Real-time pedestrian detection based on edge factor and Histogram of Oriented Gradient

This paper reports a real-time coarse-to-fine pedestrian detection method which is based on edge factor and Histogram of Oriented Gradient (HOG). In this method, first, the edge factor of detect window is used in the coarse detection, and then HOG combined with linear SVM (HOG/linSVM) is used in fine detection. The HOG/linSVM is one of the most popular video based pedestrian detection methods, which has advantage in high resolution images but low processing speed. To overcome this problem, a simple and run fast algorithm is developed—the edge factor, which is the normalized edge number. The edge factor can distinguish pedestrians from some uniform background. Edge factor is used in coarse detection to filter out some edgeless sub-windows; the fine detection which HOG/linSVM is used to do final detection on windows that has passed coarse detection. Owing to some sub-window has be eliminated by the coarse detection, the processing time of fine detection can be shortened greatly and so that the whole detection speed is improved. In addition, preprocessing such as image zooming, extract region of interest have be used for further acceleration of the detection process. Experiments show that our method has good performance and greatly improved the detection speed especially in videos of simple environment.