Semantic segmentation of surface from lidar point cloud

Abstract

Mapping the environment for robot navigation is an important and challenging task in SLAM (Simultaneous Localization And Mapping). Lidar sensor can produce near accurate 3D map of the environment in real time in form of point clouds. Though the point cloud data is adequate for building the map of the environment, processing millions of points in a point cloud is found to be computationally expensive. In this paper, we propose a fast algorithm that can be used to extract semantically labelled surface segments from the cloud in real time for direct navigational use or for higher level contextual scene reconstruction. First, a single scan from a spinning Lidar is used to generate a mesh of sampled cloud points. The generated mesh is further used for surface normal computation of a set of points on the basis of which surface segments are estimated. A novel descriptor is proposed to represent the surface segments. This descriptor is used to determine the surface class (semantic label) of the segments with the help of a classifier. These semantic surface segments can be further utilized for geometric reconstruction of objects in the scene or for optimized trajectory planning of a robot. The proposed method is compared with a number of point cloud segmentation methods and state of the art semantic segmentation methods to demonstrate its efficacy in terms of speed and accuracy.