Abstract
In this research, we study the effectiveness of images and point clouds for drivable area estimation. In previous approaches, the effectiveness of image and point cloud cannot be calculated and algorithm designers cannot interpret the role of image and point cloud. Therefore, we propose a network structure that explicitly calculates weights when integrating these features and analyze the effectiveness of image and point cloud. Experiments show that image information has more impact than point cloud for the recognition results. Especially, images features are used for recognizing surface of road, while point cloud features are used to recognize tall-objects.
