BUILDING FOOTPRINT EXTRACTION USING LIDAR DATA AND SPECTRAL INDICES FROM AERIAL IMAGERY

ABSTRACT: Light Detection and Ranging (LiDAR) is a remote sensing technology that determines 3D position of points by pulsing a laser to the earth’s surface. Building footprint extraction is an application of remote sensing that is useful in urban planning and disaster management. This study uses a computational pixel-based approach to create an initial building mask. An object based image analysis (OBIA) approach was then employed for refinement. The study site is near the Himogaan River in Sagay City, Negros Occidental with an area of approximately 300 square meters. LiDAR derivatives were generated from the LiDAR data while spectral indices were computed from an RGB orthoimage. The normalized digital surface model (nDSM) was created by subtracting the DSM and DTM in order to get the actual height of the objects. A ratio index (RI) image was computed by dividing the LiDAR intensity at 1064 nm with the red band of the orthoimage. A three by three (3x3) range filter was then applied to the resulting RI image to get a range image of the RI. Thresholding was done to the nDSM, number of returns, LiDAR intensity and range of RI for the initial building mask. Using OBIA, the initial mask was refined using geometric features like the ratio of perimeter to area. Accuracy assessment was done by comparing the result with manually digitized building polygons of the area. Classification results indicated higher accuracy in the extraction of building footprints with an overall accuracy of 94.76 % compared to a point cloud classification done in TerraScan of the same area which had a producer accuracy of 89.47%. With these detaile