Face Detection System Based On DM642

From the 21 century to the present, human face detection has been a hot point of research and its development is also particularly fast. The target of human face detection is to determine whether it contains human face from the video, in the case of detecting human face was marked in all the face of the location, size and posture from the video image. The basic idea is to use knowledge or the method of human face statistical model, according to the degree of matching between the image area and human face models to determine the face area of the image. Face detection has been widely used in the field of information and communication,video monitor etc.This topic studies how to quickly and accurately detect the human face in the image or video on DSP platform. And analysis the Adaboost human face detection algorithm in detail, and optimized. The main task is divided into three parts of algorithm, hardware and software :Hardware: according to the customer’s demand and the realities of today’s microprocessor development to build a platform for video input and output, and select the Texas instruments DSP TMS320DM642 chip as the core chip of the system. In the hardware system is also supplemented by CCD camera, video codec chip and video display.Software: the development platform of this system is in CCS(Code Composer Studio) 2.2, using real-time operating system DSP / BIOS to develop. And the system has carried on the modular design. The system is divided into video capture module, video display module, and face detection and tracking module.Algorithm: analyzes the theory and characteristics of a variety of face detection algorithm in detail, and has carried on the demand analysis of system, determine the use Adaboost algorithm based on the characteristics of Haar. At the same time, in the further study of Adaboost algorithm, through improved algorithm, and it can shorten the testing time, reduce the error rate, improve the ability of detection.Face detection system designed in the project can be input and output video at 20 frames of /s with 720 * 576 PAL format, with detecting and tracking on time.The largest number of face detection is 5.