Adaptive ROI-based autonomous pedestrian detection system

This paper presents an efficient algorithm to set adaptive ROI for detecting pedestrians in a moving vehicle environment. The algorithm analyzes the centroid of detected pedestrian in current frame and define centroid region where centroids of detected pedestrian are concentrated. Based on centroid region, adaptive ROI is updated for each different size of detection window in next frame. Experiments are conducted with the Caltech pedestrian dataset and proposed method not only reduces computation time but also maintains performance of conventional methods.