Research On Visual Tracking Algorithm Based On Particle Filter

With the increasing improvement of computer science and sensor performance,the technique of visual tracking plays a more and more important role in practical applications,such as intelligent transportation system,security surveillance system,etc.,becoming an important and difficult problem in the research fields of computer vision and artificial intelligence.The task of visual tracking is to sequentially estimate the positions and the scales of the targets in the image sequence based on the priori information about the apparent features and the motion model of the target.In order to achieve a long-time,stable,accurate and efficient tracking in a complex environment,it is required to make efforts in the self-adaption of the model and the real-time performance of the algorithm.For these purposes,the dissertation utilizes particle filter(PF)to carry out the research on visual tracking methods.The main contributions of the dissertation are as follows:1.Particle filter is confronted with the problem of particle shortage.For this reason,a novel particle filter tracking algorithm based on particle swarm optimization is proposed.In the proposed algorithm,the particles are optimized using the particle swarm optimization before resampling,by keeping the particles with larger weight unchanged and the particles with smaller weight moving to the larger weight particles.It can keep most of the particles staying at the high likelihood area,reducing the elimination rate in resampling,and alleviating the problem of particle shortage.Experimental results show that the proposed algorithm can improve the utilization ratio of the particle and realize a faster and more accurate tracking performance by using fewer particles in all of the complex situations.2.Particle filter is also faced with the problem that the spatial coverage of particles and the ambiguity of measurement are not match.To handle the problem,a novel particle filter visual tracking algorithm based on particle swarm is proposed.In the proposed algorithm,the point particle is extended to particle swarm which occupies a non-zero area,and it can improve the spatial coverage of particles.Experimental results show that the proposed algorithm can improve the sampling efficiency and provide a more accurate tracking performance without increasing the runtime.3.Particle filter has an inherent drawback in the aspect of run time.For this purpose,the CSK algorithm with the better performance in runtime is studied,and a novel algorithm with self-adaption in scale is proposed.In the proposed algorithm,a scale filter based on MOSSE is designed,which is utilized after the location filtering of CSK algorithm.The scale filter can output different response values by inputting features in different scales,and the scale of the maximum response values is the real scale of the target.Experimental results show that the algorithm not only can locate the target at real time speed but also calculate the scale of the target.