Design And Implementation Of Moving Target Tracking System Based On OEM-GMM And Particle Filter

In recent years, with the rapid development of information and social, intelligent monitoring system plays an increasing important role in military and civilian fields. Moving object tracking in video sequences is one of the core content in intelligent monitoring technology. So, moving target tracking has got widespread concern by the research scholars.Due to the complexity of the actual environment, the target tracking still faces many difficulties, such as the complex background, the goal of mutual occlusion, staggering, posture change and the changing lighting. Although many effective tracking methods have been proposed, there are still a lot of key issues need to be resolved. In this paper, Overall Expectation Overall Expectation Maximization and Gaussian Mixture model (OEM-GMM) is proposed to obtain the moving targets and fuzzy metric multi-feature adaptive integration strategy in the particle filter is put forward to complete the tracking.In moving target detection, in order to reduce the waste of resources caused by fixed numbers of Gaussian distribution in the Gaussian Mixture Model and improve the computation efficiency, a method that the numbers of the Gaussian distribution will be changed according to the environment is designed in this paper. Based on the minimum description length criterion, the local optimum drawback of the usual EM algorithm can be avoided. Compared to the GMM, average frame processing speed is increased by23.85%.In moving target tracking, in order to solve the fact that tracking failure of particle often happens in the case of low discrimination abilities of the observed features, this paper focuses on the appearance model and presents a fuzzy metric multi-features adaptive fusion model. The color, texture and edge gradient of the targets are fused according by fuzzy metric adaptive strategy, using the likelihood probability function as the target tracking basis. When the particle number is300, the accuracy of the improved algorithm is about90%. Arm to enhance the speed of the execution of the particle filter algorithm, this paper adopts OpenMP multi-threaded programming language to parallel the process. The average frame processing speeds of the target tracking model and the whole system are separately increased by36.12%and17.54%.Through improving the Mixture Gaussian Model methods and particle filter theory, the moving target tracking system designed and completed in this paper can achieve the rapid detection and robustness of the tracking of moving targets under different tracking conditions.