Research On Moving Target Detection And Tracking Based On Background Subtraction And Particle Filter

Moving target detection and tracking is the research hotspot in the field of computer vision.Traditional detection and tracking algorithms are only for a specific environment, and when theenvironment becomes complicated, the results of detection and tracking will decline. Aiming at theproblem above, some improved methods are proposed in this paper, and the major work is as follows:Firstly, a target detection algorithm based on Gaussian mixture distributions and backgroundsubtraction with adaptive thresholds is proposed. In this algorithm, the number of distributions isadaptively chosen by the complexity of the environment in the background modeling. After obtainingthe difference image by background subtraction, it is divided into two parts according to the pixels’values. The two parts are separately segmented by the adaptive threshold, then the completeforeground image is obtained after shadow elimination and morphological processing. Theexperiments show that the improved algorithm has a strong robustness when the target’s grayscale isclose to the grayscale of the background or the target changes a lot itself.Secondly, aiming at the particles becoming single which is caused by the traditional resamplingalgorithm in the particle filter, an improved resampling algorithm based on the judgment of weights isproposed. A certain number of particles are drawn randomly from a set of particles, they are sorted bytheir values of weights, and then some particles ahead are selected as the new set of particles. Theexperiments shows that the improved algorithm can maintain the diversity of the particles to someextent.Thirdly, multi-feature fusion and adaptive number of particles are used to track the target. Inorder to guarantee the stability of the matching, the color, edge and texture features are fused todescribe objects; to improve the efficiency of individual particles, the matching degree of objects andtemplate is defined, which makes the number of particles change adaptively, and the number increaseswhen the matching degree becomes large and decreases on the contrary. The experiments show that inthe non-Gaussian nonlinear environment, the improved algorithm effectively solves the target rotation,target occlusion and background confusion and many other issues.Finally, a moving target tracking algorithm based on background subtraction and particle filter isproposed. In this algorithm, the detection results have been integrated into the tracking. The weightsof the particles which are on the pixels detected as the target become large and the others stayunchanged. The efficiency of the particles can be improved and the number of particles is reduced by this measure. Experiments show that the improved algorithm has great real-time performance with thetracking accuracy guaranteed.