Robust Visual Positioning of the UAV for the Under Bridge Inspection With a Ground Guided Vehicle

Regular defect inspection of the bridge's bottom is necessary for the maintenance of the bridge. Usually, conducting such an inspection with traditional under bridge inspection vehicles (UBIVs) is high-cost and laborious. To improve the inspection efficiency, the newly developed unmanned aerial vehicle (UAV) technique may provide a promising alternative solution. As the global navigation satellite system (GNSS) is not available under the bridge, the robust positioning of the UAV is still a challenge during fully autonomous inspections. Although accumulated works have attempted to utilize visual odometry to localize the UAV, their performance may easily deteriorate under simultaneously existed varying illumination and intense light noises. To cope with this issue, we design a ground-air mobile system and a dual-source positioning algorithm to enhance the robustness of the UAV's positioning. Specifically, the ground part of the mobile system is a ground vehicle (GV) equipped with infrared markers, which provides the referential fiducials for the relative positioning of the UAV. To well identify the infrared markers, the spatial relationship between the UAV and the GV is first optimized by an observation model. Then, to guarantee the robustness of marker detection, both color image and infrared image are simultaneously captured, and a dual-source algorithm is proposed accordingly. To implement the algorithm, a candidate region containing the infrared markers is first identified by a deep convolutional neural network. Subsequently, this candidate region projects to the infrared image, and a searching-tracking-aiming algorithm robustly detects those infrared markers. Following the marker detection, the position of the UAV can be finally estimated by a perspective-3-point algorithm of the marker and an inertial measurement unit (IMU) compensation. To verify the performance of the developed positioning approach, we conduct various real-world experiments in challenging light conditions under the bridge. The results demonstrate that our ground-air system and dual-source algorithm can provide robust positioning for the UAV in intense light noises and varying illumination during the bridge inspection.