基于车载LiDAR数据的公路竖向净空自动化评估

现役道路基础设施管理过程中缺乏大范围区域内不同路段的现状或实时的竖向净空数字化资料，导致部分过高车辆撞击跨线桥或其他上空构造物的事故时有发生，造成了重大财产损失与人员伤亡。针对该问题，基于车载LiDAR数据构建公路竖向净空自动化评估方法框架。通过数据重构方法将复杂道路线形的车载LiDAR点云转化为简单的直线形式，利用基于线性索引的点云数据分块方法实现重构场景下车载LiDAR数据的条形、柱形与体素单元的快速分块，建立柱形单元非平面点初步滤波、基于K-Means与体素聚类的复杂LiDAR点云环境中路面优化分割流程。在基于条形单元划分提取道路边界后，利用体素聚类将路面上方点云进行划分。以提取的路面点云作为二维插值基准面，完成不同物体的竖向净空计算，并利用江苏南京市内的2条公路LiDAR数据的对算法框架进行测试。研究结果表明：所提方法在噪音存在的复杂LiDAR环境中可以有效分割出道路上方物体并完成竖向净空的计算；通过部分算法提取与人工标注结果的对比，显示公路1与公路2的竖向净空平均绝对误差分别为0.94、1.57 cm，具有较好的可靠性；在32 GB内存、Intel® Xeon® E5-1650 v4@3.6 GHz六核处理器的计算机上完成公路1与公路2竖向净空评估的平均时间分别为6.62、7.83 s·km^-1，算法效率可满足大尺度场景下的公路竖向净空自动化计算；相比于已有研究方法，所提方法框架考虑了车载LiDAR点云环境内的路面上测量噪音的存在，对变宽度路面条件复杂场景下的公路竖向净空评估具有更好的适用性。

Automated Estimation of Highway Vertical Clearance Using Mobile LiDAR Data

Owing to a lack of up-to-date and easy-to-update database of vertical clearance (VC) information on as-built large-scale highway infrastructure, over-height vehicles sometimes collide with overhead objects such as overpasses and cause severe casualties and property loss. To address this issue, an automated framework was proposed in this study to achieve VC measurements on existing highways using mobile LiDAR data. First, a data restructuring technique was applied to convert the complex highway alignment into a simple straight line. Next, a linear indexing-based method was used to partition restructured LiDAR data into different units (i.e., bars, pillars, and voxels) in an efficient manner. After pillar-based non-surface point filtering, a K-means and Supervoxel-based method was proposed to segment the pavement surface from LiDAR points in the presence of on-road noises. The road boundaries were then delineated as the road surface points were partitioned into a number of bars. The overhead objects within the road boundaries were clustered using a Supervoxel-like technique. Subsequently, the distance between each object and the road surface was measured, and VC estimation was achieved. Two LiDAR datasets of highways in Nanjing, Jiangsu Province, China, were used to test the proposed framework. The results indicate that the proposed workflow can effectively segment overhead objects and accomplish VC estimation on highways with complex road environments. The mean absolute errors (MAE) between the generated and manually measured VC results on Highway 1 and Highway 2 are 0.94 and 1.57 cm, respectively, which demonstrates the effectiveness and accuracy of the proposed method. The mean execution time of VC measurement on Highway 1 and Highway 2 is 6.62 s·km^-1 and 7.83 s·km^-1, respectively (RAM of 32 GB, CPU of Intel® Xeon® E5-1650 v4@3.6 GHz), which is acceptable with respect to practical application in large-scale highway projects. Compared with previous methods, the proposed framework considers the on-road noises in mobile LiDAR data and can be applied to highways with variable road widths.