DARTBOARD BASED GROUND DETECTION ON 3D POINT CLOUD
Keywords: Point cloud processing, ground detection, bird eye view, adaptive grid, quasi-flat zones
Abstract. LiDAR (Light Detection And Ranging) laser scanners acquire 3D point clouds of real environments. The process consists in sampling the scene with laser beams rotating around an axis. By construction, the point density decreases with the distance to the scanner. This density heterogeneity is a major issue, in particular for mobile systems in the context of autonomous driving, as usually a single scan is processed simultaneously (instead of mapping applications that can integrate several scans, reducing the density heterogeneity). We propose a dartboard grid with cell size increasing radially in order to adapt the grid size to the point density. The effectiveness of this strategy is demonstrated by means of a ground detection task, a fundamental step in many workflows of analysis of 3D point clouds.