Road Segmentation Using Point Cloud BEV Based on Fully Convolution Network

Road segmentation based on deep learning methods is an important task in the field of unmanned driving using 3D Lidar data. Point cloud bird's eye view (BEV) is one of 3D Lidar data's import representation methods. In this paper, we introduce a new road segmentation model using point cloud BEV based on fully convolution network (FCN). We use the road data in the KITTI dataset to train a road segmentation model and analyze the impact of different feature fusion methods on segmentation performance. The FCN8s model was used to test the road segmentation performance of different road scenes. Experiments show that the multi-level feature fusion method can increase the performance of road segmentation, and the FCN model has a general segmentation effect in complex road scenes.