Power Facility Safety Protection Monitoring System Based On Binocular Vision

Power facilities are an important basis for the production and transmission of electric energy.With the development of UHV transmission technology,it is of great significance to detect the safety distance between large construction machinery and electrical equipment.Based on binocular vision,the process of autonomous recognition and ranging algorithm under ambient light was developed in this paper.Take the intelligent detection of the cable as the entry point,an effective security protection monitoring system that combines embedded development boards and terminals was built.Firstly,aiming at the lighting effect and noise interference problems of cable images under environment light,Multiscale Retinex algorithm（MSRCP）with color protection can reduce image distortion caused by color casts while solving the light normalization.Then,noise is filtered using a bilateral filtering algorithm.The experimental results show that the MSRCP algorithm has a higher effect than histogram algorithm and gamma algorithm for the enhancement of image.Then,the traditional segmentation algorithm based on edge or threshold has a single feature.Cable image segmentation model based on Deep Lab V3+,using Mobilenet V2 as a feature extraction network and combining the Void Space Pyramid Pooling Module as the encoder part.The model reduces the amount of parameters to improve training efficiency.In this study,the Deep Lab V3+ network reaches 91.52% and 93.91% on MIOU and MPA indicators respectively,effectively splitting cable targets under multibackground,multi-distance and multi-attitude.Finally,Morphological algorithms process the mask map.Afterwards,the pixel coordinates at the center of the cable target are calculated by means of contour fitting and the construction of the smallest rectangular geometryFurthermore,binocular stereo vision ranging accuracy is related to focal length,baseline and parallax.The binocular camera baseline and focal length are fixed according to the ranging range.Zhang’s calibration method calibrates the camera and the optimal distortion parameters are corrected by experimenting with the size,position,and number of rescales on the checkerboard.After the image is corrected by the polar line,the semiglobal stereo matching algorithm SGBM is used and the parallax search range,SAD window and post-processing parameters are adjusted to obtain the optimal parallax.The experimental results show that the maximum error of SGBM algorithm is 5.92%,which is 3.61% lower than the maximum error of 9.53% within the specified distance of BM algorithm and meets the requirements of real-time.Finally,between the Latte Panda development board and the PC terminal,the server and the client transmit the image encoding through the UDP protocol.The terminal decodes the image after receiving and calculates it in the process of recognition ranging algorithm.Through outdoor experimental testing,the system has the functions of autonomous recognition ranging and host computer display,which can play a role in safety protection monitoring.