Detect residential buildings from lidar and aerial photographs through object-oriented land-use classification

Relating less directly to the physical reflectance from remote sensors, land-use analysis is comparably more challenging than land-cover studies, especially for residential land-uses. This research presents an object-oriented approach to detect residential land use of buildings directly from lidar data, aerial photography, and road maps to enhance urban landuse analysis. Specifically, the proposed methodology applies a multi-directional ground filter to generate a bare ground surface from lidar data, then uses a morphology-based building detection algorithm to identify buildings from lidar and aerial photographs, and finally separates residential buildings using a supervised C4.5 decision tree analysis based on seven land-use characteristics of buildings. Experiments based on the 8.25 km² study site located in Austin, Texas proved the possibility and efficiency of directly detecting and identifying residential buildings from remote sensing images with 81.1 percent of residential buildings correctly classified.