Multi-Pedestrian Detection from Effective Proposal in Crowd Scene

Pedestrian detection is a challenging issue, especially when multiple pedestrians are close in space. In this condition, the detection score generated by a single pedestrian (SP) may be too low to indicate those mutually occluded pedestrians. To address the above problem, multi-pedestrian (MP) detector is applied to detect group pedestrians through sliding windows search, which heavily increases the computation burden caused by tremendous amount of candidate windows. To well balance the effectiveness and efficiency, we present a novel MP detection method motivated by the effective proposals idea. In order to localize and segment high-quality candidate windows, we introduce bottom-up region proposals (which are called effective proposals) into the MP detection process. From which, high detection rate as well as low computation complexity can be simultaneously achieved. Experimental results on two standard datasets have validated the effectiveness and efficiency of the proposed method.