Slam Technology For UAV Distribution Line Inspection Based On IMU And Binocular Vision

In recent years,with the development of civil UAV(Unmanned aerial vehicle,UAV)technology,UAV line inspection has become an important way of power line inspection.Power lines are divided into high-voltage transmission lines and distribution lines.The main inspection object of the traditional UAV power line inspection is the high-voltage transmission lines,which are located in the open air,so it is easy to receive GPS signals to realize the UAV self-positioning.Distribution lines are generally located in urban areas,villages,industrial areas and other residential areas,where the environment is generally narrow and surrounded by numerous obstacles.GPS signals are easy to be missing and the positioning accuracy is insufficient,it causes UAV cannot patrol distribution lines.In order to solve the problem of UAV localization under distribution lines,based on binocular vision SLAM(simultaneous localization and mapping,SLAM),a binocular vision SLAM algorithm incorporating IMU(inertial measurement unit,IMU)is proposed in this thesis.Firstly,the theoretical basis of binocular visual SLAM with IMU information is introduced.The working principle of the SLAM system is introduced through motion model and observation model.Then,according to the sensor selected by the system,the working principle of the camera and the IMU as well as the coordinate system conversion relationship are respectively introduced,and on this basis,the stereo vision principle of binocular camera is discussed and the IMU is modeled.Then,the three working threads of binocular vision SLAM are studied deeply.In the tracking thread,Fast feature points and Brief descriptors are optimized to extract and describe ORB feature points,and then matching them according to hamming distance.ICP algorithm is used to calculate the pose of binocular camera,and the key frame generation rules are introduced.In the local mapping thread,the Bo W(Bag of words,Bo W)model and dictionary tree are introduced to describe the key frames,the rules of map point generation are introduced,and the system’s key frames and map points are further screened by local BA optimization.In the loop closing thread,the Bo W model is used to detect the loop of the system,and the loop correction is carried out on this basis.After that,a fusion scheme of IMU and visual information is proposed.At first,the pre-integration model of IMU is proposed to solve the problem of different working frequencies between IMU and vision system.Then it introduces the way of IMU system initialization to prepare for information fusion.Finally,the IMU information and visualinformation are fused in a tight coupling way with nonlinear optimization,and the information fusion efficiency is improved through the marginalization of sliding windowsFinally,experiment have been conducted to evaluate the proposed SLAM technology.A MYNT binocular camera integrated with IMU is used as the experimental platform of the system,and a simple distribution line environment is simulated with a street rack and a distribution line.Firstly,Kalibr tool is used to calibrate the binocular camera and IMU.Then,the distance measurement experiment of binocular camera is carried out under different ambient light or background.The experiment showed that the distance measurement accuracy of the algorithm is within 0.08 m.Finally,through the Eu Roc data set,the experiment of binocular vision SLAM with IMU information is carried out.The experimental results show that binocular vision SLAM fused with IMU information has higher localization accuracy and robustness than pure binocular vision SLAM..