Fast pedestrian detection using a modified WLD detector in salient region

The research of pedestrian detection has attracted more and more attention in recent years. This paper proposes a fast pedestrian detection method using a modified Weber local descriptor (MWLD) and salient region detection. Significantly different from the previous pedestrian detection methods which usually scan the whole image with the sliding window, our method only scans the salient region, which can speed up the pedestrian detection. Inspired by the HOG method and Weber local descriptor (WLD), we present a MWLD method to character the sliding window for pedestrian detection. Experimental results on INRIA pedestrian dataset and Daimler Chrysler (DC) pedestrian Benchmark dataset validate the effectiveness of our proposed MWLD detector. Comparing with the HOG and HOG-LBP methods, our MWLD method performs better. Furthermore, we also validate that using saliency detection is effective for speeding up the pedestrian detection.