Research For Face Detection Based On Adaboost Algorithm And Its Implementation With DM642

Face detection is a special object detection, which is based on digital image processing. Its purpose is to exclude environmental background image information on the static image or the dynamic video, while accurately detecting and locating the human-face image region. As an important technology of face information processing, face detection has become a popular research project in the field of computer vision. With the rapid development of DSP embedded system, it gradually established DSP’s dominant place in the field of digital image processing application. Compared to the face detection system based on PC, the face detection system based on DSP has many advantages, such as much cheaper price, more light-weight system, more efficient process capacity.The detection speed and the detection rate are important indicators when we evaluate the performance of DSP face detection system. Under the guidance of this idea, this article utilize Gentle Adaboost face detection algorithm to study and train a set of face detection cascade classifier, which is suitable for real-time detection on DSP. In order to achieve efficient and accurate face detection in face detection system based on DM642, this article apply the reasonable strategy of image detection and the DSP detection codes which are developed by author, and focus on optimizing the detection project.The main study of this article is divided into the following three categories:(1) On the basis of a deep understanding of Gentle Adaboost detection algorithm, utilize VC++and Opencv software development tools to train the suitable face detection classifier in PC platform, and the reasonable training samples guarantee the quality of classifier. After this work, complete the simulation of detection algorithm. In order to prove the feasibility of face detection classifier when it is applied on DSP detection system, this article compare efficiency with one of classifier trained by expert, aims to analyze the pros and cons of the article-trained cascade classifier in terms of detection rate and detection speed. (2) The system is with TMS32QDM642at the core, and this article select system peripheral chips reasonably, aims to propose own platform design ideas in terms of hardware module. Design the critical data structure of detection algorithm code and properuse the API code language of DM642, this article aims to introduce the code development in terms of software module. Successfully design the face detection system based on TMS320DM642, while implementing and ensuring correct operation of system.(3) In order to improve the detection speed of system, this article starts with the following three aspects:optimizing data structure of algorithm code, optimizing the parameters of CCS projects, decreasing the amount of image information on detecting video. Finally, detect the different human faces and several human faces of multi-poses with detection system. According to the recording results of detection, this article inspects and verifys the actual detection performance of detection system.