Cascaded Sliding Window Based Real-Time 3D Region Proposal for Pedestrian Detection
Jun Peng Hu, Tao Wu, Hao Fu, Zhiyu Wang, Kai Ding · 2019
Pedestrian detection is an indispensable technology for designing autonomous driving systems. This paper proposes a cascaded sliding window based real-time approach to generate 3D pedestrian proposals on point clouds only. After rasterizing the raw point cloud, a 3D sliding window is adopted to extract pedestrian candidates. Two features are proposed to improve the proposal performance. One is the central points density feature, which acts as a filter to speed up the process and reduce false alarms. The other is the location feature, including the density distribution and height difference distribution of the point cloud, which describes the profile and location of an object in a sliding window. The scores generated by this feature are then applied to non-maximum suppression (NMS) to remove sub-optimal boxes. Experiments on the KITTI 3D object detection benchmark show that our approach achieves state-of-the-art results among comparable methods, while maintaining the speed and accuracy trade-off.