LAUSR

In this article, we present a framework to detect pedestrians in presence of various real world challenges. The depth-level occlusion is addressed by a stereo-aided triangulation mechanism, where the ORB (Oriented FAST and Rotated BRIEF) descriptor is used to speed up the disparity estimation. An empirical formulation has been made to compute the maximum feasible window size during region proposals generation. The variation of unusual articulated postures is tackled with a shape-histogram representation that uses a set of oriented, high-frequency kernels to compute the gradient details; a set of co-occurrence texture cues is further taken into consideration to strengthen the resulting descriptor. We validate the efficacy of our method on three benchmark pedestrian datasets, where the obtained results are expressed in terms of five performance metric.