Human Detection Based on Modified C4 and Head-Shoulder Feature

Over the past few years, significant research has devoted to the human detection area. This paper proposes a fast and accurate human detection approach called C3 based on modified C4 framework. The C3 combines the features of ROI extraction, Sobel transform and the CENTRIST descriptor, together with the HIK SVM classifier. Besides, the head-shoulder detection method is also adopted in the proposed technique. It is shown in the experiments that the C3 successfully detects almost all the pedestrians and consumes detection time of 104ms/frame on a 640x480 image.