Research On Moving Target Tracking Algorithm Based On C-V Model

With the development of computer technology, the moving target tracking technologyhas become a hot research in the field of video surveillance, among which tracking based onactive contour model is one of the tracking technology. In recent years, because of thesimplicity, efficiency and the ability to extract arbitrary shape deformation profile, thegeometric active contour model has been widely used and greatly developed in the imagesegmentation and tracking of moving targets. In this paper, an improved C-V(Chan-Vese)model has been proposed and combined with the classical gauss mixture model to realizemotion detection and tracking. Then, it combined with the kalman filtering algorithm toachieve the continuous tracking of the movements. Finally, the embedded system was usedas a development platform to implement the application of active contour model in intelligentdevice.First, this paper introduces the traditional C-V model which has the followingproblems:(1) The level set function needs to be periodically initialized to sign distancefunctions, which leads to slower segmentation;(2)The traditional C-V model sets internal andexternal value of the initial contour as a constant, and for the uneven gray object this globalapproximation method is not able to accurately segment the image;(3)The method is not goodfor the image which contains noise. View of the above problems, we introduce an gausskernel function and an edge stopping function on the base of the traditional C-V model. Usingthe weighted average of the local window function instead of the global mean in C-V modeland joining the distance function compensation term to avoid the re-initialize of the level setfunction, and then we integrate the edge of image information into the geometric activecontour model, so it will overcome the problem of traditional C-V model which can’t use theimage gradient information.Second, as video tracking of the target position in each frame changes leading to unable to determinethe initial curve, we combine the gauss mixture model algorithm with the improved C-V modelalgorithm. First, using the gauss mixture model algorithm to detect the moving object, thenmark the center and moving target area of motion after the morphological filtering and so on;Then, the motion area is set as initialize curve and we use the improved C-V model to evolvethe initialize curve so that we can accurate segment each frame of the moving objects, thus wecan realize the moving target tracking.Then, combining the improved C-V model algorithm with kalman filtering, using theimproved C-V model algorithm to segment the current image and determine the target’scentre of mass, and then using kalman filter to predict the centre of mass, besides initializingthe curve and again for segmenting evolution, thus we can achieve the goal of continuoustracking.Finally, using the Embedded Wince system as a development platform, researching thecomponent of the hardware system and describing Windows CE program, BSP packagetransplant, the use of camera and the correspond between Mini2440development board andpc machines. In the host, we use vs2005and opencv library as a software development tracking the moving targets; in the development board, we have realized the goal that we canpause and play a video camera, besides we can also realize some simple image processingoperations such as stretch and shorten.