Research On Multi-target Tracking Based On Adaptive Fragment In Complex Scenes

With the development of computer hardware and multimedia technology,as well as the governments and people's great attention to security,the applications of intelligent video surveillance become more and more extensive.As the most fundamental and core technology of intelligent video surveillance,research on multi-target tracking has strong theoretical significance and broad application prospect.So it has received extensive attention and research from academia and industry all over the world.At present,the research of multi-target tracking has made great progress,but there are still some problems,such as complex tracking scene,the gesture change of non-rigid targets,occlusion,real-time tracking and so on.This dissertation focuses on multi-target tracking in complex scene that there are occlusion,appearance change and color similar targets.The main research contents and results of this dissertation are as follows:Firstly,the fundamental theory of multi-target tracking is introduced in this dissertation.The fundamental principle of Kalman filter and particle filter under Bayesian theory framework is briefly introduced,and their advantages and disadvantages are analyzed.Furthmore,this dissertation introduces the fundamental principle of mean shift and fuzzy C-means,and studies their basic steps.Secondly,for the problems of occlusion and color similar targets in multi-target tracking,we proposed a multi-target tracking method with particle filter based on adaptive fragment.We divide each target into a few fragments adaptively according to its gray projection,and this method can improve the accuracy of multi-target tracking in the presence of occlusion.During particle filter tracking,we obtain particle sets of each target by mean shift algorithm and FCM clustering algorithm,and the optimal state estimation of each target is calculated through particles state in subgroup.The similarity of each fragment can be obtained by adopting weighted Bhattacharyya Coefficient that considers the influence of fragment reliability to particie weignt.Finally,in order to solve the problems of color similar target occlusion and appearance change also existing multi-target tracking,a multi-target tracking method with particle filter based on adaptive fragment and multi-feature fusion is proposed.This method adds a multi-feature fusion strategy based on the previous method,fusion color histogram and HOG feature to describe the target.During particle filter tracking,the weight of each fragment in target model is dynamically adjusted according to its spatial reliability and all particles' distribution.And to reduce the impact of target change on the tracking results,the weighted update method is used to update the target feature model in time.Experimental results show that the proposed method can track the targets accurately and robustly on many kinds of complex circumstances,such as appearance change,color similar targets,occlusion and even color similar target occlusion.