A Bridge Crack Detection and Localization Approach for Unmanned Aerial Systems Using Adapted YOLOX and UWB Sensors

The management and maintenance of the aging bridges can benefit from an efficient and automatous bridge inspection process, such as crack detection and localization. This paper presents a robust and efficient approach for unmanned aerial vehicle (UAV)-based crack recognition and localization. An adapted YOLOX model is used in the proposed approach to improve accuracy and efficiency of crack recognition, and hence to enable real-time crack recognition from the captured UAV images at the edge-computing devices. In this way, non-crack images can be recognized in real-time during data acquisition and be filtered out to relieve the burden of subsequent data recording. In addition, a self-organizing positioning system based on ultra-wide-band (UWB) sensors is employed in the proposed system to enable real-time UAV positioning and crack localization in GNSS-denied areas such as spaces underneath the bridge deck. Experiment studies were carried out to investigate the impact of the quantities of employed UWB base stations on the UAV positioning accuracy. Finally, the proposed approach is tested on a self-developed UAV system and the effectiveness is validated through laboratory tests and real-world field tests.