Large-scale building dilapidation assessment for high-density cities: An urban visual intelligence approach

Abstract

Understanding the dilapidation condition of existing building stock is the first step in urban renovation and renewal. However, current assessment methods rely on tedious in-situ inspections, and manual inspection reports filing, which are difficult to scale for community- or city-wide implementation. Based on a proved association between urban phenomena with their appearance, this paper proposes an urban visual intelligence approach for large-scale building dilapidation assessment, leveraging widely accessible geographical big data (e.g., street view images and footprints). This method is easy to scale for assessing the building mass without a need of labor-intensive and expensive site survey. A deep learning model was trained to automatically process the large volumes of street view images and detect building defects with a precision of 90.4 % and F1 score of 80.7 %. The detected defects are positioned in a geo-spatial context for large-scale mapping and assessment with a building-level granularity. Piloted in the Kowloon Peninsula of Hong Kong, the proposed approach successfully evaluates the condition of over 9,172 buildings within an area of 47 million m² in 4 h. Theoretically, the paper enriches the emerging field of urban visual intelligence by extending its realm to urban renewal. Practically, this research provides a robust and scalable solution for mass dilapidation assessment to inform policy-making and urban planning. The mapping results offer a new stream of quantitative data for future studies to understand the mechanism of urban decay.