Research On Algorithm Of Moving Object Detection And Tracking Methods In Complex Condition

With the ever increasing requirements of modern life, people pay more attention to the development of information technology, especially the development of intelligent video surveillance system. In the intelligent video surveillance system,moving target detection and tracking technology is the key and the hot issue in computer vision field. They have played an important role in the range of national defense military, medical image and social security. Now the technology has gained many results, but many deficiencies still need to be improved.The work of this paper is as follows:This research mainly studied the optical flow, frame difference method and background difference method. Due to the bad effects of the traditional detection algorithm under the dynamic background, the research focused on target detection algorithm based on mutual information under dynamic background and put forward improvements, firstly, the image of once Mallat wavelet decomposition which is the approximate component, Pso and Powell hybrid optimization strategy was used to look for the maximum mutual information value of the image, then completed space conversion model and PV interpolation method,in this way the registration accuracy could be improved. Finally, used improved three difference method which avoided detection information loss to complete the target detection.The experiment proved that the improved algorithm can finish moving target detection under dynamic background.And this research also respectively studied Mean Shift algorithm and Camshift algorithm, because Mean Shift algorithm was easily affected by the background pixels and failed to track targets in complex conditions, corresponding improvement were put forward.Firstly, established a target model through the method of background weighted, secondly, according to the judgments of occlusion,combined Mean shift algorithm with kalman filter to track the target in the state of occlusion. Mean Shift algorithm failed to track targets when they were shielded, so we improved the Mean Shift algorithm based on kalman filter.That was to say,used kalman filter to predict the state of target and the possible location point would be obtained, then started to predict the position of target with the Mean Shift algorithm,and at the same time judged whether occlusion occurred, if occlusion happened,used kalman filter to predict the position of target and tracking results was the predicted results, so in this way we could track moving targets accurately when occlusionoccurred, and so was in the situation of fast moving targets. In order to further improve the accuracy of target tracking, we regarded an adaptive kalman factor as a judgment, then adjusted the parameters of the kalman filter flexibly, so the target could be tracked accurately. In addition, in the condition of background interference,improved Mean Shift algorithm based on background-weighted histogram.Established target model with the use of background-weighted histogram and established candidate target model with the use of nuclear-weighted histogram, then combined it with kalman filter, eventually this improved method not only reduced the interference of background pixels, and also enhanced the real-time performance of the algorithm.