Estimating First-Floor Heights of Residential Houses from Google Street View Imagery Using Object Detection

Determining a building’s first-floor height (FFH) is essential for assessing its risk from coastal flooding. While elevation certificates provide accurate FFH data, they are only available for a small subset of buildings, causing a need for estimation methods. We developed a machine learning framework that uses object detection on Google Street View images to predict FFH for residential structures. The model detects houses and their front doors, calculates the pixel distance between the door’s base and the building’s base, and scales that distance using an assumed door height to estimate FFH. Validation against elevation certificates in Florida and Virginia yielded mean absolute errors of 0.28 and 0.31 m, respectively, outperforming estimates from the National Structures Inventory. Although the approach is limited by images where the house or door is significantly obscured, it offers flexibility by allowing for imagery from any source. This framework provides an efficient, scalable, and effective tool for estimating FFH, a critical attribute for coastal flood risk assessment.