Design And Implementation Of Obstacle Identification Module For Inspection Robot Based On 3D Lidar

In recent years,robot technology has become increasingly mature,and more and more substations have begun to use inspection robots instead of manual operations.However,with the in-depth use of substation inspection robots,inspection robots have encountered new problems in identifying obstacles.At present,substation inspection robots mainly rely on visible light images to complete obstacle identification.This method cannot effectively identify obstacles in the case of strong light or poor light conditions at night,which seriously affects the robot’s inspection work.To enable the substation inspection robot to distinguish obstacles more stably,it is necessary to rely on sensors that are not restricted by light conditions.At present,3D lidar has been widely used in the navigation tasks of substation inspection robots.The data obtained is not affected by light.On this basis,the use of lidar data to complete obstacle discrimination can avoid information redundancy and light influence at the same time.Therefore,this article is based on 3D lidar sensors to obtain data,and applies deep learning technology to the inspection robot obstacle identification.At the same time,the relevant software and hardware environment of the substation inspection robot obstacle identification module is built,and finally a stable working substation is obtained.Obstacle discrimination module for inspection robots.The main contents of this paper are as follows:1.Construction of Obstacle Data Set for Substation.Summarizes the current research status of obstacle identification of substation inspection robots,and analyzes the functional requirements of obstacle identification of substation inspection robots.Collect 3D lidar point cloud data and camera images containing obstacles in the actual substation to construct a substation obstacle data set,and provide data support for the subsequent study of the identification method.2.Improved method for identifying obstacles in substation inspection robots.The obstacle identification method and deep learning network of the substation inspection robot are analyzed,and a deep learning network suitable for the obstacle identification of the substation inspection robot is selected and improved.The network improvement draws on the two improved versions of the SSD(Single Shot Multi Box Detector)network.The backbone network draws on the DSOD(Deeply Supervised Object Detector)network and is modified by Dense Net to improve the feature extraction capabilities of the network.The prediction module draws on the DSSD(The residual prediction module of the Deconvolutional Single Shot Detector network is designed to reduce calculation costs and improve recognition accuracy.Experiments show that the improved obstacle discrimination network effectively improves the accuracy of obstacle discrimination and reaches the preset accuracy index.3.The design and realization of the obstacle discrimination module of the substation inspection robot.Set up the hardware and software environment of the obstacle discrimination module of the substation inspection robot,transplant the cut and compressed obstacle discrimination network into this environment,and test the module.The experimental results show that the obstacle discrimination network after trimming and compression,with less accuracy loss,the network efficiency is greatly improved and the preset efficiency index is reached.At the same time,the various functions of the module are in line with expectations and can complete the entire process from data collection to recognition result display.