Lidarbox: A fast and accurate method for object proposals VIA lidar point clouds for autonomous vehicles

Haziq Razali, Nizar Ouarti

Abstract: The use of LIDAR technology for autonomous navigation has gained prominent status over the recent years due to its sensing capability. In this paper, we present an object proposal pipeline for LIDAR point clouds catered towards the domain of autonomous vehicles. Our method is built on the assumption that the distribution of point clouds (in the nearest neighbours sense) with respect to its depth can be modelled by a linear function with added gaussian noise. Combined with our occlusion handling algorithm, we show the efficacy of our object proposal pipeline on the KITTI dataset with extremely competitive results.