Research On UAV Autonomous Navigation Technology For Bridge Detection

With low cost,high extensibility and efficiency,the Unmanned aerial vehicle(UAV)is gradually applied in the field of bridge detection.Not only has it reduced the cost of testing,it has also been able to offset the detection efficiency by looking at the parts of the bridge that humans cannot reach.However,because of the complexity of the bridge structure,it is easy for a drone to collide with a bridge.At the same time,the bridge detection is a repetitive task,and the route is usually fixed.The ability to enhance autonomous navigation in complex environments will greatly improve the safety of the UAV and its usefulness in bridge detection.So the autonomous navigation of drones has been studied.The main research is as follows:By reviewing the relevant literature,the characteristics of UAV navigation were analyzed.Combined with the application scenario of bridge detection,a navigation scheme of UAV based on bridge point cloud model registration is proposed.The scheme uses the drone to fly around the bridge,obtain the whole point cloud model of the bridge,and then register the point cloud of the bridge and the local point cloud obtained by the UAV in real time to get the position of the UAV relative bridge.Aiming at the bridge point cloud model generation method,the method of generating point cloud model based on binocular stereoscopic vision technology is proposed,and the key algorithm of the method is analyzed.The three-dimensional measurement technology based on binocular stereoscopic vision is realized by software.And the feasibility of the method is verified by binocular stereo vision measurement system.For solution of the UAV attitude solving method of point cloud registration,RANSAC algorithm is used for coarse registration of point cloud,and then the coarse registration results as Kd-tree improved ICP registration algorithm input,so as to improve the accuracy and speed of registration.The point cloud processing algorithm is also studied.Finally,in the Visual Studio 2013 development environment,combined with OpenCV and PCL library,using C++ to achieve the algorithm in the program.The experiment was carried out by using the drone of M100.The results of the experiment show the reliability of binocular stereo vision technology.Point cloud registration experimental results show that the binocular camera panning in turn 50 mm and 100 mm,150 mm and 200 mm case,through the calculation of point cloud registration of binocular camera positioning error of 2.2 mm and 2.8 mm,3.4 mm and 6.6 mm,proved the feasibility of the proposed scheme.