Research On Tilt Detection Method Of Power Towers Based On Binocular Vision

Power poles and towers are used to support overhead transmission wires and are essential equipment for the normal operation of the power supply system.The collapse of the tower will cause serious safety accidents.The tilt detection of the power tower is the guarantee for maintaining the safety of the power supply system.Nowadays,the total station or theodolite is commonly used to measure the inclination of the tower on the spot,which is labor-intensive,low-efficiency,and needs to be improved in convenience and intelligence.With the rapid development of today’s computer vision technology,non-contact measurement technology based on computer vision has received extensive attention.In this paper,computer vision technology and deep learning methods are used to study the tilt detection of power towers.Firstly,the structure and inclination characteristics of various power towers are deeply analyzed,and the specific location of the tower characteristics is determined as the diagonal side edges of the tower body or tower legs.Considering the influence of complex external environment,tower material and lighting,the characteristics of tower images in different scenarios are summarized,binocular vision technology and image segmentation method are studied,and a method for detecting tower tilt based on binocular vision technology is proposed.A comparative study of traditional image feature extraction methods and deep learning methods is carried out.Based on the maximum inter-class variance method and the probabilistic Hough transform,a traditional algorithm for feature extraction of towers is designed,and experiments are carried out using real images of power towers.According to this,a feature area segmentation algorithm of power towers based on Deep Lab V3+ network is proposed,but there are still problems of inaccurate edge extraction and great influence by illumination.Aiming at the shortcomings of the original algorithm,the Deep Lab V3+ network structure was improved,and the network was optimized by adding low-level features and incorporating the IBN-Res Net(Instance and Batch Normalization-Residual Network)structure.The accuracy of feature segmentation is improved,and the accurate segmentation of the side edge area of the power tower is realized.A binocular parallel tower tilt detection model is established.Based on the principle of polar line matching,the side-edge matching algorithm of the power tower is designed,and the fitting equation of the side-edge line of the tower is obtained by fitting the three-dimensional point cloud using the singular value decomposition method.The geometric relationship before and after the inclination of the tower is analyzed,and the inclination calculation model is constructed.According to the electric power industry standard,the absolute value of the measurement error of the inclination angle of the tower should not exceed 0.05°.The experiment shows that the system can accurately measure the inclination of the tower.This paper completes the design of the power tower inclination detection system,which meets the needs of engineering measurement and improves the efficiency and intelligence of the power tower inclination detection.