Resarch On Moving Objects Detection And Target Tracking Based On The Video

In recent years, with the rapidly development of the computer science, image processing, artificial intelligence and pattern analysis, there rise the wave of the digital video surveillance. The technology of video based on object detection and tracking does not require human intervention, which using computer vision and video analysis method to analyze the image sequences recording from the camera automatically, and large numbers of researchers devoted themselves in the area and have already achieved many progresses. The dissertation is mainly to resolve the problem of how to improve the self-adaptive and real-time capability in moving target detection and tracking algorithm. In this paper, a new algorithm of target detection based on compressed sensing and a fast Kalman filter-based tracking algorithm is achieved based on the analysis and further research of current conclusion. Besides, we did further research in Mean-shift algorithm which is based on color pattern matching.The main content of this dissertation can be summarized as followings:1,Firstly, the ordinary techniques of moving target detection are briefly introduced. Then an improved method of moving objects detection based on background subtraction and frame difference is achieved by using a self-adapt threshold and the means of limited frames method.It can make up the limitation when the video sequence contains a moving target at the first frame, and can avoid the inadaptability when use an experiential threshold effectively.2,The moving target detection system offers enormous pressure due to the large amounts data of video image processing. Based on the recent new theory called compressed sensing(CS), a novel background subtraction way for detecting moving objects has been introduced in this article. The CS theory, which is different from the traditional Nyquist sampling theory, point that it's possible to reconstruct the signal from small number of non-traditional samples in form of randomized projections, as long as the signal is sparse or compressible. The experimental results show that the algorithm can be used to detect moving target by using a small amount of sample.3,We do further research on the Kalman filter algorithm which is used in tracking moving objects, and an improved algorithm of fast and efficient Kalman filter algorithm is achieved in the paper. Kalman filter is used to do the optimal estimation and correction at the next time based on the system's current state. Observing the iterative process, we see that because of too many parameters for each calculation, the computation is very cumbersome, and it reduces the real-time of tracking moving objects. However, by the formulas and many time's experiments, we get that the gain equation is going to become a constant value rapidly when the number of iterations involved is 20% of the total processed sequences. In this paper, the algorithm take the top 20% of the original frames are updated by the gain equation, then the rest of the video sequences are supposed as a reasonable constant value. The experiments show that the algorithm used in this article can achieve the tracking of moving objects fastly and effectively.4,The Mean-shift algorithm based on color pattern matching is reseached deeply. Then Mean-shift combined with fast Kalman filter predicting algorithm is achieved, it can reduce the search time in tracking moving objects to some extent. And it also can solve the problem when the objectives speeded-up rapidly or there have a large block curtain while the Mean-shift algorithm can't track effectively.