Research On Face Tracking And Recognition Technology

With the development of information technology and universality of video conferencing, remote teaching and other media forms, image tracking, target identification and information security have been put forward higher requirements. Human face is ideal basis for tracking and recognition because of its own uniqueness and stability as vital biological characteristics of people. But fast real-time face tracking and accurate face recognition technology is facing many problems. For example, how to accurately detect position of the face in the image, how to solve the coverage, rotation and tilt in face tracking, and how to improve feature classification in face recognition, and so on.To solve the above problems, we propose a face tracking algorithm based on real-time Camshift and Harr detection and a near-infrared face recognition algorithm based on SVM and AdaBoost, and design some experiments. Experiments show that the algorithm can effectively overcome the target lost problem in state of coverage, tilt and rotation, and can meet the real-time requirements with better lighting robustness. Besides above, it can also improve the ability of feature classifier and reduce the calculation complexity, so that it has higher recognition rate than other algorithms. In face detection, several image pre-processing methods combined with adaptive illumination compensation and two-dimensional wavelet is used to reduce the adverse effects from changing light and background noise. Integral graph algorithm to Haar detection enhanced by AdaBoost training method can accurately locate face with high detection rate and achieve friendly non-contacted detection.In face tracking, a time backtracking algorithm based on Camshift is proposed to solve the problem of face coverage. Changing-weighted histogram model is introduced to effectively overcome the target missing problem in state of rotation and tilt. Oval template is created with angle variable for good tracking results can be closer to actual facial features.In face recognition, firstly, a image acquisition hardware system is designed. Secondly, the popular two-dimensional principal component analysis (2DPCA) algorithm in the near-infrared light is taken for face recognition. SVM classification algorithm using AdaBoost training method proposed for characteristic data can be quickly and accurately identify the characteristics in order to improve the overall recognition rate.Finally, we design some experiments to verify the face tracking algorithm and the performance of face recognition systems. Such as blocked face tracking, rotation and tilt experiment, stability test and real-time experiment. These tests show that the loss rate of real-time tracking algorithm is low. Experiments of recognition rate and algorithm complexity are designed to compare recognition rate and calculation of different algorithm. The experiments show that the algorithm is lower complexity, and the recognition rate can be up to 90%.