AUTOMATIC BUILDING EXTRACTION USING LiDAR AND AERIAL PHOTOGRAPHS

Abstract

ABSTRACT This paper presents an automatic building extraction approach using LiDAR data and aerial photographs from a multi-sensor system positioned at the same platform. The automatic building extraction approach consists of segmentation, analysis and classification steps based on object-based image analysis. The chessboard, contrast split and multi-resolution segmentation methods were used in the segmentation step. The determined object primitives in segmentation, such as scale parameter, shape, completeness, brightness, and statistical parameters, were used to determine threshold values for classification in the analysis step. The rule-based classification was carried out with defined decision rules based on determined object primitives and fuzzy rules. In this  study, hierarchical classification was preferred. First, the vegetation and ground classes were generated; the building class was then extracted. The NDVI, slope and Hough images were generated and used to avoid confusing the building class with other classes. The intensity images generated from the LiDAR data and morphological operations were utilized to improve the accuracy of the building class. The proposed approach achieved an overall accuracy of approximately 93% for the target class in a suburban neighborhood, which was the study area. Moreover, completeness (96.73%) and correctness (95.02%) analyses were performed by comparing the automatically extracted buildings and reference data.