Implementation Of Real-time Pedestrian Detection System Based On INTEL ATOM Smart Camera

In the field of computer vision, pedestrians are important object of image and video analysis. To fulfill removable and real-time pedestrian detection, not only software and arithmetic methods need to fulfill detection function but also a small intelligent vision platform needs to carry pedestrian detection arithmetic methods and obtain detection fluency,taking into account of the rate of correction.This paper mainly analyzes current mainstream pedestrian det ection arithmetic methods,screens of ACF+AdaBoost with high calculation efficiency and validate and test it with computer, takes targeted analysis on the vision platform hardware structure and miniaturization requirement of small camera and designs a hardw are system based on INTEL ATOM architecture. After the transplantation of pedestrian detection to small smart camera platform, relevant performance test showed satisfying results.The main work of this paper includes the following:The arithmetic part: Pedestrian detection arithmetic method describes image feature using ACF ten channel pyramid feature, increase detection speed using AdaBoost cascade classifier method and get the final result of pedestrian detection through data synthesis. To obtain better and faster effect, INTEL hardware pip acceleration and face detection part was included.Hardware platform design: we designed small hardware carrier based on pedestrian detection requirement. INTEL ATOM BayTrail series processor was chosen for customized smart camera. The independent hardware system not only fulfilled the function like video and image capture, resolution switch, auto exposure and auto white balance that normal camera has, but also has the function like hardware H.264 compression, storage and information transfer.