Precise free space detection and its application to background extraction

In an autonomous driving system, a precise reference map with static environment information is necessary to perform accurate localization and path planning. To build a static reference map, only the background parts in the scene must be extracted by removing dynamic objects in the scanned data. However, filtering dynamic objects out from the scene is currently an extremely difficult task. To solve this problem, we propose a ray binding method to detect precise free space and a background extraction method based on the free space information. The ray binding method is a novel approach to detect the free space of environments precisely and reliably by constructing a spherical triangular pyramid set from adjacent laser beams. This precise free space information can be utilized in the process of map building to remove dynamic objects automatically, enabling us to extract residual background points reliably. We experimentally verify that the proposed background extraction method using the free space information operates well in diverse real scenarios.