Hardware architecture for high-accuracy real-time pedestrian detection with CoHOG features

Abstract

Co-occurrence histograms of oriented gradients (CoHOG) is a powerful feature descriptor for pedestrian detection. However, its calculation cost is large because the feature vector for the CoHOG descriptor is very high-dimensional. In this paper, in order to achieve real-time detection on embedded systems, we propose a novel hardware architecture for the CoHOG feature extraction. Our architecture exploits high degree of fine-grained parallelism and adopts an efficient histogram generator combined with a linear SVM classifier. The proposed architecture is implemented on a Xilinx Virtex-5 FPGA and it achieves real-time pedestrian detection on 38 fps 320×240 video. That is more than 100 times faster than the execution on a state-of-the-art Intel CPU.