The Research Of Video Face Detection Method Under Complex Background

Face detection and tracking in the field of video surveillance, age verification, medical diagnostics, human-computer interaction has broad application prospects. Although there are a lot of face detection algorithm, but face image can be affected by various factors, including:lighting, perspective, attitude, aging, face, hair, face accessories and the like. The complexity of the human face of the non-rigid and the environment, so that even if the same person to detect differences in different conditions there may be huge, so use an algorithm to carry out all the circumstances of the face detection is impossible.In this paper, the moving target detection algorithm has been improved, and the integration of color feature, geometric feature, template matching three kinds of face detection algorithm of layered filtering and realize the face detection in static and dynamic images. Which includes complex background illumination changes, trees disturbance, noise and so on. Motion picture include existing online and real-time video sequences video sequences with a fixed camera shot. Details are as follows:(1) the video sequence pretreatment performed separately using the reference white illumination compensation algorithm, average filter, smooth it.(2) for video redundancy background information, principles and limitations of background subtraction method, using the background subtraction method and the improved Combination of three frame difference algorithm to overcome the limitations of a single detection method, comparison with other methods.(3) Comparative analysis of models and colors skin model, with improved YCbCr color space to build thresholding segmentation model of skin color area, then skin color area opening and closing operation, to obtain a crude face candidate area, and then labeled connected region complexion, make the face template matching, enabling precise positioning face.(4) Details of the MeanShift algorithm principle CamShift algorithm and KLT tracking algorithm, and finally using a modified P-KLT face tracking algorithm, method and article of the CamShift algorithm, MeanShift algorithm experimental comparison, It proves the effectiveness of the improved process.