Pedestrian detection from sparse point-cloud using 3DCNN

This paper proposes a LIDAR-based pedestrian detection method using 3DCNN. The proposed method converts a sparse point-cloud obtained by a low-resolution LIDAR to two-channels voxel representation that consists of the 3D object probability channel and the reflection intensity channel. To evaluate the performance of the proposed method, an experiment using real-world LIDAR data was conducted. The results show that the proposed method is able to detect pedestrians more accurately than detectors trained by other conventional features.