Research Of Video Face Detection Based On DIM3517

With the rapid increase of computational powers and availability of modern sensing,analysis and rendering equipment and technologies, computers are becoming moreand more intelligent. Face detection is one of an important field of computer vision.Face detection is the step stone to all facial analysis algorithms, including facealignment, face modeling, face relighting, face recognition, face verification, headpose tracking, and many more. By researching face detection algorithm, not only canunderstand a variety of face detection algorithm, and also can be applied it to computervision and image processing, to reduce the computational complexity and saveresources.This paper analyzes the algorithm of face detection based on AdaBoost, the currentmost algorithm of AdaBoost are only detect for frontal face, to limit its robustness inthe video. This paper combines the ellipse fitting and target tracking algorithm for anyhuman face pose detection and tracking problems in the video, introduce Houghellipse detection and Camshift target tracking algorithms, to improve the algorithm ofAaBoost detection in a single frame image detection and robustness in the videodetection. Finally, by transferring the improved algorithms of AdaBoost detection toDIM3517, which in the ARM Cortex-A8platform, this paper implement facedetection algorithm in the video in the embedded platform.The second chapter introduces face detection algorithm on AdaBoost, training ownAdaBoost face detector by using the MIT face database, and test at CMU facedatabase. Experimental results show that the algorithm has better performance forfrontal face, but performance is not satisfactory for non-frontal face and profiled face.The third chapter introduces the random Hough ellipse detection algorithm, bydetecting ellipse area in a single frame of the video, to determine the candidateregions of face, and then detect the candidate regions of face by AdaBoost detector,improve the detection rate of AdaBoost detection algorithm for a single frame ofvideo face.The fourth chapter introduces Camshift tracking algorithm, proposed Camshfittarget tracking algorithms and AdaBoost face detection algorithm, to solve AdaBoostdetection algorithm in any of the video Multi-pose face detection undesirableproblems and improve the detection algorithm robustness in the video. The fifth chapter demand for low-cost applications, on the cost-effective DIM3517ARM Cortex-A8platform, completed transfer and optimized the detection algorithmfor the hardware platform, realized the video people face detection system in theembedded platform.