Research On Multiple Objects Detection And Tracking Algorithm Of Visual Surveillance

Recently, Video Surveillance (VS) has widely application in various fields. Object detection and tracking are the key technology of video surveillance. Effectively detection and tracking multiple objects are the hot and difficulty problem. Especially effectively track multiple objects in occlusion is a difficulty problem. In this thesis, we firstly research the common algorithm of multiple object detection and tracking. Then we proposed an effective algorithm to dissolve the occlusion issue in tracking multiple objects, combing the feature of overlap region with the feature of RGB-based dominant color histogram (DCH). The primary work of this thesis included:Firstly, the thesis analysis three common object detection algorithm, Background Subtraction, Temporal Difference, Optical Flow and comparison the advantages and disadvantages of three algorithms. The foreground is regarded as the background of the traditional Background Subtraction when the gray of them is similarity. So the thesis proposed background subtraction based on RGB color space. The experimental showed that: the algorithm of this thesis is valid and robustness.Secondly, the thesis analysis the algorithm of model-based tracking, region-based tracking, feature-based tracking, and active contour based tracking, discussed the difficulty of object tracking. Aim at tracking multiple objects in occlusion scenes, the thesis proposed the algorithm which combing the feature of overlap region with the feature of RGB-based dominant color histogram (DCH). From the experimental results, the proposed algorithm can improve the validity and robustness of tracking multiple objects.