Research On Moving Target Tracking Technology In Video Images

As an important branch of Computer Vision, target tracking technology is the foundation of high-level video processing and image understanding, such as object behavior analysis, the compression coding of video image, besides it is the key technology in the automation and real-time application of video surveillance, Therefore it is widely applied in military and civil areas.This paper analyzes the research status of target tracking and difficulties it faces. Aiming at eliminating the effect brought by illumination and occlusion, the paper puts forward an improved approach on the basis of research on classic tracking methods. Experimental results show that the improved approach performs better than original approach. The main contents are as follows:○1 This paper introduces research situation on target tracking areas both at home and abroad, and analyzes systematically the advantages and disadvantages of main classical tracking approaches. Based on it, the paper focuses on two approaches-one of them is based on the target representation of region covariance matrix, and the other is based on SURF matching.○2 Region covariance matrix is a new form of target representation, it can blend many target features together easily and efficiently. So as to enhance the resolution of covariance matrix, by introducing the conception of block combination into the matrix representation, and using SURF description for reference, the paper presented a new feature vector form. Meanwhile it also appends sub-region covariance matrix to represent the target. Experimental results demonstrate that the new covariance matrix representation is helpful for target tracking in complex scene.○3 In view of common problems in target tracking on the basis of SURF matching, the paper suggests many measures: at first, in order to enhance the accuracy and robustness of matched SURF points, the paper takes some simple methods to delete the wrong matches according to some restricted condition, then improves target location approach which can adapt to the situation when matched feature points are few, at last put forwards a simple template updating approach which plays an important role in avoiding tracking drift.○4 The paper blends the above two approaches to improve the tracking performance when the target is occluded. The fusion method takes an advantage of the two tracking approaches: the stability and accuracy of the former method (correspond to covariance matrix representation) when the target is not occluded and the stableness when target is occluded of the latter one (correspond to SURF matching). The tracking strategy of fusion method is as follows: the fusion method judges the situation of target according to the number of matched SURF points firstly, and then takes corresponding measures to tracking the target.