Building detection in Digital surface model

Abstract

Containing elevation information, Digital surface model (DSM) data plays an increasingly important role in remote sensing image processing, especially in the field of building detection. However, due to the lack of spectral and textural information from DSM data, few attempts have been made to use DSM itself for building detection with satisfactory results. Most researchers fused airborne images and DSM data to enhance the detection accuracy. In this paper, a novel building detection technique by only using DSM data has been proposed. The building detection process is performed in three steps. First and foremost, the ground surface is reconstructed by employing the Minimum Description Length (MDL) principle. After that, the non-ground objects including buildings and trees are extracted from the original data. Finally a morphological method is applied to remove trees so that buildings can be recognized. To validate the performance of the proposed method, gray values of randomly selected points have been compared between original DSM and reconstructed ground surface and the building detection rate of proposed method has been calculated. The validation result shows that our proposed method, which only uses DSM data, has satisfactory detection rate and acceptable detection error. We prove that DSM data by itself is enough for building detection.