Moving Target Tracking Algorithm Under Occlusion

In the intelligent video surveillance system, target tracking is one of main researchaspects. Occlusion problem in the scene is always a difficult problem of tracking system.The article is the study of moving object tracking in the transportation. Mean Shiftalgorithm and particle filter theory are detailed introduction. Based on the occlusionanalysis and algorithms analysis, the algorithm is improved. The next is the main works ofthis thesis:Firstly, improved mean shift algorithm: in the relatively simple traffic system ofbackground, combining the scale space information and color information on the meanshift algorithm of histogram is improved, and deal with the problem that movement targetis far from and close to. In this process, the target is amplification and narrow, mainly tosolve the movement process of tracking window problem. The article respectively in theabsence of light and light conditions to tracking, compared to the method with only thecolor information of mean shift algorithm, experimental results show that the algorithmcan get better tracking effect.Secondly, improved particle filter algorithm: we use the simulated annealing methodto improve the proposal distribution of particle filter. In anneal layer, scale spaceinformation and color information integration are weighted to get weight function, thenthe proposal distribution is improved and the degraded phenomena of particle filteralgorithm is solved.Finally, combining particle filter and mean shift algorithm: RPF is further iterativeoptimized by improved mean shift algorithm, and using similarity factor to judge thedegree of occlusion. We consider the trajectory prediction to get the target of there-emergence of the location and number of fragment. The algorithm is real-time, andovercomes the degeneracy phenomenon of particle filter, shorten the time of calculation.Improved algorithm can effectively deal with the tracking process of the moving target infull occlusion.