Detection and Location of Steel Structure Surface Cracks Based on Unmanned Aerial Vehicle Images

Cracks of steel structure usually came from fatigue or material failure, which may cause structural failure and were not allowed to occur. Aiming at the flexible and convenient monitoring method of unmanned aerial vehicle, this paper proposed an image-based detection and location method for the cracks on the surface in time and avoid the potential safety accident. Considering the compatibility between the detection of crack and other surface damage such as the corrosion, the method was composed of four steps: super-pixel segmentation, damage area primarily detection, pixel-level identification and crack location. Firstly, the UAV images were pre-segmented into suitable and universal size by simple linear iterative clustering. Then, different crack segmentation data sets were established for detection and pixel-level identification, on which YOLO V3 and DeepLab V3+ models were trained. Finally, combining images with UAV flight records, a panoramic crack location and presentation method was proposed. The method was verified by the case of Urban exhibition hall of Beijing-Tianjin Cooperation Demonstration Zone. The results showed only the close-range monitoring method could obtain the relative structure position of crack area and realization the monitoring of cracks.