Research On Human Detection And Tracking Based On Infraced Imaging

Intelligent video surveillance has made great progress in recent decades. It has beenwidely used in medicine, astronomy, security checks, visual navigation, military guidance,and many other production areas of life. With lower prices, as well as the continuous progressof science and technology of infrared imaging devices, infrared imaging conditions targetdetection and tracking increasingly attracted the attention of researchers. Particularly thosebased on infrared imaging of the human target detection and tracking has become a hotspotamong the field of visual intelligence in recent years.This paper focus on the infrared image of the human body target detection and trackingtechnology.Firstly, infrared image segmentation algorithm has been researched and summarized, andthe corresponding experimental results are provided. Then it selects the k-means clusteringthreshold segmentation algorithm to discuss and to improve them: through the derivation ofthe threshold value of the k-means clustering algorithm measure function into the test clustercenter value and the theoretical center value of the partial is selected The shift amount as themeasure function, thereby improving the effect of image segmentation.In target tracking, we have adopted an algorithm based on particle filter infrared humantracking algorithms and fusion particle filter and meanshift advantages of infrared body targettracking algorithm. Using brightness statistics of human target on the origin to the humantarget area center ring-from the space, infrared human detection algorithm based on particlefilter constructs the human characterization model in intensity-distance space, and then therobustness of infrared human target tracking is finished by fusing the model with particle filter. In order to solve the problem of increasing amount of calculation and system haltingcaused by particle number required in particle filter and status dimension, we propose theinfrared body target tracking algorithm integrating particle filtering and meanshift advantages.This method firstly builds movement-brightness body characterization model by integratingthe human target gray information in the brightness-distance space with the motioninformation of human target fusion to. Then mean shift algorithm integrated into the particlefilter is used to re-assign random particle samples, making it move to the direction of themaximum posteriori probability density of target state. The experiments show that thismethod can not only quickly and accurately track the infrared human target, but also excludethe interferences of halos and background.