Research On Multi-Target Detection And Tracking Algorithm Under The Framework Of Hough Forest

Visual object detection and tracking is a key task within the field of computer vision. It has been extensively concentrated and researched, since its wide applications in many fields in our daily life, such as visual surveillance, medical diagnose and virtual reality. So the research of target tracking has great research and application values. In this thesis, with intelligent video surveillance system as the main research background, on the basis of Hough forest algorithm,problems of samples selection and target searching were researched and with the above as base, an robust and accurate moving target tracking system were constructed. The main content of this thesis as follows:(1) In the view of the problem that the traditional algorithm could make the target tracking fail when the color of the background is similar to the target’s, an improved online hough forest tracking algorithm is proposed. When the detection of each frame is completed, we extract the selected samples roughly using entropy characteristics, thereby, training samples with a more texture informative are obtained. Then sample similarity is calculated in the histogram feature space, while leaving the ones with low similarity and the negative samples are used to update the classifier, thus the problem of performance degradation of the classifier is eliminated due to inaccurate sample. The experimental results validate that the proposed method is able to solve the problem of color similarity and has wide practicality when compared to tradition online hough frorest tracking.(2) Aiming at the problem of inaccurate detection results caused by fixed target searching window in traditional hough forest, an iterative searching strategy using the weighted center of the target patch is presented. So the failurer of tracking caused by the the speed of the target changed fast is soved.The experimental results show that the iterative searching strategy can improve the tracking accuracy in the guarantee of the stability of the algorithm.(3) Based on the Hough forest multi-target detection algorithm and the single target tracking algorithm of hough forest, a multi-target tracking system was designed, and this system verified the feasibility and practicability of the algorithm.In this thesis, traditional hough forest tracking algorithm has been researched and improved from samples selection and searching window updating. The changed algorithm can track target well even when the target’s color is similar to the background’s or the speed of the target changes fast. At last, a practical multi-target tracking automatic system has been constructed.