Automated Extraction of Forest Road Network Geometry from Aerial LiDAR

We developed an algorithm that was designed to create a spatial database of a forested transportation network using aerial LiDAR. The algorithm uses two main attributes, LiDAR intensity values and ground return density. The road extraction process was developed using aerial LiDAR from McDonald-Dunn Research Forest near Corvallis, Oregon, U.S.A. The road extraction process requires X, Y, Z coordinates, intensity values, canopy type, and the maximum road grade. To compare the results of the process, nine road segments were field surveyed with terrestrial LiDAR. The result of the road extraction process resulted in 80% true positives, 34% false positives, 20% false negatives, and 38% true negatives in identifying forest roads. The average absolute value difference in the road width between the two data sets were 1.1m, while the cut/fill slope differences were minimal (> 4%) and the difference in road cross slope was two percent. These results were comparable with other published studies that examined differences between LiDAR measurements and field measurements.

Raw project data is available by contacting [e-mail]

本研究团队研发了一种算法，旨在构建一个由空中激光雷达数据生成的森林交通网络空间数据库。该算法主要依赖两个关键属性：激光雷达强度值和地面回波密度。道路提取过程是基于位于美国俄勒冈州科瓦利斯附近的麦当劳-邓恩研究森林的空中激光雷达数据进行开发的。该过程需要X、Y、Z坐标、强度值、冠层类型以及最大道路坡度。为了对比该过程的结果，我们使用地面激光雷达对九个道路段落进行了实地勘测。道路提取过程的结果在识别森林道路时，达到了80%的真阳性、34%的假阳性、20%的假阴性和38%的真阴性。两个数据集中道路宽度的平均绝对值差异为1.1米，而切割/填充斜率差异极小（>4%），道路横坡度的差异为两个百分点。这些结果与其他研究文献中探讨激光雷达测量与实地测量之间差异的发现相媲美。原始项目数据可通过联系[e-mail]获取。