Research On Multiple Object Tracking Under Complex Background

As we all known,the research about multiple object tracking has become a hotspot in the field of computer vision.Thus,we have conducted a thorough research on multiple object tracking under complex background based on the pedestrian tracking in this paper.The main work is as follows:(1)Developing an improved GMMCP tracker for multiple object tracking based on the Generalized Maximum Multi Clique Problem(GMMCP)tracker.In order to improve the tracking performance of target occlusion and reduce the computational complexity,we employ the Aggregated Dummy Nodes(ADN)based on the dummy nodes.What's more,we modify the measures of object similarity in order to improve the confidence of weight set and the tracking performance of tracker.(2)Developing an improved Discrete-Continuous Energy Minimization tracker for a joint representation of data association and trajectory estimation based on the Continuous Energy Minimization tracker.We use cubic B-splines to represent the trajectory in order to ensure the continuity.Besides,we integrate global constraints of trajectory by modified label cost term in energy function in order to represent the real world.(3)Performing experiments and evaluating the performance of our proposed methods based on publicly available video sequences including TUD,PETS09 and so on.According to the tracking results,we find that our methods can perform a robust multiple object tracking.Additionally,we perform the quantitative evaluation of the methods using CLEAR MOT metrics and compare the performance of our methods with other state-of-art methods,which shows that our methods have extremely reached or been close to the advanced world level.In summary,this article has conducted a thorough research on multiple object tracking under complex background and made some achievements.The proposed methods can perform a robust multiple object tracking which has been verified by experiments.