A Multicriteria Evaluation Method for 3-D Building Reconstruction

Abstract

This letter proposes a multicriteria system to evaluate the accuracy of reconstructed 3-D buildings. Current 3-D evaluation methods are derived from 2-D pixel-based evaluation; however, the difference between 2-D and 3-D evaluation methods is not well presented in previous literature. Most 3-D building evaluation methods concentrate solely on rooftop accuracy while ignoring the degree of accuracy found with regard to walls. To address these problems, this letter designs a multicriteria evaluation system based on three components: volume, surface, and point. The volume accuracy component represents the traditional classification accuracy based on random samples. The surface accuracy component evaluates shape similarity which compares sample and reference buildings, including rooftops and walls, in a true 3-D environment. The point accuracy component measures distance at feature points between the sample building and the reference building. This multicriteria system aims to provide an improved evaluation method for building reconstruction using advanced algorithms and multiplatform data. The system is also expected to provide valuable information to guide applications with different accuracy requirements.