Design And Algorithm Of Stereo Vision Processing System For High Voltage Line Inspection Platform

The rapid development of society relies on a stable electricity supply.The national power system is responsible for the production and delivery o f electricity,and also has great significance for the national economic development,national security,light and heavy industry,the steady improvement of people's livelihood and so on.With the increase in mileage and area of transmission lines,the equipment used for automatic inspection of high voltage lines has received more and more attention.At present,some inspection platforms detect the on-line obstacles using the equipment such as ultrasonic carried by themselves,which cannot classify specific barriers and have poor adaptability.With the development of computer vision and image processing,more and more researches at home and abroad have merged into high-voltage line inspection equipment and improve the autonomy of robot inspection.In this thesis,a solution to the obstacle identification and avoidance in high-voltage line inspection is put forward.The stereo camera is used as the hardware platform,and the research is carried out based on multiple view algorithm.In the outdoor experiment,the platform would swing due to lateral wind,vibration of body and other factors,which lead to the blur of picture.Ensuring that the image is true and clear,this thesis presents a method to the restoration of fuzzy images.First,model the fuzzy images,and analyze the cause of degradation and parameters needed to be calibrated.The fuzzy images are estimated by the quadratic Fourier transform and autocorrelation function of the image,and the estimated fuzzy angle and length are put in the image restoration algorithm.After verification,the restoration process can restore the fuzzy image accurately and ensure the normal inspection.Dampers,spacers,and suspension clamps are conventional obstacles on the high-voltage line,and have certain structural features and size characteristics.This inspection platform uses stereo camera as a hardware platform,obtain the depth information by acquiring the left and right images simultaneously and three-dimensional matching.In the three-dimensional matching process,the accuracy and computation time of local matching,semi-global matching and global matching are synthetically compared,finally the semi-global matching algorithm is adopted considering the real-time and accuracy requirement.In the obstacle detection,a method of obstacle detection with SVM and ANN is proposed.The HOG descriptor of the obstacle in the scene is extracted first and trained using SVM.Hard example is introduced in the training process to improve the initial inspection accuracy.In the primary barrier ROI range,use ANN to make a second judgement on the candidate area,ensuring the accuracy of location and category information.