Research On The Moving Detection And Target Tracking Based On Video Sequences

The technology of video based on object detection and tracking is one of the hotspots in the field of computer vision, which is also the basic in the applications of smart surveillance, human-machine interface, mobile robots navigation, industrial robots hand-eye system and so on. In real life, moving target detection and tracking extensively involves in human tracking and identification, intelligent transport, traffic flow monitoring, etc. During the past few years, much effort has been made to enhance the performance of visual tracking by combining multiple features. In this paper, a fast algorithm of target detection in static background video sequences and a tracking algorithm is proposed based on region correlation descriptor, which can provide an elegant solution to fuse multiple features for representing the object. Besides, we suggest a method to find the best matching region by calculating a "centroid" estimated from a set of candidate points in searching window. Main content can be summarized as follows:1. In view of the fixed video sequence, a rapid method of moving target detection algorithm is proposed. Due to the moving target appears in different locations in each frame, it can be innovatively viewed as random noise. Therefore, the background model can be extracted quickly through the method of the means of limited frames. Then, this paper presents high-bit-layer exclusive OR(XOR) algorithm to extract moving target. Compare to traditional background subtraction algorithm, new algorithm avoid the influence of camera jitter and the sun shine. The effective and accuracy of new algorithm has been demonstrated by experiments.2. Considering the insufficient of extracting feature only from the gray level image traditionally, this paper presents a region correlation descriptor to fuse multiple features to represent the object. Firstly, the algorithm establishes a template, including the information of color, texture and edge gradient. Secondly Forstner distance is used as a criteria to measure the matching extent between template and the searching area . The experimental results show that the tracking efficiency has been enhanced in a large scale.3. In order to reduce the computation, searching area need to be set in the smallest external rectangular. Beside, to further narrow the search region, a pair of "return" window is proposed , which is the intersection of the smallest external rectangle and twice size of the templates. The effectiveness of new algorithm has been demonstrated by experiments.