利用激光雷达点云生成城市级三维道路地图

Combine Laser Scan Data with Open Street Map to Produce a Three-Dimensional Road Map

摘要: 本文提出了结合激光雷达点云拓展开源地图(Open Street Map, OSM)生成城市三维道路地图的算法，称为PHT (Projection and Hough Transform)。该方法可分为室外通道处理，室内通道处理和坡度异常再计算三部分。室外通道主要是基于局部路面可用平面近似的假设，利用正交投影将三维道路投影成直线，霍夫变换(Hough Transform, HT)提取道路候选点集合，再拟合平面计算高度；室内通道主要是基于相关联的室外通道的高度由投影距离加和得到。最后针对坡度异常的道路，利用带权重的霍夫变换(Weighted Hough Transform, WHT)再计算。本文使用德国科隆市政府提供的机载激光雷达点云(误差约为20 cm)，为亚琛市建立了三维道路地图。结果表明，与OPTICS算法(Ordering Points to Identify The Clustering Structure)相比，PHT成功预测了87%的场景，大于OPTICS算法13%的成功率；该算法准确率更高，对于点云被遮盖的情况，点云密度的变化以及噪声点的干扰更具有鲁棒性。

Abstract: With the continuous development of computer technology, the method to acquire spatial data has updated rapidly. Three-dimensional digital map attracts so much attention to be developed. Generating a three-dimensional digital map requires a basic map. Because the Open Street Map (OSM) is open-source and free, it has received widespread attention. However, the height information of the road is very sparse in the OSM, and the mean square error is higher than 5 meters, which makes more and more researchers focus on the generation of high-precision three-dimensional maps. Due to the Light Detection and Ranging (LiDAR) point cloud’s high-precision characteristics whose average square error is about 20 cm, it can extend the OSM to generate high-precision 3D maps. This paper studies the method of OSM combined with LiDAR point cloud to generate a three-dimensional digital map. Due to the sampling characteristics of the airborne LiDAR used in the overhead view, the occluded area cannot be sampled. The method proposed in this paper can solve the challenge of occlusion. It is composed of 3 main parts: 1) dealing with indoor area; 2) handling with outdoor area; 3) applied Weighted Hough Transform (WHT) for recalculation. The main steps for dealing with indoor area are as follows: 1) The three-dimensional road surface is projected into a two-dimensional line by orthogonal projection. 2) To find a set of road candidate points, the line is fitted by Hough Transform (HT). 3) Random Sampling the Uniform Sample Consensus (RANSAC) combined with the least squares method (LSM) is used to fit the road plane according to the obtained set of candidate points. This paper proposes a method for estimating the height of an indoor road using the height of the associated outdoor channel which is added up with different weights according to their projection distance. For the road with abnormal slope, the Weighted Hough Transform (WHT) is used for recalculation. This paper uses the airborne lidar point cloud (root mean square error is about 20 cm) provided by the municipal government of Cologne, Germany, to establish a three-dimensional road map for the city of Aachen. The results show that compared with the Ordering Points to Identify The Clustering Structure (OPTICS) algorithm, PHT successfully predicts 87% of the scenarios, which is greater than the 13% success rate of the OPTICS algorithm. In conclusion, the accuracy of the PHT algorithm is higher. In addition, PHT is more robust to the occlusion problem, change of point cloud density and the interference of noise points.