The Algorithm Of Track And Record Moving Object In Video Processing

Moving object tracking and recording are the main topics of computer vision and image coding; they concern many subjects such as image process, pattern recognition, AI and so on, so this subject has important theoretical significance and wide practical value. Because of the complexity of video processing, there isn't a universal method. So we should do more research of this subject.Track moving object is the first step of video segmentation, it is to divide an image or sequence into some areas according to the specifications, in order to extract some significant entities, called video object in digital video processing. Active Contour Model, or Snake model, which was first introduced by Kass in 1987, could track moving object effective and credible. The strongpoint of the algorithm is it that the ability of using the information of image data and the position of object to divide, match and track object. Based on the in-depth research of active contour models which have been extensively studied and applied as image segmentation methods during the past decades.The interior limitation of the traditional Snake model has caused the severe disadvantage1. It is sensitive to the initial position of curve, and it must depend on people to initialize it;2. It is also sensitive to the parameter. It is difficult to choice a right parameter;3. If there are many objects in a image, they will disturb each other, and the algorithm couldn't get anyone.Image matching is an important technology of image process, it has been widely used in tracking, identification and medical image analysis. The image matching is to match the model with an image, and find the conditional displacement. The Hausdorff distance has strong ability of anti-jamming, and can enhance the robust of matching. So it has been used in many domains. Hausdorff distance is suit for the situation which has a little noise and disturb from the edge. The part Hausdorff which is introduced by Huttenlocher, has solved the problem of noise, and the algorithm is very simple. So it could be well used in practical matching.In the contrail recording algorithm, if we use video to record, it must use too much system resources, so it isn't adapt to long time recording. It could release the limits of hardware that if we only record one point's contrail, but the observer couldn't get enough vision information, such as: the shape or the background.After carefully reading a lot of papers on the moving object track and record in video processing, we give a comprehensive sum-up of existing methods, and have validated the results and practicability of each algorithm.Based on the former research, the main innovation of this paper represents in the following three aspects.1. In this paper, we introduce an improved tracking algorithm which based on the Snake. The algorithm first use Sobel arithmetic to get the edge in the key frame, and this edge information can be used to the constraint condition. We get the initial points of Snake by the difference of three frames, then use median filter to get a better image and use the Greedy Snake to get the real contour of object.Compare to traditional Snake, the improved tracking algorithm could automatically, quickly and exactly get the initial points which are near the edge of object. It solves the problem of initial points'choice, and avoid the Snake curve converges to a wrong position. Meanwhile, it reduces the times of iteration. So this tracking algorithm which based on Snake could do real-time tracking and get an exact contour.2. In order to do a long time recording of aim's contrail, in this paper, we introduce an improved recording algorithm which based on the Hausdorff distance matching. This algorithm records the position of aim's center point in each frame, composes them to the motion contrail and save them in text.Use the method to record contrail of moving object, the observer could see the contrail clearly and also could get the more vision information such as: the shape of object, the background etc. This algorithm enhances the robust of matching, but it only needs approximate 0.001 percent of usual video recording memory. The improved recording algorithm which based on the Hausdorff distance matching could use in the long time recording which mainly concern the trail of motion.We used Visual C++ 6.0 to experiment on the algorithms which we had referred in this paper, and had validated the effect of algorithms which we introduced in this paper.