Image-based Face Detection And Tracking Algorithm

With the development of information technology, a variety of information devices become the necessities of our life, and more and more attention has been paid on information security issues. As an authentication methodology, the technology of face recognition has become a hot research topic in applied mathematics, pattern recognition, computer application and other cross-disciplinary. The technology of face detection and tracking has a wide range of applications in many different fields including face recognition, content retrieval, video surveillance and so on. In recent years, a growing number of scholars devote on the research of face detection and tracking technology. Nevertheless, due to these and other factors such as facial expression, gesture, light, complex background, how to establish a real-time and efficient system of face detection and tracking becomes a formidable challenge.The thesis investigates the problems mentioned above, and then proposes a joint algorithm of face detection and tracking based on Real AdaBoost and improved Camshift from a practical perspective, along with an algorithm of pedestrian detection based on HOG and SVM. The main contribution of this thesis is list as follows:The face detection algorithm of Real AdaBoost is studied, thereby realizing the face detection system based on the Real AdaBoost algorithm. From the comparative experiment, we demonstrate that Real AdaBoost algorithm has higher detection rate and lower false alarm rate than AdaBoost algorithm. The detection rate of the face detector trained in this thesis reaches98.56%through testing on the9031images in the frontal subset of CAS-PEAL-R1database.The face tracking algorithm of Camshift is thoroughly investigated, and it has been improved to overcome the shortcomings in the specific application. From the comparison of tracking effect of the two algorithms in the recorded video images, it can be proved that the improved Camshift algorithm is more robust and accurate than the original algorithm. In addition, in order to enhance the accuracy of tracking, we construct a skin color model and a closes color model to check the validity of the tracking result and whether occlusion occurred in the process of tracking, respectively. Once occlusion occured, the algorithm of Kalman prediction would be used to predict the location of the tracking target. The experiment result shows that the inclusion of such a judgment can effectively enhance tracking accuracy and reduce error tracking.Long-distance face detection is widely used in many practical situations. However, small scale face detecting in the whole image is time-consuming and the detection rate is low. In view of this situation, this thesis proposes the thought of pedestrian detection for assisting face detection. To be more specific, we detect pedestrians in the whole image at first, and then locate the face accurately on this basis, the algorithm of pedestrian detection based on multi-scale HOG, SVM and AdaBoost is realized.