Research On The Key Technology Of Visual System Of Obstacle-crossing Robot Along High-voltage Ground Wire

High-voltage transmission line inspection robot can assist or replace manual inspection,but for the lines across the primeval forest and big rivers,communication is blocked,and machine vision becomes an effective method to ensure the safety of robot walking and obstacle crossing.In this dissertation,a vision system of power line inspection robot is proposed to enhance the image under complex lighting environment.On this basis,obstacles in front are detected and identified when runnubg along the ground wire and the ground wire of the hand eye is accurately positioned when crossing obstacles.The main contents of this dissertation are as follows:(1)The proposal of key problems of vision system for the inspection robot.Firstly,the operating environment,composition and operation of the inspection robot are analyzed.The visual system of the inspection robot is constructed from two aspects of hardware and software.Then through the analysis of the image of the actual application scene,the research needs of three key issues in the visual system are proposed.(2)Image enhancement algorithm in complex lighting environment.Aiming at the problems of uneven illumination,low contrast and image noise in the field complex illumination environment,an image enhancement method based on adaptive MSR and fuzzy enhancement was proposed.Firstly,Haar wavelet transform is used to decompose the image into low frequency and high frequency information to separate the image illumination from details and noise.Then,for the low-frequency signal,a multi-scale Retinex fusion coefficient adaptively adjusted according to the light intensity information is adopted.The resulting multiple reflection images are weighted and fused to realize the correction of non-uniform illumination.For high frequency signals,according to the distribution rule of wavelet coefficient histogram corresponding to noise and edge detail,the fuzzy enhancement algorithm adaptively adjusts the index of nonlinear stretch power function to realize noise suppression and detail enhancement.Finally,the final enhanced image is obtained by inverse wavelet transform of the image enhanced by low frequency and high frequency information.The image enhancement experiments were carried out on the ground wire and obstacle images,and five objective image quality evaluation criteria were used to evaluate the enhancement effect.At the same time,the validity and feasibility of this method are verified by comparing with other seven advanced image enhancement algorithms.(3)Obstacle detection method based on feature fusion.Aiming at the problem that most of the obstacle detection methods of transmission line are still in the research stage of close range and single view angle,an obstacle detection method based on feature fusion is proposed.Firstly,the image enhancement algorithm is used to preprocess the image to correct the non-uniform illumination and enhance the image details.Then,based on EDLines algorithm,the ground wire is detected quickly.The bounding box algorithm is improved through the ground position,the area and the aspect ratio range of the obstacle target in the image,and the obstacle target area is proposed to quickly and accurately detect the position of the obstacle target in the image.For each detected obstacle image block,the global feature Hu moment and the local corner feature ORB optimized by bag of visual words method were extracted,and the two were weighted and fused.Finally,the "one-to-many" support vector machine algorithm of decision tree type is adopted to classify the shockproof hammer,obstacle group,suspension clamp and background,among which the particle swarm optimization is adopted to optimize the key parameters of the SVM of the radial basis kernel function.The experimental results show that the detection and identification rates of anti-vibration hammer,suspension clamp and obstacle group within the range of 1 to 5m reach 96.4%,92.6% and 90.5% respectively,which verify the effectiveness and practicability of the method.(4)Hand-eye ground line positioning method based on texture recognition.In order to solve the problem of inconsistency of ground texture caused by different types,different surface conditions and different illumination conditions,a new method of ground texture visual location based on texture recognition is proposed.Firstly,the image enhancement algorithm is used to preprocess the image to correct the non-uniform illumination and enhance the image details,so as to provide high-quality image for the location of ground wire.Then,the ground image is segmented into image blocks of the same size by moving the scanning window,and the texture feature vector of each image block is extracted.Then,naive bayesian binary classifier is used to classify the ground image block.In the construction of feature vectors,statistical texture features,GLCM and local binary mode features are optimized by genetic algorithm.Finally,the location parameters of the ground line in the image are obtained through the linear fitting of the center point of the ground line image block with error detection through the random sampling consistency(RANSAC)algorithm.The validity of feature selection based on genetic algorithm is verified by experiments.In the case of different illumination conditions,different angles and different heights,the positioning experiments were carried out on the protection wire of steel-cored aluminum strand,ground wire of steel strand and OPGW ground wire,and the positioning accuracy was 82.4%,84.6% and 79.5%,respectively,which verified the effectiveness and engineering practicability of the method.