| Along with the continuous reform and innovation of computer vision technology, moving target detection and tracking which based on video surveillance system has become one of the hot topic computer vision research. In the process of video moving target detection and tracking, the surrounding environmental factors such as the light, weather,the behavior and the movement speed of moving target always affect the result of the detection and tracking. Therefore, in this paper based on the complex observation condition as the research background, starting from the theory and combine with the practical significance, in depth analysis and research of moving target detection and tracking.The study of this subject has important scientific significance and broad application prospects.The mainly work of the subject have the following aspects:(1) The part of image preprocessing. In order to guarantee the follow-up work smoothly, in this paper, the selected video frames have to be processed in advance. The image processing is divided into image gray scale processing and noise processing. In researching and analyzing the methods of image gray scale processing and noise processing at the same time, by using Matlab7.0 tool for simulation and comparison. Then using the weighted average method and adaptive median filtering method for video frame image gray processing and noise processing.(2) The part of video moving target detection. Analysis and research the three methods of the optical flow field, the inter-frame differentiation and the background differentiation with these detection algorithms implementation process. Then, analyze their respective advantages and practical scope and the comprehensive performance of contrast. On the premise of complicated observation condition and high accuracy and real-time performance for research purposes. This article adopted the background difference method which based on gaussian mixture model for moving target detection. Multiple gaussian distribution model is established for each of the pixels of the image. From judging the model matching degree to update the parameters. After repeated training the parameters, can get more accurate background model. By using Matlab7.0 tool for moving target detection experiment which under the condition of different illumination and different character actions.(3) The part of video moving target tracking. In this paper, the classification of the tracking algorithm is analyzed. At the same time, emphatically to introduce the mean-shift, the kalman filter and the particle filter algorithm. Then, their respective advantage and disadvantage are analyzed in the process of practice. By comparing to the comprehensive performance of these algorithm, choosing the particle filter as the basis of tracking algorithm.Meanwhile, in view of the particle degradation and shortage problem of the original particle filter algorithm which in the process of running, the paper first present a new solution called the particle filter algorithm based on optimization of diversity. In effectively solve the problem, at the same time, integrated into the cuckoo search optimization algorithm based on levy flight mechanism. It can not only expand the search scope, but also fully retain the effective particle number and the diversity of particles, so as to reduce the tracking error of the particles. By the operating conditions of Matlab7.0, compare the algorithm with original particle filter and the current advanced particle filter based on particle swam optimization algorithm for state estimation and moving target tracking. The simulation results show that the proposed algorithm has a good tracking accuracy and a good real-time performance and robustness. |
