Based On Compressive Tracking Combined With Multi-feature

Target tracking is an important research direction in the field of computer vision, it has important application value and research value in intelligent video surveillance, intelligent transportation, human-computer interaction, military and so on, so it has drawn many domestic and foreign scholars’attention. So how to design a kind of high stability, high real-time tracking algorithm has been the hotspot and difficulty of target tracking research due to the uncertainty of the target tracking scenario (light, shade, appearance, etc.).By the very sparsely measured matrix to extract the Haar features of image, Compress Tracking is a real-time algorithm and could deal with part of the occlusion to a certain extent, but the tracking effect is unsatisfactory in the conditions which severe occlusion, because it extract Haar feature which based on color information, sensitivity to light. And compression tracking algorithm using the motion model of fixed the size of the target tracking box, it does not accord with the actual, severely limits the applications of this method.This paper analyzes the advantages and disadvantages of the Compress Tracking to adopt the idea of sparsely measured matrix and Bayesian classifier, and puts forward a stronger robustness, better real-time tracking algorithm which based on compressive sensing.namely based on compressive tracking algoriyhm combined with multi-feature.Firstly, we use a more robust motion model, namely particle filter, because adoption of the monte carlo technology of particle filter, make the particles closer to the distribution of target, what’s more, it make it successful coping with the effects of various kinds of mutations. In addition, the motion parameters is affine transformation,successfully cope with the effects of the target deformation and ensure the consistency of the tracking rectangular box with the size of tracking target.Secondly, HOG feature is based on the characteristics of the texture information, not sensitive to light, so in this paper adopts two features (HOG+Haar) combined to together, increased expressive of feature group, to increase the robustness of the algorithm.Finally, we will propose the algorithm compared with other popular tracking algorithm from two aspects of tracking accuracy and tracking speed, and have done a lot of contrast experiments through the video sequences with different challenges, thus verify the feasibility of this algorithm.