Inspection Robot Based On Machine Vision

With the rapid development of China’s industry,the number of places where electricity is used continues to increase,and a large number of high-voltage electrical towers and substations have been built.The emergence of intelligent inspection robots effectively solves this problem.The inspection robot has the advantages of stability,reliability and high efficiency.It can realize unmanned on-duty inspection of substations,and at the same time,it can perform continuous high-intensity work without being disturbed by ordinary external factors.In this paper,an inspection robot based on machine vision is designed,which is mainly used to help the substation staff to complete the inspection work.The robot has two modes: automatic inspection of preset sites and manual control inspection,which helps staff complete inspection tasks and saves manpower.First,understand the basic requirements of substation inspection,and formulate the functional framework of the inspection robot.Through the functional framework,formulate the relevant hardware framework to understand the technology involved in the relevant hardware,and carry out the hardware selection.In this paper,the hardware framework is divided into three layers: chassis,torso,and robotic arm gimbal.The chassis part is the motion control hardware layer with STM32 as the core;the torso part is the navigation recognition layer with jetson Nano B01 as the information processing center;above the torso is the gimbal layer that collects image data with the robotic arm and RGBD depth camera.After that,in order to match the hardware,the software control system framework is designed.The system used by the inspection robot is ROS,and each module is divided by functions,which are divided into basic motion control module,encoder feedback speed module,map creation module,and global path planning module.,Local obstacle avoidance path planning module,interactive software development and display module.This paper expounds each of these modules in detail,analyzes and explains the principle,which mainly focuses on the calculation method and theoretical basis of the speed control module of the motion control layer;map creation,global path planning and local path planning algorithms and algorithm workflows.Finally,the recognition algorithm used by the robot is introduced.The YOLO algorithm is selected by comparison,and the network model of the recognition algorithm is improved by adding an attention module and changing the loss function used in the image output.The performance improvement of the improved model is proved by the COCO data set.By using Label Img to make and calibrate the verification set required for the reading,the improved algorithm model can identify the reading of the digital diode,and at the same time make a training set to train it.Using the trained model to test the actual reading of the digital diode,the effect is good,and the reading accuracy rate reaches 92.8%.After the completion of the robot performance test,the main test items include start-stop,speed range,steering,obstacle crossing,obstacle avoidance,manual control of the upper computer and automatic site inspection.Analyze the experimental results,and think about the shortcomings and improvement methods by comparing with mainstream products on the market.