A Study On Problems Of Splitting And Merging In Multiple Human Tracking

Multi-target tracking under complex scenes is a research hotpot in the field of computer vision. To overcome the problems of tracking failures resulting from splitting and merging in multiple human tracking, in this paper, we analyze the classical algorithms and the latest research progress of multi-target tracking, and present a multiple human tracking method based on JPDA and weight filter. We design and implement a prototype system of multiple human tracking.Our method is composed of three steps. First, it identifies tracking windows and measurements through the study of multiple frames, and then applies the JPDA algorithm for calculating the joint association probabilities between each measurement and the corresponding possible source targets. Finally, it combines the joint association probabilities with the weight filter to accomplish the tasks of predicting and updating the status of each target. The results on the general test sets PETS2003 show that our method effectively solves accurately tracking problem when multiple humans split and merge frequently, and achieves a good balance in robustness, accuracy and speed of the method.