An Improved Corner Point Detection Particle Filter Target Tracking Method

The tracking of moving target in video image sequence is a core problem of computer vision research,in the national defense security,medical diagnosis,human-computer interaction and intelligent transportation system and so on has a very important application value.At the present stage,the difficulties and problems of moving target tracking are: object occlusion,complex background,light change and similar interference,etc,to design a high robustness and accurate tracking is the pressing needs of the current stable accurate algorithm.For moving targets by occlusion or similar to the background color cases unstable,inaccurate problem,an improved particle filter target tracking method based on improved Harris corner detection is proposed,tracking of moving object in video image sequence simulation experiment,verified the feasibility and accuracy of this algorithm.The main research work is as follows:(1)Improved algorithm for corner detection.For Harris corner detection points deficiencies: using single threshold method to screen corner,single threshold setting is too big or too small,will influence the accuracy of corner detection and use of smoothing gauss window function,the gaussian window function is also not easy to control,a greater than small will affect the corner detection;Double threshold method is proposed to deal with initial candidate corner points,reusing SUSAN's(Small univalue segment assimilating nucleus)idea to deal with candidate corner points,eliminate false angles and improve detection accuracy,the advantages and accuracy of the improved Angle point detection algorithm are verified by simulation experiment.(2)Improvement of particle filter target tracking algorithm.With the shortcomings of the existing algorithms for target tracking and the difficulties of moving target tracking faced,by manually selecting the moving target in the first frame of video image sequence,making angular point detection and processing,the HSV color space model is established.Manually select the center and width,height of the rectangular area as the state of the particles,the observed value is HSV color histogram,and the observation equation is the likelihood function of HSV color histogram,particle filter is used to tracking,and the particle's constant self-renewal in the tracking process ensures the reasonable distribution of particle weight,thus achieving stable tracking of the moving target.(3)Simulation experiment.On MATLAB platform,dozens of video image sequences are used,two of which are taken as examples,by simulation experiment compared with the classical particle filter target tracking algorithm,verifies the feasibility,stability and accuracy of the moving target tracking algorithm when the target is occluded or the target is similar to the background color.