Comparison of Quality Measures for Building Outline Extraction

Abstract

To date, numerous automatic building detection methods using lidar data have been developed. However, in most cases the parameters used for the evaluation of the results are not specified, accuracy assessment results cannot be readily compared or no result evaluation is performed at all. In this study, accuracy assessment aspects have been analysed in order to find a generally applicable approach for the comparison of building outline extraction methods. An area‐based evaluation of accuracy has been selected as the most suitable due to its robustness, predictability and comparability between datasets. This assessment can be further supplemented with both object‐ and area‐based evaluations, procured solely with those buildings that were classified as true positives, in order to interpret the extraction results in more detail. In cases where a reference dataset of superior quality (such as low‐altitude aerial imagery) is available, an absolute accuracy evaluation is also relevant.

Building detection and regularisation using DSM and imagery information

“…Several evaluation approaches have been used in the literature, including object‐based, area‐based and RMSE (Rutzinger et al., ; Rottensteiner et al., ), together with polygon and line segments (PoLiS) (Avbelj et al., ). The area‐based approach (using completeness, correctness and quality) was chosen to cover all quality aspects (Potůčková and Hofman, ) and the results are presented in the left‐hand part of Table . The results of the object‐based approach (applied twice for the two cases of buildings larger than 10 m 2 and then larger than 50 m 2 ), together with the method using the RMSE of the orthogonal distances between the extracted vertices and the closest vertices in the reference, are presented in the right‐hand side of Table .…”

“…(c). Quantitative evaluation : Building polygons are rasterised. The rasterisation process may introduce errors, but it provides a more robust outcome (Potůčková and Hofman, ). Values for area‐based completeness, correctness, quality, RMSE of vertices and RMSE of the centroids of the polygons, as well as the execution time, are presented in Table .…”

Comparison Between Airborne Lidar Data and Digital Image Matching for Surface Reconstruction

“…The Presidente Prudente dataset reference polygons building roof outlines were derived using a manual image restitution process, following the Photogrammetric pipeline, performed in the ERDAS IMAGINE system. According to Potučkvá and Hofman (2016), there are several metrics available to perform the quality assessment of the extracted polygons that represent building roofs. This study will focus on the Polygons and Line Segments (PoLiS) metric proposed by Avbelj et al (2015) and in F-score (Sokolova et al, 2006).…”

Comprehensive approach for building outline extraction from LiDAR data with accent to a sparse laser scanning point cloud

“…In order to evaluate the quality of outline extraction, an automated area-based approach is applied. Its advantages in comparison with other evaluation methods are discussed in detail in [10].…”