Real-time pedestrian detection on a xilinx zynq using the HOG algorithm

Advanced driver assistance systems (ADAS) are the key to enable autonomous cars in the near future. One important task for autonomous cars is to detect pedestrians reliably in real-time. The HOG algorithm is one of the best algorithms for this task; however it is very compute intensive. To fulfill the real-time requirements for high resolution images an efficient parallel implementation is necessary. This paper presents an efficient hardware implementation as well as a parallel software implementation of the HOG algorithm for pedestrian detection on a Xilinx Zynq SoC. The hardware implementation achieves a speedup of 2x compared to the parallel software implementation for high resolution images (1920 x 1080). Against state-of-the-art a speedup of 1.32x is achieved. The hardware implementation has a reliable detection rate of 90.2% using a classifier trained by an AdaBoost algorithm and a minor false positive rate of 4 %.