Real Time Low-level Pedestrian Detection on a Moving Car

This paper presents a real-time pedestrian candidates detection algorithm, which is capable of handling the challenges of stationary as well as moving objects, utilizing a single car mounted monochrome camera. First, the system detects obstacles above the ground plane by utilizing the inverse perspective mapping (IPM) algorithm. A fast digital image stabilization algorithm is used to compensate the oscillation of the camera on the vehicle. Afterward, a low level pedestrian segmentation algorithm is developed to extract bounding boxes of potential pedestrians. Finally, robust symmetry search and symmetry filtering algorithms are applied to align the detection result to the pedestrian. Further more a novel "Pedestrian Detection Strip (PDS)" is introduced to reduce the calculation time by factor of six. The performance test methodology and experimental results are also provided. For the covering abstract see ITRD E140665.