UAV-Based Non-Contact Fatigue Crack Monitoring of Steel Structures

A novel computer vision methodology is proposed to monitor fatigue cracks of steel structures using unmanned aerial vehicles (UAV). Through a short video recorded by a consumer-grade camera carried by a UAV, this methodology tracks the surface motion of the structure under fatigue loading and detects fatigue cracks by revealing the differential motion patterns induced by the cracks. Three main steps are developed to process the video for fatigue detection. First, the global motion of the UAV embedded in image frames in the video is compensated through geometric transformation. The transformation matrices are computed and applied to the video frames to obtain motion-compensated frames. Next, feature points are identified from each image frame as natural targets to track the surface motion. Subsequently, with the compensated surface motion tracked by the feature points, a localized circular region scans the entire surface to reveal differential movement patterns for crack identification. Compared with traditional computer vision techniques that are based on edge features for crack detection, the proposed method detects the breathing behavior of fatigue cracks; therefore, it is able to distinguish true cracks from crack-like features and structural boundaries. Laboratory validation is performed to validate the proposed methodology.