DGPolarNet: Dynamic Graph Convolution Network for LiDAR Point Cloud Semantic Segmentation on Polar BEV

Semantic segmentation in LiDAR point clouds has become an important research topic for autonomous driving systems. This paper proposes a dynamic graph convolution neural network cloud semantic using polar bird’s-eye view, referred to as DGPolarNet. are converted coordinates, which rasterized into regular grids. The points mapped onto each grid distribute evenly solve the problem of sparse distribution and uneven density clouds. In DGPolarNet, feature extraction module is designed generate edge features perceptual interest sampled by farthest sampling K-nearest neighbor methods. By embedding with original cloud, local obtained input PointNet quantize predict results. system was tested on KITTI dataset, accuracy reached 56.5%