Research On Algorithm Of Face Detection And Tracking Based On OpenCV

The technology of face detection and tracking is hot in the field of machine vision and pattern recognition. They have broad application on the aspects of automatic face recognition replacing the working-card, remote video conference, and smart home security system. The purpose of face detection and tracking is to make some systems to detect and track target faces in the video. But the exsiting face detection methods need too much time to detect multi-view faces and the accurancy is bad under the complicated environment. The face tracking technology can not track the target face in real time and exactly.Against these problems, the the Ada Boost algorithm is used to train three kind of classifiers in this paper. These classifiers are the frontal face classifier, the half-profile face classifier and the full-profile face classifier which can detect multi-view faces in the video. The Cam Shift algorithm(Continuously Adaptive Mean Shift) and other methods are utilized to track the detected faces exactly. For improving the speed of the face detection and tracking time, the skin search or image sampling preprocessing method is used to the origin images to decrease the search region. The Open Source Computer Vision(Open CV) Library is transplanted to the embedded system and then a face-detection-and-tracking hardware system on ARM-based embedded platform is built to track the detected face based on the result of face detection and tracking.In the experiment part, through comparing the result of skin search and image sampling, it demonstrates that the image sampling preprocessing is better than the skin search preprocessing in terms of face tracking accuracy and speed. Therefore, before face detection the image sampling preprocessing method is applied into the captured images with different compressing scale. In order to reduce the face detection time and tracking time without destroying the images’ quality greatly, 0.5 is choosed as the compressing scale in this paper. The image sampling increases real-time of the system efficiently. Through comparing the experimental data, it shows that the detection time and tracking time in the personal computer is less than face-detection-and-tracking hardware system that we built. But our system can satisfy the actual need and detect the multi-view faces with high accuracy and real time under the complicated circumstance.