Fusion of laser and monocular camera data in object grid maps for vehicle environment perception

Abstract

Occupancy grid maps provide a reliable vehicle environmental model and usually process data from range finding sensors. Object grid maps additionally contain information about the classes of objects, which is crucial for applications like autonomous driving. Unfortunately, they lack the precision of occupancy grid maps, since they mostly process classification results from camera data by projecting the corresponding images onto the ground plane. This paper proposes a modular framework to create precise object grid maps. The presented algorithm creates classical occupancy grid maps and object grid maps. In a combination step, it transforms both maps into the same frame of discernment based on the Dempster-Shafer theory of evidence. This allows fusing the maps to one object grid map, which contains valuable object information and at the same time benefits from the precision of the occupancy grid map.