A Software Tool for Automated Extraction and Classification of Road Surface Markings from Mobile LiDAR Point Clouds

要使无人驾驶汽车能够自主工作，需要高度详细、最新和准确的道路地图。SUV搭载的移动的激光扫描（或LiDAR）系统可以在SUV沿沿着移动时每秒捕获超过600，000个点以及360度全景图像。如今，商用移动LiDAR系统可以在一天的驾驶中积累100千兆字节或更多的3D点云数据。所有这些点云都可以构建高清道路地图，帮助自动驾驶汽车在道路上定位，精度在10-20厘米以内。鉴于大量的移动的LiDAR数据，自动化的androbust软件工具，可以处理移动的LiDAR点云，以支持高清晰度道路地图的制作和更新，在一个有效的和高效的方式是迫切需要的。本项目的目标是利用3D移动的LiDAR点云开发一种具有成本效益的基于点的路面标记提取技术。要开发的软件工具包括点云预处理，自动提取和分类的路面标记的新算法。提取的道路标记信息可以用于建立和更新道路数据库。该项目的成果将提供一种新颖、具有成本效益和节省时间的方法，用于制作高清道路地图，以支持加拿大道路上的无人驾驶汽车。这种新工具可以补充或替代目前使用传统调查或基于图像的方法制作路线图的做法。

The highly detailed, up-to-date and accurate road maps are required to enable driverless cars to workautonomous. A SUV-borne mobile laser scanning (or LiDAR) system can capture over 600,000 points persecond together with 360 degree panoramic images while the SUV moves along the road. Today a commercialmobile LiDAR system can accumulate 100 gigabytes or more of 3D point cloud data in just one day's driving.All these point clouds allow to build up a high-definition road map that can help position a self-driving car onthe road to an accuracy of within 10-20 cm. Given a huge amount of mobile LiDAR data, the automated androbust software tool that can handle the mobile LiDAR point clouds to support the high-definition road mapmaking and updating in an effective and efficient fashion is urgently needed. The objective of this project is todevelop a cost-effective, point-based road surface marking extraction technology using 3D mobile LiDARpoint clouds. The software tool to be developed consists of novel algorithms for point cloud preprocessing,automated extraction and classification of pavement markings. The extracted road marking information canthen be used to make and update the road databases. The outcomes of this project will provide a novel,cost-effective and time-saving method for making high-definition road maps to support driverless cars on theCanadian roads. Such a new tool can be supplementary or replaceable for current practices of making roadmaps using either conventional surveys or image-based approaches.