The Research On Detection And Tracking Algorithms Of Motion Object

Object detection and tracking is one of the most important research projects on the fieldof computer vision, which includes moving object detection, distilling, and tracking. It iswidely used in military, video surveillance, human-machine interaction, and industrialcontrols. However, numerous factors affect the performance of a detection and trackingalgorithm, such as the uncertainty of the object itself, background clutters, as well asillumination variation, and there exists no single detection and tracking approach that cansuccessfully handle all scenarios. Therefore, object detection and tracking algorithm hasimportant theoretical significance and practical value. The main goal of this paper is topresent some novel algorithm based on a single camera and a single video moving object,which can improve the accuracy and reduce time complexity of the existings. The major workincludes:(1) On the research of the motion detection, three main methods of the motion detection,background subtraction, temporal differencing and optical flow, are introduced. Afteranalyzing their basic principles, this paper proposes a novel object detection algorithm basedon scene illumination invariant to solve the problem that they are all sensitive to illuminationchanges. The core of the proposed algorithm is to modify illumination invariant features withthe blocking mechanism and locality gray sensitive histogram. Unlike the intensity values,this features remain the same even under dramatic illumination changes, and are used asinvariants for matching to detection algorithm. The experimental result shows that theproposed algorithm is much lower than fixed threshold in terms of the object false rate, andcan effectively eliminates the influence of illumination changes on the object detectionalgorithm. (2) On the research of the motion object tracking, the object sparse representation andthe particle filter framework are studied. After analyzing the weakness of L1tracker, a novelobject tracking method based on sparse representation is proposed to avoid thetime-consuming. In the particle filter framework, tracking problem can be cast as finding asparse approximation. Specifically, to find the tracking target at a new frame, each targettemplate is sparsely represented in the space spanned by target candidates and trivialtemplates. The sparsity is achieved by solving an L1-regularized least squares problem. Toreplace by the traditional reconstruction error, the multi-target coefficient vector is used tocomputer the weight of target candidate. The proposed tracker shows excellent performancein comparison with two previously approach and saves time at least80%in comparison withL1tracker.