A framework of road extraction from airborne lidar data and aerial imagery

Abstract

This paper presents a new framework of road extraction from airborne lidar data and aerial imagery, consisting of five main procedures: (1) data fusion of lidar data and aerial imagery, (2) creation of pseudo-scanlines from the fused lidar data, (3) initial extraction of road segments per pseudo-scanline, (4) refinement of road extraction to enhance the initial extraction results, and (5) final extraction of road surface and centrelines. A rule-based edge-clustering algorithm with various constraints is proposed to obtain smooth, elongated road segments per pseudo-scanline. Multiple refinement processes such as k-nearest-neighbours clustering, rule-based identification of intersection nodes and spatial interpolation are implemented to eliminate false positives and to connect the misclosures caused by the occlusion from trees, buildings and vehicles. Finally, curve fitting is employed to obtain accurate road centrelines. The quantified completeness and correctness of five test results are 82.6 and 87.4 percent, 89.2 and 91.2 percent, 80.7 and 87.6 percent, 84.2 and 90.4 percent and 79.5 and 89.5 percent, respectively.

Keywords:

• Airborne lidar
• road extraction
• edge clustering
• k-nearest-neighbours clustering
• curve fitting