
image stabilization systems are mature and
proven technologies that address the
quality of images. OIS has been
penetrating the DSC market rapidly, and
as camera resolutions continue to increase,
optical image stabilization is expected to
become as standard a function as
autofocus on every DSC. As engineers
struggle to pack advanced technologies
into the scarce and premium real estates of
handsets, small size and low cost are at the
top of their lists. With the fast pace of
increased CMOS sensors pixel densities
and feature offerings, such as auto focus and optical zoom, OIS is expected to
enter into
the camera phone market very shortly. Many system designers are faced with the
challenges of introducing image stabilization into already crowded broad spaces.
Until
recently, there have been no viable options that meet the demanding size and cost
requirements. Thanks to InvenSense’s innovative MEMS dual-axis solution, the
barrier to
the implementation of OIS into a new generation of high performance camera phones
has
been lifted, and solutions as early as spring o 2007 are expected.
Figure 19. InvenSense gyroscope wafer
17
About the Authors
David Sachs is a Systems Application Engineer at InvenSense. He has designed many
gyroscope-based systems at InvenSense and MIT’s Media Lab. His work ranges from
low-level algorithmic development and multi-sensor fusion of gyroscopes,
accelerometers, and magnetic field sensors, to complete motion-sensing systems
including image stabilization systems, user interfaces, and gaming devices. He has
a MS
from MIT, and a BA from Oberlin College.
Steve Nasiri is the Founder, President and CEO of InvenSense Inc. and a 26 years
veteran
in the MEMS industry with expertise in fabrication and packaging of several MEMS
products. His extensive knowledge of MEMS fabrication and high-volume packaging
was instrumental in conceiving a novel mirror design for Transparent Networks,
where
Mr. Nasiri served as the Vice President of Operations and MEMS Development and
where he derived the idea for the dual-axis gyroscope from optical device designs.
He has
co-founded several successful Silicon Valley MEMS startup companies including
Sensym, NovaSensor, Integrated Sensor Solutions (ISS), ISS Nagano GmbH,
Intelligent
MicroSensor Technology, and most recently Transparent Networks. He has an MBA from
Santa Clara University, and an MSME from San Jose State University, and BSME from
University of California Berkeley. He has 28 issued patents.
Daniel Goehl is the Director of Sales for InvenSense Inc. He has held senior level
sales
and marketing management positions at several successful startup companies selling
into
the wireless market. He has a degree in Economics from the University of Illinois.
References
Nasiri, Steven. "A Critical Review of MEMS Gyroscopes Technology and
Commercialization Status".
Nasiri, Steven and Seeger, Joseph, May 17, 2005. U.S. Patent #6,892,575, "X-Y axis
dual-mass tuning fork gyroscope with vertically integrated electronics and wafer-
scale
hermetic packaging."
Nasiri, Steven and Flannery Jr., Anthony, September 6, 2005. U.S. Patent
#6,939,473,
"Method of making an X-Y axis dual-mass tuning fork gyroscope with vertically
integrated electronics and wafer-scale hermetic packaging."
Furukawa, Hiroshi et al. March 9, 1976. U.S. Patent #3,942,862, “Image stabilizing
optical system having a variable prism”.
Hara, Yoshihiro et al. October 4, 2001. U.S. Patent Application #20010025477,
“Control
mechanism with actuator employing shape memory alloy and method for adjusting
servo
control of the control mechanism”.
18
Kakiuchi; Shinichi. January 12, 2006. U.S. Patent Application #20060008263,
“Camera
provided with camera-shake compensation functionality”.
Kawahara, Hideo et. al. November 8, 2005, U.S. Patent #6,963,361, “Image sensing
method and apparatus capable of performing vibration correction when sensing a
moving
image”.
Kitazawa, Tomofumi et. al, September 6, 2005. U.S. Patent #6,940,542, “Imaging
apparatus, and method and device for shake correction in imaging apparatus”.
Konica Minolta Photo Imaging, 2004. “Digital SLR Camera with World’s First Body-
Integral Anti-Shake Technology”.
London, Barbara and Lipton, John. Photography, 8th Edition. Prentice Hall, 2004.
Masuda, Kazunori. January 9, 2003. U.S. Patent Application #20030007075, “Camera
system, camera, and lens apparatus”.
Micron Technology, Inc., 2006. “Lens Selection and Suppliers”.
Nikon Corp, 2005. “Precise camera-shake compensation at every angle”.
Noguchi, Kazuhiro. October 7, 2003. U.S. Patent #6,631,042, “Lens barrel having
image
shake correcting function and optical device having same”.
Semiconductor Insights, September 20, 2005. “Micron MT9D011 CMOS Image Sensor
awarded Most Innovative Image Sensor”.
Toba, Akira. June 15, 1997. U.S. Patent #5,648,815, “Video camera with electronic
picture stabilizer”.
Washisu, Koichi. June 29, 2004. U.S. Patent #6,757,488, “Control apparatus for
image
blur correction”.
Stavely, Donald J., Image Stabilization Technology Presentation, September 6,
2005.