Particle Filter And Research On The Methods Of Adaptive Window

Moving object tracking in video is the focus issue in the field of computer vision,which is a variety of theories, involving image processing, artificial intelligence andpattern recognition and other interdisciplinary technology. It is widely applied in the fieldof video analysis and processing of intelligent monitoring systems, human computerinteraction, medicine and biology, navigation, guidance.Particle filter is widely used in moving target tracking system because it caneffectively solve the non-linear, non-Gaussian filtering problem and its algorithm hasexcellent robustness. In the traditional particle filter tracking method, the size of trackingwindow is determined by the initial target size, during the whole process of target tracking,the tracking window size does not change, which will lead to inaccurate tracking, eventracking failure especially the target size becomes much big. In recent years, a lot ofresearches for solving this problem are based on Mean-shift tracking algorithm adaptivewindow, which put forward some changes to the kernel function bandwidth algorithm. Inview of the particle filter tracking window, by contrast, is litter research of adaptive change.This paper around the particle filter tracking algorithm and its window adaptive adjustmentwindow problem has carried out the following related research work.Systematically study of the particle filter and tacking algorithm based on particle filtertracking window adjustment. This paper summarizes the existing tracking windowadjustment method and their characteristics, focusing on the particle filter trackingadaptive algorithm window, putting forward a kind of adaptive tracking windowadjustment. The algorithm in my paper is analyzed in detail the relationship between theaverage distance of particles to the center of the target and the target size, then established the mathematical model of the size of the tracking window adaptively adjusted. On threekinds of target model for the established mathematical model has carried on by thesimulation, the experimental results show that, my algorithm can well with the changes inthe target size and the adaptive adjustment of the tracking window, and does not increasethe amount of additional calculation and has good real-time performance.