Moving Object Tracking In Video Sequences Based On Particle Filter

Moving object tracking is one of the attractive research subjects in computer vision,which combined many advanced technology in the field of signal processing, pattern recognition and automatic control. It has been widely used in the areas of intelligent video surveillance, safety monitoring, medical diagnosis, public safety, meteorological analysis,human-computer interaction and video compression.Particle filter is a method of Monte Carlo simulation based on recursive Bayesian filtering. It is widely used in moving object tracking as its better performance in processing the estimation of non-linear and non-Gaussian system. It can closely keep approaching the posterior probability density function via a group of particles with weights so that the object state estimation can be obtained.The research in this thesis is mainly about moving object tracking in video sequences based on particle filter. The contents can be divided into three parts as follows:Firstly, for the problem that object tracking by using single object feature often be considered as the key reason leading to the most object tracking algorithm in a poor performance, an algorithm based on multi-features fusion is proposed which combines the color histogram and edge histogram under particle filter framework was studied in order to tracking the object stability and accurately under the conditions of complicated background?abrupt changing of illumination and moving background.Secondly, for the problem that moving object tracking in occlusion cases is difficult, a method based on block matching using object color was studied. An object tracking algorithm for detecting and tracking effectively during occlusion in the following frame was realized.Thirdly, a new multiple objects tracking method based on background modeling and particle filtering is proposed to reduce the amount of computing by traditional particle filtering algorithm in multiple objects tracking. This method can keep tracking the objects effectively even in a faster computing speed.The experiments had been carried on for the research above separately and the results manifest that methods from the thesis are effective.