Face Detection System Based On ARM Platform

Face detection and facial expression recognition is an important field of scientificresearch, which uses the machine in place of human to detect, identify and track target.It is mainly used in industrial production, safety testing, bank security system, liferescue, automobile electronic industry and so on. Since the1970s, with the ubiquity ofnew information technology and media, face and facial expression recognition arebeing developed. Furthermore, the ever decreasing price/performance ratio ofcomputing coupled with recent decreases in video image acquisition cost imply thatcomputer vision systems can be deployed in desktop and embedded systems.Face detection means that given an arbitrary image, the detection is to determinewhether or not there are any faces in the image and, if present, return the imagelocation and extent of each face. Face detection from an image is a challenging taskbecause of variability in scale, location, orientation (up-right, rotated), and pose(frontal, profile), facial expression, occlusion and light conditions also change theoverall appearance of faces. How to design fast, more accurate face detection andtracking system is an important mission of this thesis.The thesis designs an intelligent face detection system based on OmniVisioncamera, which detect and track faces in the view of camera. ARM platform obtainvideo scenes through the camera, face detection algorithm used to determine andanalyze the video which containing human faces and locating the human face. TheDMA is used to transfer the image between the system bus and the T35-LCD. For thissystem, the core research mainly includes three parts: first, determining the displayprogram in the conditions of a fixed hardware; second, the face detection algorithm istransplanted to embedded platforms; third, improving real-time of overall systemthrough optimization algorithm.Several commonly face detection algorithms are firstly introduced, includingKnowledge-based methods; Appearance-based methods, Template matching methodsand Feature invariant approaches. We analyzed of algorithm complexity and scope ofthe code transplantation. The method of feature invariant approaches is emphasizedespecially. Then the basic principle of face detection algorithm is recommended. Onthis basis, the paper has been fulfilled following works:(1)With skin color model and facial features, first, we use skin color model toidentify and extract the region of the color of the video scene, the skin regions; then, we calculate the location and size of the target face, the color area which based on thefacial feature through the region marking.(2) The algorithm porting. Since the image processing on the PC and embeddedplatforms are different, image format and the selected chip limits the real-time of theembedded platform and has great relevance to the image output format, so, algorithmon the host computer must be to a large number of improvements.(3) In the real-time of the face detection systems, hardware architecture and codespace complexity and time complexity impact of his performance. In this paper, wehad developed the detection algorithms, camera drivers, display drivers, high effectivecomputation method and compiler features optimization algorithms to save computingtime, improve the system performance.