Face Detection System Design Based On The SOPC

The digital surveillance market is expected to reach110billion RMB yuan by2015in china. The digital surveillance is increasingly important in heavily traffickedareas, airports, campus, public buildings and so on. The real-time detection of humanface is one of the most important applications of the digital surveillance. With thedemand of portability, miniaturization and low price for face detection system, facedetection is particularly important in embedded systems. Embedded face detectionsystem will greatly expand the applications of face detection to people’s lives. But thegeneral embedded systems is not good at processing large amounts of data, especiallyinvolving the algorithm with high computational complexity. FPGA+ARMarchitecture-based embedded systems can effectively improve the speed of thesystem.Face detection technology on PC is mature because of the face detection andrecognition algorithms proposed in the past few decades. One of the most popularface detection algorithms is the Adaboost classification technique, offering significantadvantages in terms of speed and accuracy over other algorithms. Therefore, the facedetection algorithm used in this paper is based on Adaboost algorithm architecture.This design is based on FPGA+ARM SOPC’s face detection system todetermine the order of Xilinx’s Zynq-7000processor chip, hardware developmentplatform is Digilent’s Zedboard development board. The main content of this paperare:(1) The principles of face detection algorithm based on AdaBoost that proposedby Viola and Jones are introduced. The two key factor that influence the effect ofAdaboost face detection algorithm are clarified: Haar-like features and Integral Image.In-depth study of the weak classifier selection, building and other issues.(2) The Zynq-7000chip architecture, the internal processing system (PS),programmable logic (PL), PS and PL interconnect interface and development kits areintroduced in the paper. The on-board resources of Zedboard development board ispresent. A complete hardware development platform is built by using thedevelopment kit provided by Xilinx after its interface resources configuration(3) By using the Intel’s OpenCV on PC, the design of face detection systembased on Adaboost algorithm is finished.The procedure of training sample usingOpenCV is described. (4) Using Zynq-7000as the carrier, the design of the face Detection System isrealized according to the method of co-design of hardware and software in this paper.The PS fulfill the design of video capture module and face detection module. The PLcompleted the design of video output module and hardware acceleration module.A SOPC-based Face Detection System using Zynq-7000as its key processor isdesigned, which realized the image acquisition, processing, testing, result displayingand other functions. Finally, the hardware platform of face detection was built. Theexperimental results proved the detecting speed can reach17.8frames per secondwhen the input image is640x480pixels. The detection rate is93%,which can thedemand of real-time face detection.