The Research For Adaboost Face Detection Algorithm And Achieve It In Hardware Platform

Face detection refers to the process of determining the presence of all the face size, attitude and position information from the input images at any face recognition systems. Face detection is the key to achieving an intelligent face recognition, and it’s also a cutting-edge topics in the field of pattern recognition at present. With the improvement of science and technology, development of social, face detection technology has been widely appreciated and researched, it has many important applications in many situations. For example:index on digital figures, intelligent video surveillance, e-commerce, etc. But now, the traditional face detection systems mostly use the software on a computer methods to achieve, because of badly real-time,these systems can’t achieve SME-oriented and real-time requirements in reality engineering environment. With the continuous development of programmable logic devices, people expect to be able to get the miniaturization and high real-time face detection system based on FPGA.In this paper, starting from the classical AdaBoost algorithm, the algorithm was a comprehensive analysis and presented optimization methods. On the basis of the optimized Adaboost algorithm, use Canada’s mature DALSA Anaconda dedicated image processing card, the Matlab code in the optimized Adaboost algorithm is converted into efficient hardware description language--VHDL by the assistance of AccelDSP synthesis tool. Then face detection module was integrated into the FPGA of Anaconda image processing card by using Xilinx ISE. Finally, the face detection system based on FPGA was developed, it’s real-time and can meet the requirements of face recognition or meet a higher level of demand. The main contents are as follows:1.Algorithm Analysis:firstly, the development of face detection and the current mainstream face detection algorithm is studied, and select the most appropriate Adaboost algorithm as the theoretical basis of this article. Then I present and analysis the integral image, Haar features, weak classifiers, strong classifier, cascade classifier and other key technologies theoretical in Adaboost algorithm. Experimental analysis shows it has some shortcomings:integral image calculation too slow in Adaboost algorithm, and face windows have insufficient face. For this situation, in the case of detection rate was guaranteed, this paper presents an optimized approach to meet real-time requirements, while laying a foundation for higher levels of demand.2.Collaborative software and hardware environment:this study is based on the Canadian company DALSA Anaconda dedicated image processing card, use onboard Xilinx’s Virtex2P_VP20 as FPGA platform. Firstly, optimize and verify the Adaboost algorithm in Matlab, and convert it to a hardware description language--VHDL, by the help of Acce1DSP synthesis tool. Then synthesize the VHDL to the bit file by using Xilinx ISE, the bit file was integrated into the FPGA of Anaconda image processing card, test and verify the face detection module finally.3.Build system:On the basis of the resulting hardware-based face detection system, converge DALSA company’s Pantera SA 2M30 Series DS-21-02M30 surface scan camera, and verify the detection results of face detection system.