基于语义分割的高精度地图自动生成方法  

Automatic Generation Method of High-Precision Maps Based on Semantic Segmentation

作　　者：陈玉雨 杨洲 胡坚 周春城[1,2] CHEN Yuyu;YANG Zhou;HU Jian;ZHOU Chuncheng(Aerospace Information Research Institute,Chinese Academy of Sciences,Beijing 100094,China;Key Laboratory of Technology in Geo-Spatial Information Processing and Application System,Chinese Academy of Sciences,Beijing 100094,China;School of Electronic,Electrical and Communication Engineering,University of Chinese Academy of Sciences,Beijing 100049,China)

机构地区：[1]中国科学院空天信息创新研究院,北京100094 [2]中国科学院空间信息处理与应用系统技术重点实验室,北京100094 [3]中国科学院大学电子电气与通信工程学院,北京100049

出　　处：《自动化应用》2024年第16期241-244,251,共5页Automation Application

摘　　要：在自动驾驶领域,高精度地图因其高精度和丰富语义信息成为必不可少的组成部分。而现有制图方法往往针对单一研究对象,需要人力参与信息提取及绘图过程,难以满足自动化生成丰富语义信息地图的需求。因此,提出一种基于多类别点云语义分割的高精度地图自动生成方法。首先基于KPConv深度学习网络,加入特征提取模块和数据增强模块,提升多类别语义分割的效果。其次,使用语义分割后的不同类点云提取特征点,通过贝塞尔曲线拟合建立矢量化模型,最终生成带有丰富语义信息的高精度地图。该研究使用城市道路环境激光雷达点云数据实现33个类别的点云语义分割,MIoU达到70.59%,多类别分割表现良好,可自动化生成带有丰富语义信息的高精度地图。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 highprecision 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 highprecision maps with rich semantic information can be automatically generated.

关 键 词：激光雷达 点云 语义分割 深度学习 矢量化 高精度地图 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] TN958.98[自动化与计算机技术—控制科学与工程]