Moving Human Tracking Based On Mean Shift Algorithm

Intelligent video surveillance has been one important component part of security protectionarea so far, which can guarantee public safety effectively and guard against attack of terrorism.The human body tracking in intelligent video surveillance systems is researched in the thesis. Inaddition, the current situation of the human body tracking and mean shift algorithm in bothChina and foreign countries is summarized in the paper. Improvements have been made toovercome the defect of the mean shift algorithm when it was used to track the fast motion humanor partial occlusion human. The improved algorithm is applied to a human body tracking systembased on the smart ball machine. The main content of this thesis are shown as follows:To overcome the defect of the mean shift algorithm that was used to track fast motionhuman, the improved mean shift tracking algorithm combined with skin detection is proposed.The initial skin model is created by the skin pixels of human face, and the skin model is used todetect the skin area in the current frame. The morphological method and geometriccharacteristics method etc are used to find the position of skin area which belongs to the humanface. Then the position of human can be calculated based on the position of face, which is treatedas the initial position of the mean shift iteration, thus the goal of tracking the fast motion targetcan be achieved.To overcome the weakness of the mean shift algorithm that was used to track partialocclusion human, the human body tracking algorithm based on dynamic selection of fragment isproposed. The human body is divided in the algorithm based on the vertical color information,then the discrimination between fragments and background is calculated. To suppress thebackground color interference, the fragment which has best discrimination to the background isselected to track. When the target is occluded partially, the unoccluded fragment which has bestdiscrimination to the background is selected to track.Finally, the two improved algorithms are synthesized to the human body tracking systemwhich is based on the smart ball machine.