Road network extraction from airborne LiDAR data using scene context

We presented a novel procedure to extract ground road networks from airborne LiDAR data. First point clouds were separated into ground and non-ground parts, and ground roads were to be extracted from ground planes. Then, buildings and trees were distinguished in an energy minimization framework after incorporation of two new features. The separation provided supportive information for later road extractions. After that, we designed structure templates to search for roads on ground intensity images, and road widths and orientations were determined by a subsequent voting scheme. This local searching process produced road candidates only, in order to prune false positives and infer undetected roads, a scene-dependent Markov network was constructed to help infer a global road network. Combination of local template fitting and global MRF inference made extracted ground roads more accurate and complete. Finally, we extended developed methods to elevated roads extraction from non-ground points and combined them with the ground road network to formulate a whole network.