Computer-Aided Civil and Infrastructure Engineering, Journal Year: 2024, Volume and Issue: unknown

Published: July 26, 2024

Abstract Street view imagery has emerged as a valuable resource for urban analytics research. Recent studies have explored its potential estimating lowest floor elevation (LFE), offering scalable alternative to traditional on‐site measurements, crucial assessing properties' flood risk and damage extent. While existing methods rely on object detection, the introduction of image segmentation expanded utility street images LFE estimation, although challenges still remain in quality capability distinguish front doors from other doors. To address these this study integrates Segment Anything model, foundation with vision language models (VLMs) conduct text‐prompt estimation. By evaluating various VLMs, integration methods, text prompts, most suitable model was identified estimation tasks, thereby improving coverage current based 33% 56% properties. Remarkably, our proposed method, ELEV‐VISION‐SAM, significantly enhances availability almost all properties which door is visible image. In addition, findings present first baseline quantified comparison image‐based The not only contribute advancing but also provide novel approach tasks civil engineering infrastructure tasks.

Language: Английский