Research And Implementation Of Video Face Detection Algorithm Based On DSP

Video face detection, which is the first step of the automatic human face analysis and processing system, means face detection in a dynamic image sequence, and accomplishment of the goal of detected face positioning and background separation. Its speed and accuracy directly affects the overall system performance, so video face detection has an extremely important role in the field of human face information research.Currently under the situation of controllable illumination and the cooperation of users, static face detection technology is relatively mature, but the video face detection algorithm performance needs to be improved. Video face detection system requirements both reliability and real-time characteristic, and there are still some difficulties in doing video face detection to achieve high-performance video face detection with changing light conditions, complex background, etc.The system’s hardware platform uses the high performance CCD camera for video capture and TVP5146 video decoder chip for video decoding, and the main processor is TT’s TMS320DM6437 processor, at last the result will be displayed on the TFT-LCD panel.The real-time video resolution is 720 * 480 pixels.Based on the extensive research on domestic video face detection method, using the light compensation pretreatment added skin color detection method as the basic algorithm. In the image pre-processing stage, the chrominance components of the real-time video is compensated through illumination compensation algorithm, and the impact of light is weakened. Besides the flaws of skin color detection method for light sensitive is made up. In the face detection stage, using the clustering of human skin color in YCbCr color space, this paper chooses skin color detection method to complete the candidate face region detection.After filtering out the salt and pepper noise through the opening and closing operation, combined with regional growth connected domain method to determine the candidate face block, and using the feature parameters of face shape to exclude some non-face region. Finally to achieve precise positioning of the face by the human eyes detected. Color detection method has the advantages of small amount of calculation and high speed, the system can meet both the reliability and real-time requirements.This paper first completed video face detection algorithm and simulation tests in the MATLAB platform. And then the algorithm and system functions are implemented in the CCS 3.3 DSP integrated development environment, and carried out the real-time debugging by its integrated real-time operating system DSP/BIOS.Based on the hardware characteristics of the DSP, in the candidate face region detection stage,using down-sampling processing method and the optimization of the original C language code,the efficiency of detection system is significantly improved, and the goal is achieved that a real-time human face detection in dynamic video sequences.Experimental results show that the face detection system designed had a 90% accuracy rate, with a considerable robustness and practicality. The efficiency of optimized code has been significantly improved to process almost six frames per second, which is capable to meet the basic real-time requirements.