Research On Real-time Face Detection And Tracking Based On Video Surveillance

Face detection and face tracking are important research themes in the topic of ComputerVision and Pattern Recognition. They have a broad application prospect in many fields such ashuman face recognition, facial expression recognition, video surveillance, human-computerinteraction, medical diagnosis and so on. Take the Video Surveillance Field for example, facedetection and tracking has greatly enhanced the Intelligent Video Surveillance System, enablesthe system to achieve target detection and tracking in the dynamic scene without people. Onthis basis, some follow-up operations can also be achieved, such as trajectory analysis, targetrecognition, alarm. This paper focuses on building a face object-oriented video surveillancesystem. Based on the domestic and foreign academic papers and research reports on human facedetection and tracking, the dissertation explores the issue of face detection and tracking with arelatively static background from video sequences. The main results of this paper aresummarized as follows:(1)Face detection is a complex problem because of it influenced by background and light.After study on pre-processing in video sequences, average evaluation algorithm and referencewhite algorithm are used to filter the interference color and compensate the picture's light.Background is separated by mean algorithm. Background subtraction method is used to extractmoving information. By pre-processing it can remove redundant background information andexpedite the follow-up face detection.(2)Analyze and compare the superiority of difference color space and propose an approachof skin segmentation based on improved YCbCr color space which instead Cb of Cg. Meantime,using threshold segmentation to establish a complete skin color model. The results ofexperiments show that skin color has better clustering in improved color space, and caneffectively segment skin color regions.(3)Combination with the skin segmentation method introduced in (2), we propose ahierarchical filtering face detection method. This method adopts a coarse strategy to finestrategy: First, mark color connected region. Secondly, candidate faces are roughly located withthe help of face geometric characteristics that are face area and face aspect ratio and Eulernumbers. Last, use the color distributing character of face to build the template matrix fortemplate matching, the faces are accurately located.(4)In the phase of face tracking, CamShift algorithm is used for fast tracking in video sequences that faces are searched in a local area instead of the whole image. To make up for thedisadvantages of the traditional CamShift algorithm, the CamShift based on the Kalman filterare put forwarded. Constantly updating the motion model can solved fast moving objectstracking and temporally occluded objects in face tracking process. This algorithm improves thetracking matching degree and accuracy, while also make the face tracking system to obtain ahigher reliability and robustness.By researching and realizing related face detection and tracking algorithm, this thesisdesigns and completes an experimental system using Motion Tracking Framework provided byOpenCV. By experimental verification, the system achieved good results and can be adopted inactual detect and track human face.