16-05-2012, 03:19 PM
Analysis and Compensation of Rolling Shutter Effect
[attachment=22162]
INTRODUCTION
THE DEMAND for image sensors is growing rapidly due
to the increasing popularity of digital video cameras in
surveillance, personal communications, consumer electronics,
etc. Although the charge-coupled device (CCD) has been the
dominant technology for image sensors, the complementary
metal-oxide semiconductor (CMOS) technology is becoming
a popular alternative because CMOS sensors can directly
integrate with image processing or communication circuits,
providing low cost, low power solutions for various applications.
RELATED WORK
The rolling shutter effect can be prevented by using a mechanical
shutter. Alternatively, one can design a new circuit with a
local sample-and-hold for each photo sensor so that the whole
sensor array experiences a single exposure [7], [8]. However,
this method reduces the fill factor of the photo sensor and degrades
the efficiency of photo collection. As a result, the readout
rate is still limited by the transmission rate.
MODELING OF ROLLING SHUTTER EFFECT
We focus on the scenario in which the camera undergoes with
respect to the object a translational motion parallel (or nearly
parallel) to the image plane. The planar motion model is based
on two observations. First, the rolling shutter effect is noticeable
when the camera or the object moves rapidly in a single
direction. It is less noticeable for complex motions such as rotating
and zooming. Second, the difference in image distortion
is hardly noticeable when depth range of the objects in the scene
is moderate. Such a planar motion model has been commonly
adopted for many image processing problems, such as image
deblurring, motion-compensated video coding, and video stabilization.
DISTORTION CORRECTION ALGORITHM
Undoing the rolling shutter effect is a challenging task. There
are three primary problems to be addressed. First, from (9), we
know that if and are identified and the velocity is
known, the displacement between and can be derived and
corrected. However, a pair of and need be identified first.
Second, because the input sequence is geometrically distorted
and motion blurred, velocity estimation by feature detection and
matching is difficult. An alternative solution has to be sought.
Third, the temporal sampling of must be at the scanline
level, which is much higher than the frame rate. However, traditional
motion analysis methods can only produce motion information
at frame rate.
[attachment=22162]
INTRODUCTION
THE DEMAND for image sensors is growing rapidly due
to the increasing popularity of digital video cameras in
surveillance, personal communications, consumer electronics,
etc. Although the charge-coupled device (CCD) has been the
dominant technology for image sensors, the complementary
metal-oxide semiconductor (CMOS) technology is becoming
a popular alternative because CMOS sensors can directly
integrate with image processing or communication circuits,
providing low cost, low power solutions for various applications.
RELATED WORK
The rolling shutter effect can be prevented by using a mechanical
shutter. Alternatively, one can design a new circuit with a
local sample-and-hold for each photo sensor so that the whole
sensor array experiences a single exposure [7], [8]. However,
this method reduces the fill factor of the photo sensor and degrades
the efficiency of photo collection. As a result, the readout
rate is still limited by the transmission rate.
MODELING OF ROLLING SHUTTER EFFECT
We focus on the scenario in which the camera undergoes with
respect to the object a translational motion parallel (or nearly
parallel) to the image plane. The planar motion model is based
on two observations. First, the rolling shutter effect is noticeable
when the camera or the object moves rapidly in a single
direction. It is less noticeable for complex motions such as rotating
and zooming. Second, the difference in image distortion
is hardly noticeable when depth range of the objects in the scene
is moderate. Such a planar motion model has been commonly
adopted for many image processing problems, such as image
deblurring, motion-compensated video coding, and video stabilization.
DISTORTION CORRECTION ALGORITHM
Undoing the rolling shutter effect is a challenging task. There
are three primary problems to be addressed. First, from (9), we
know that if and are identified and the velocity is
known, the displacement between and can be derived and
corrected. However, a pair of and need be identified first.
Second, because the input sequence is geometrically distorted
and motion blurred, velocity estimation by feature detection and
matching is difficult. An alternative solution has to be sought.
Third, the temporal sampling of must be at the scanline
level, which is much higher than the frame rate. However, traditional
motion analysis methods can only produce motion information
at frame rate.