The Realization Of A Face Detection System Based On Adaboost Algorithm With Using DSP

As an essential part of the face recognition technology,the face detection technology has developed from a newly emerging thing into current research focuses at home and abroad in recent ten years, especially being extensively developed in the security surveillance and other fields. Currently, there are much mature face detection algorithms, among of which the Adaboost learning algorithm is the most popular method. Later, a variety of Adaboost algorithms are formed on the basis of the algorithm. The algorithm processing performance and data storage of board of DSP has been greatly improved since the 20th century, which guarantee that face detection algorithm of the complex data structure can be implemented in the embedded systems, in this paper, the face detection algorithm has realized on the DSP development board which core chip is the DM642 chip of TI Company.The training process of face detection algorithm is introduced in detail, including the select of the Haar features and the method of integrogram, and the MIT face training library is also used to training the classifier. Then according to the Adaboost algorithm.the Haar features and the method of integrogram calculate the rectangular Haar features that are used to realize the process of training classifier. Including the training process of weak classifier.stage classifier and cascade classifier composed of a series of stage classifier. And then using the trained cascade classifier to detect face in VS2010 which embedded the Open Computer Vision(OpenCV) Library,at the same time, both of the classification results are compared. We find that the trained cascade classifier by author using Adaboost algorithm is better than that in OpenCV library, whether to a frontal, tilt or to a rotate face. And the miss rate of face detection by auther is lower half than that, to reach 4.3%. The detection library only containing the frontal face even does not exist the false detection. However, the false detection rate of 11.7% is higher than 5.3% of the classifier offered by OpenCV.After that the face detection system is constructed based on the DM642 development platform. The platform is composed by DM642-PCI development board, video camera and the LCD monitor. And the face detection software system is established by the software CCS2.2 which embedded the DSP/BIOS. Making reference of the face detection function code implemented on PC and according to the features of the DM642 chip to program and make an optimization which contain data conversion of the float-point to fixed-point, using look-up data, inline function, reduce the number of array dimensions, using the DM642 built-in EDMA function and so on. After the joint debugging of hardware and software, the function of the face detection is achieved ultimately. For the size of 352x288 image, the time of detecting a image is 300ms and the running time of the algorithm is 200ms. The running time of capture?processing and display basically satisfy the real-time video processing.