Data

[D5-01] Enhancing large-scale footprint extraction evaluation: a two-level approach with proxy data and building-unit matching

*Shenglong Chen¹, Yoshiki Ogawa¹, Chenbo Zhao¹, Yoshihide Sekimoto¹ (1. University of Tokyo)

キーワード：footprint quality evaluation, completeness assessment, building matching

Evaluation on large-scale extracted building footprints remain several challenges. Traditional accuracy evaluation metrics fails to fully reflect the accuracy and does not capture the complicated semantic relationship. Besides, the lack of reference data in some developing countries exacerbates these challenges. Therefore, this study proposes a two-level large-scale footprint evaluation method. In the first level, population and land use data are employed as proxy data to evaluate the grid-level completeness. In the second level, an optimized two-way area overlap method is used to evaluate the accuracy of the entire area and obtain an object-based semantic accuracy. To validate the effectiveness, Hyogo Prefecture and Susono City in Japan are used as study areas. The results indicate that the first-level method exhibits an improvement of 0.05 in grid classification accuracy and a 0.55 increase in completeness coefficient correlation. Furthermore, the improved matching method achieves a matching accuracy of 99% and takes only 13s to match 5,000 buildings.