Automated filtering of fa?ade defect images using a similarity method for enhanced inspection documentation

Unmanned Aerial Vehicles (UAVs) are routinely utilized to capture images of building facades for structural health inspections. These images are often taken at regular time intervals, capturing overlapping views to ensure complete coverage of the facade. However, this practice generates numerous highly similar images, leading to inefficiencies in documentation and potentially biased analysis due to redundant data. Hence, this study proposes an automated image data filtering method based on defect similarity. Given facade images where defects are detected, this method applies key point detection and descriptor matching to measure image similarity concerning defect presence. Subsequently, a selection strategy is proposed to produce a minimal set of representative images covering all defects. This method was validated using a dataset of UAV-captured facade images across three facades, successfully deriving the set to six representative images. Overall, the proposed method offers a practical solution to reduce the quantity of images required for documentation, thereby enhancing the efficiency of UAV deployments for facade inspection.