Reducing search space for fast pedestrian detection

Region of interest (ROI) generation used to extract candidate windows which are likely to have pedestrians is an important step in pedestrian detection systems. In this paper, we propose a novel algorithm to reduce the number of candidate windows extracted as ROI to reduce the computation time of pedestrian detection. In our proposed method, we take advantage of the temporal correlation between the adjacent frames to discard the candidate windows in regions where pedestrians cannot be found. To do this, we compare the confidence scores computed for the ROIs in the previous frames in two steps and ignore the regions where all of the neighboring ROIs are classified as non-pedestrian. Simulation results show that our proposed algorithm can improve the speed of pedestrian detection significantly compared to the existing systems while maintaining comparable accuracy.