Research On Obstacle Detection And Positioning System For Power Transmission Line Inspection Robot Based On Deep Learning And Binocular Stereo Matching Algorithm

The domestic consumption of electricity,the scale of power transmission line and the size of the national power grid have gradually expanded year by year with development of society.Therefore,ensuring the stable supply of electricity is of great significance to today’s social life,which led to much attention to the research of automatic inspection robots for power transmission line.The obstacles cannot be accurately classified by the traditional Ultrasonic and Lidar equipment which will also make it difficult to cross the obstacle accurately for the inspection robot.Such issues have been solved to a certain extent with the development of computer vision,but the traditional image processing methods are sensitive to changes in scale and lighting which make it difficult to adapt to the field work environment.Based on these considerations,this paper proposes a power transmission line obstacle recognition and location algorithm based on deep learning and binocular stereo matching algorithm,which uses embedded devices and binocular cameras as hardware platforms.The main research contents are as follows:1.An obstacle recognition algorithm of power transmission line based on improved SSD is proposed which increased online calculation speed.Having a deep network layer and a large number of parameters,the SSD algorithm is not suitable for direct deployment on embedded devices.So,in this paper the SSD algorithm has been improved.The convolutional layer is reduced from the original 23 layers to 7 layers,and the model size been reduced from the original 301 Mb to 2.05 Mb.The actual measured frame rate on the Jetson TX2 increased from 4.5FPS to 15 FPS.2.The semi-global matching algorithm SGM(semi-global matching)is proposed to extract the disparity map,and the WLS(weighted least squares)algorithm is used to smooth the disparity map,and the distance of the obstacle is calculated.3.The 4G technology and UDP protocol have been proposed to develop video remote monitoring system.The end-to-end data transmission has been achieved by using ipv6 address,which avoid the shortcomings of ipv4 to transfer data through an intermediate server.Therefore,the video transmission rate is greatly speeded up.Finally,the obstacle recognition and positioning results are displayed on the computer in real time.The research results show that the improved SSD algorithm can be easily deployed on embedded platforms,and the detection speed is much faster than the original algorithm.The combination of the stereo matching algorithm can basically achieve the real-time detection of the distance of obstacles.At the same time,the 4G technology and ipv6 technology also realized the real-time transmission of video.