Research On UAV Distribution Line Inspection Based On Multi-dimensional Vision

China’s distribution lines cover a wide area and involve complex terrain.Power supply reliability is closely related to residents’ production and life.In the coastal areas of Zhejiang Province,typhoons attack every year.After the typhoon passes through,a large number of power distribution equipment and communication base stations are seriously damaged.The emergency repair personnel cannot obtain the equipment damage information at the first time.They must visit the site in person to obtain the first-hand damage information,and cannot prepare the emergency repair materials in advance.Under such circumstances,it is impossible to complete the request of Zhejiang provincial government to resume power transmission within 72 hours after the typhoon transit.How to quickly enter the site in extreme weather,collect the damage on the site,prepare rescue materials in advance,and ensure that the damaged line returns to normal power transmission within the specified time is the difficulty and pain point of the State Grid Corporation of China in building a disaster prevention and mitigation power grid at this stage.Therefore,starting from the fact that the multi rotor UAV can realize manual replacement,this paper enters the disaster scene in advance through image recognition and obstacle avoidance inspection,so as to realize the goal of identifying the fault point of distribution line and returning relevant information,and create huge direct and indirect value for the society.This paper mainly proceeds from the hardware construction and software algorithm of the multi-rotorless drone,and cooperates with the function of the drone in the distribution network inspection,and the related work scene is used as an verification case.The hardware construction is the basic condition of the algorithm.Effective image recognition is the prerequisite for avoidance,and the high success rate avoidance is the premise of inspection path planning.Efficient posture control is a solution to a stable inspection,four Part of the loop ring is deducted.Four aspects are mainly done from the hardware construction and software algorithm of multi rotor UAV.The hardware construction is the basic condition for the implementation of the algorithm,effective image recognition is the prerequisite for obstacle avoidance,obstacle avoidance with high success rate is the premise of patrol path planning,and efficient attitude control is the solution to realize stable patrol.The four aspects are closely linked and indispensable.The main work is done in four aspects: first,comprehensively sort out the application scenarios of UAV at this stage,and analyze the technical difficulties of UAV flying at low altitude.The second is to build the overall framework of UAV operation according to the inspection requirements of UAV distribution line.Aiming at the images collected in the inspection process,the image processing methods are studied,and the methods of scan line optimization,edge correction and sub-pixel lifting are proposed to realize the optimization and effective correction of parallax map.Thirdly,for the obstacles in the inspection process,based on the depth information obtained by binocular vision,the scene approximate element decomposition is carried out,and the solution of feasible discrete path is studied.Fourthly,the attitude control of UAV is studied based on multi-sensor,the time synchronization mechanism is proposed,and the visual observation matrix is optimized by Kalman filter.The above research greatly improves the image recognition ability and obstacle avoidance level of multi rotor UAV in distribution line patrol inspection,and realizes the independent patrol inspection of distribution line after extreme weather through independent route planning.