Abstract
Computer vision-based intelligent video analysis combines many technologies
including image processing, pattern recognition, artificial intelligence, automatic
control, mathematics and computer science, etc, and has vital scientific significance
and broad applied prospects. Along with the fast development of information
technology as well as the urgent need of security situation, the requirement for higher-
powered security equipment is ever-increasing. It is becoming the developmental
direction in the future that making video surveillance system be roboticized and
intellectualize through computer vision technology. Both motion target detection and
tracking are important tasks in video surveillance for they are the cornerstones of the
following each kind of high-level processing, like pattern recognition, behavior
analysis, etc. Since the realistic monitoring environment is often complicated and
constantly changing, it is the main research content that searching the method which
can manage various kinds of change in complex environment, and detect and track
target rapidly, accurately and stably.
Based on the background of realizing intellectualized video surveillance system,
this paper mainly research on two key technologies in the intelligent video surveillance
system, namely motion target detection and tracking. The purpose is to design the more
practical algorithm and form closed-loop controlling method in whole to realize the
intelligent video surveillance.
The paper has given a comprehensive summary of the motion target detection and
tracking algorithms and their development. Several commonly used motion target
detection algorithms are firstly reviewed, including optical flow, time-difference,
background-difference, then the method of foreground detection based on the Gauss
background model is elaborated. A classification of the motion target tracking
algorithms is also made, and tracking algorithm based on the Mean Shift is thoroughly
studied. The principles, advantages/disadvantages, and the sphere of applications of
these algorithms have been analyzed and compared. With this understanding, the paper
has been fulfilled following three works:
An automatic tracking system has been designed by the means of combining
differencing-frame and CAMShift. Firstly, foreground\background is detected through
sequential three-frame-double-difference. The CAMShift Algorithm is then used to
calculate the exact location of the target and adjust the size of search window.
IV
The above-mentioned improved motion target detection and tracking method has
been implemented on DSP. A closed-loop system based on DM642-PCI board and
servo cradle head has been also designed. To begin with, the system detect moving
target automatically, and then the CAMShift Algorithm is used to calculate the exact
location of the target. The information of the warp between the target’s center and the
focus of the camera’s eyeshot is transformed into the protocol commands and sent via
the serial port lastly, to rotate the cradle head to change camera’s eyeshot for
maintaining the target visible in the scene. The software algorithms have implemented
on the hardware with the core of TMS320DM642 by CCS2.2.
A kind of tactic of PTZ tracking controlling based on high speed spherical
camera has been designed. Spherical camera is active along three directions of pan, tile
and zoom, and it is able to rotate fleetly with 360. Under the situation of the sphere
camera’s mechanical parameter unknown, P/T directionless posture angle of the
spherical camera is adjusted through controlling it to rotate intermittently to maintain
the target visible in the scene. After the center of eyeshot aiming at target accurately,
spherical camera starts to zoom for observing or shooting the target with its local
details. Furthermore, a method of compute the zooming ratio based on SIFT
characteristic matching has been designed for solving the issue of the tracking
windows needing adaptively adjusted in the process of zooming, as well as a tracking
algorithm in RGB color model based on the Mean Shift. Based on PELCO protocol,
the whole flow of PTZ tracking has been implemented by VS2005 and OpenCV.
In this paper, the main research focus on the algorithms and the design of the
systems controlling flow, but implementing on the hardware is an experimental work
for just building the system platform. In order to exert the powerful performance of the
DSP, there is a further work to optimize the embedded software, which needs time and
the engineering experiences considerably. Regarding PTZ track strategy of spherical
camera, the experimental results show that it is able to tracking effectively, real-timely,
and can control the spherical camera to zoom image for shooting the target with its
partial details. It has certain reference value in the application of crime evidence
collection. Since few references about the PTZ tracking realization are available at
present, the strategy in the paper is only the preview research, still has certain gap to
the true commercialized use.
Key words: Intelligent video surveillance; Motion target detection and tracking;
TMS320DM642; Mean Shift algorithm; High speed spherical camera; PTZ controlling