AUTOMATIC DETECTION OF BUILDING PROPERTIES FROM AERIAL PHOTOGRAPHS USING COLOR AND 3D CONFIGURATION

摘要

The buildings properties in spatial databases are very useful for urban planning, disaster mitigation, insurance business etc. But the databases in some governments are not freely available since they include private information. Therefore, the data acquisition mostly depends on image interpretation of photographs and field surveys. Consequently it is time- and labor-consuming and very costly to construct databases, and it is not easy to maintain the data up-to-date and to employ them in wide use. In this study, automatic detection method of building properties based on aerial photographs is proposed and the result is verified by the database made in a conventional way of data acquisition. The method consists of the following steps: (1) geometric correction and gray scale conversion of aerial photographs, (2) stereo matching and evaluation of building heights, (3) evaluation of building areas, (4) color and texture analysis, and (5) classification of buildings. Buildings are classified into the following categories: (a) detached houses, apartment buildings and multi-storied buildings, (b) wooden and non-wooden buildings. The method is applied for a target area in Tokyo and the result is compared with the ground truth data investigated by the authors and the database provided by of Tokyo Metropolitan Government. As a result, it is found that low-storied apartment buildings and multi-storied buildings are correctly identified and they can be classified into proper structural types.