01-11-2012, 05:28 PM
Background stabilization and debris detection in launch pad video monitoring
Background stabilization.pdf (Size: 736.4 KB / Downloads: 110)
Abstract
Automatic detection of moving objects in video sequences is a widely researched topic
with application in surveillance operations. Methods based on background cancellation by
frame di®erencing are extremely common. However this process becomes much more com-
plicated when the background is not completely stable due to camera motion.
This thesis considers a space application where surveillance cameras around a shuttle
launch site are used to detect any debris from the shuttle. The ground shake due to the
impact of the launch causes the background to be shaky. We stabilize the background by
translation of each frame, the optimum translation being determined by minimizing the
energy di®erence between consecutive frames. This process is optimized by using a sub-
image instead of the whole frame, the sub-image being chosen by taking an edge detection
plot of the background and choosing the area with greatest density of edges as the sub-image
of interest.
The stabilized sequence is then processed by taking the di®erence between consecutive
frames and marking areas with high intensity as the areas where motion is taking place. The
residual noise from the background stabilization part is ¯ltered out by masking the areas
where the background has edges, as these areas have the highest probability of false alarms
due to background motion.
INTRODUCTION
The detection of movement in frame sequences is done by a variety of methods including
frame di®erencing, determination of optical °ow and by using the features of various objects
in the sequence to track them. However, it is not possible to use many of these methods in a
case where the background is semi-stable. This thesis deals with an application of this nature,
where the movement in the background is caused by ground shake due to a satellite launch.
The methods used in this thesis are simple, commonly used image processing techniques that
have been combined together to give a reasonably good solution of the problem at hand.
The problem has been divided into a stabilization phase and a detection phase thereby
providing an elegant way of reducing the complexity of the programming involved. It is likely
that this approach adds some computational ine±ciency to the algorithm, but the author
believes that the resulting simplicity of the approach makes the trade-o® worthwhile.
Organization of the thesis
Chapter 2 contains a review of the common methods in existing literature for motion
detection in video sequences. It contains a short explanation of methods based on frame
di®erencing. This is also a short discussion of motion detection based on optical °ow, more
speci¯cally the Lucas-Kanade image alignment algorithm.
Chapter 3 contains a description of the methods used by the author for background stabi-
lization. It contains an explanation of the process of feature extraction including a description
of the Sobel method of edge detection which is used to extract information about the back-
ground. The chapter then describes the process of aligning successive frames to the initial
background by shifting each frame such that the error energy between frames is minimized.
In Chapter 4 the author describes the method used to determine regions of movement
by subtracting consecutive frames and also describes ways of masking out some of the false
alarms caused by imperfections in the stabilization process.
Motivation for the thesis
The premier motivation for this project comes from the increased emphasis on safety in
NASA's space program in the aftermath of the tragic crash of the space shuttle Columbia.
The space shuttle Columbia, NASA Orbiter Vehicle designation: OV-102, was the ¯rst
shuttle of NASA's orbital °eet that went on 28 missions to space between 1981 and 2003. It
was launched for the ¯nal time on the 16th of January, 2003 on a micro gravity and Earth
science research mission, STS-107, that carried out a wide range of scienti¯c experiments
including biomedical research and earth observation. On the 1st of February, 2003 Columbia
disintegrated during it's re-entry into the earth's atmosphere, just sixteen minutes before
scheduled touchdown. All seven crew members were killed in the crash.
The video taken during take-o® showed a piece of foam from an external fuel tank strike
the shuttle's left wing. The damage caused to tiles in the thermal protection system by this
debris strike is generally thought to have caused the accident.
These events have given a greater sense of importance to the process of detecting and
analyzing debris in NASA's shuttle program. They have shown that what was previously
considered a \turnaround issue" is vital to ensuring the safety of the space shuttle and those
aboard it.
REVIEW OF EXISTING LITERATURE
We look at some of the basic ideas used in the various motion detection algorithms that are
in existence. Though there are innumerable papers in this area of research, the discussion
in this thesis is limited to the most commonly used methods. This chapter aims merely to
provide some background about the basic concepts involved in motion detection and is by
no means a comprehensive review of the current state of the art.
Methods based on frame di®erencing
A really common way of detecting movement in frame sequences is to subtract each frame
from its previous frame. Ideally this cancels out all the pixels in which there is no change
leaving non zero values only in pixels where motion has occurred. This method is central to
the work done in this thesis and merits further scrutiny by means of an example.
Methods based on features of objects in the video sequence
It is possible to use the properties of the various objects in the video to track them through
the various frames. These properties include color, shape, texture and trajectory of motion.
It is possible also to segment di®erent objects by representing them as a mixture of two
dimensional gaussians. These methods are not suitable for the problem being addressed by
this thesis as the debris from the shuttle is of unknown shape and size, and could travel in
any possible trajectory.