Automated extraction of digital terrain models, roads and buildings using airborne lidar data.

摘要

Airborne lidar has become a commercially viable remote sensing platform, and can provide accurate elevation data about both topographic surfaces and non-terrain objects. Its capability of mapping topography and 3-D models of civil objects is uncommon to other remote sensing technologies. This dissertation presents a collection of algorithms developed for automatically extracting useful information from lidar data exclusively. The algorithms focus on automated extraction of DTMs, 3-D roads and buildings utilizing single- or multi-return lidar range and intensity data. The hierarchical terrain recovery algorithm can intelligently discriminate between terrain and non-terrain lidar points by adaptive and robust filtering. It processes the range data bottom up and top down to estimate high quality DTMs using the hierarchical strategy. Road ribbons are detected by classifying lidar intensity and height data. The 3-D grid road networks are reconstructed using a sequential Hough transformation, and are verified using road ribbons and lidar-derived DTMs. The attributes of road segments including width, length and slope are computed. Building models are created with a high level of accuracy. The building boundaries are detected by segmenting lidar height data. A sequential linking technique is proposed to reconstruct building boundaries to regular polygons, which are then rectified to be of cartographical quality. Then prismatic models are created for flat roof buildings, and polyhedral models are created for non-flat roof buildings by the incremental selective refining and vertical wall rectification procedures. Many attributes of these building models are derived from the lidar data. These algorithms have been tested using many lidar datasets of varying terrain type, coverage type and point density. The results show that in most areas the lidar-derived DTMs retain most terrain details and remove non-terrain objects reliably; the road ribbons and grid road networks are sketched well in built-up areas; and the extracted building footprints have high positioning accuracy equivalent to ground-truth data surveyed in field. A toolkit, called Lidar Expert, has been developed to bundle these algorithms and to offer the capability of performing fast information extraction from lidar data.