Research On Video Object Tracking Algorithm Based On Improved Particle Filter

In The research of artificial intelligence represented by Google AlphaGo AI system has attracted much attention in recent years.Computer vision is a very important field in the research of artificial intelligence,and target detection and target tracking is one of the main research branches in the field of computer vision.This thesis using the method of global motion compensation,edge detection and morphological filtering to improve the consecutive frames subtraction as the method of moving objects detection.And target matching based on two level target matching mechanism based on image information entropy and template matching.Constituting a specific moving target detection algorithm.Traditional particle filter target tracking algorithm in the face of the problem of target occlusion requires substantially increase the number of particles and the amount of computation,but if there is a long time occlusion,tracking effect will still be significantly decreased.This thesis proposes an improved particle filter tracking algorithm based on fusion motion state information to solve this problem.Using the trajectory prediction based on least square method to improve the particle filter algorithm framework,which has more robust to against occlusion or background mutation.Secondly,fusion motion state characteristics in the main process of particle filter algorithm,which based on the color distribution histogram and the moving state information of the template set as the target model.based on this model and trajectory prediction process,the key steps of state transition model,system observation model and template updating strategy are improved.It can effectively overcome the occlusion,scale,deformation and illumination changes of the target,and make the target model to approach the true state of the target.The specific moving target detection algorithm and the particle filter tracking algorithm based on the fusion motion state information constitute the two main modules of the video tracking system.