Fast Object Segmentation Pipeline for Point Clouds Using Robot Operating System

This paper presents a method to pipeline the segmentation process for point clouds using the Robot Operating System (ROS) and the Point Cloud Library (PCL). The pipeline's objective is to optimize the run time of a conventional segmentation algorithm by working within the Robot Operating System framework. It can be implemented using any system and in conjunction with a GPU. It shows the greatest reduction in run time for the least downsampled clouds. Therefore, it can be used for real-time safety-critical applications especially in scenarios where the point cloud is sparse or of highly uneven spatial density and should thus not be downsampled. It was developed for Obstacle Avoidance for Autonomous Vehicles and Drones where segmentation is only the first step of a larger pipeline involving obstacle detection and tracking. It was observed to reduce run time up to 31.3% on the KITTI data set and up to 44.4% on data collected from a 16 channel Ouster lidar at the Indian Institute of Technology, Hyderabad.