Automatic Generation Method of High-Precision Maps Based on Semantic Segmentation
In the field of automated driving,high-precision maps have become an essential component due to their high precision and rich semantic information.However,existing mapping methods often target a single research object and require human participation in the information extraction and mapping process,which is difficult to meet the demand for automatic generation of maps with rich semantic information.Therefore,this paper proposes a method for automatic generation of high-precision maps based on semantic segmentation of multi-category point clouds.Firstly,based on KPConv deep learning network,feature extraction module and data augmentation module are added to improve the effect of multi-category semantic segmentation.Secondly,feature points are extracted from different classes of point clouds after semantic segmentation,and a vectorization model is established using Bessel curve fitting to finally generate high-precision maps with rich semantic information.In this study,33 classes of point cloud semantic segmentation are achieved using urban road environment LiDAR point cloud data,and the MIoU reaches 70.59%,and the performance of multi-class segmentation is good,so that high-precision maps with rich semantic information can be automatically generated.
万方数据
维普
