Point-Based and Object-Based Building Extractions in Urban Area Applying Airborne LiDAR Data
MetadataShow full item record
In this research, we proposed a novel point-based statistical approach for automatic building segmentation and extraction by analyzing the differences between two LiDAR returns, the change of variance, and density from LiDAR point clouds. Then we applied an object-based supervised classification algorithms namely support vector machine (SVM) with several LiDAR-derived features, such as height texture (NDSM, DEM), and contrast texture from co-occurrence matrix and intensity (amplitude of LiDAR response) to extract the building areas in comparison of the result of the statistical methods. Since the terrain is highly uneven and the normalized DSM was a crucial factor in both methods, we filtered the ground points using a new filtering method, which is a combination of the slope-based algorithm (Vosselman 2003) and statistical analysis of last-return of LiDAR data in order to establish the DEM. Furthermore, the accuracy assessment was tested using a four band high-resolution (one foot) digital aerial ortho-imagery. The results show that LiDAR data could be used as a very reliable and stable data source for building extraction in urban areas.