Research On AGV Intelligent Positioning Technology Based On Machine Vision

AGV is the abbreviation of an automatic guided vehicle,which is used to transport materials and transport goods in the industrial environment.It has been invented for many years since it was invented in 1950.In the past,the AGV's movement method was generally repeated application in a single track and fixed scene.During the movement,it was impossible to perceive the changes in the surrounding environment,and at the same time,it could not be flexibly assembled in the production line.Even if the trackless navigation method represented by inertial navigation and laser navigation is accompanied by an increasingly mature industrial environment,it only pays attention to the relative position and movement direction of the AGV in the production environment.Tasks,and cannot respond to unexpected situations or overly complex tasks.With the rapid development of computer technology,machine vision technology has many applications in the industrial environment due to its low cost,high precision,and contactless advantages.The application of machine vision in AGV's intelligent positioning has always been a research hotspot.With the development and maturity of deep learning and convolutional neural network technology,AGV combining convolutional neural network and machine vision can solve more complex problems in actual industrial production.Based on the analysis of the latest convolutional neural network target positioning method,this paper proposes an improved convolutional neural network target intelligent positioning method.After training the improved neural network model through the data set,the high-performance computing platform and The verification test is carried out on the mobile platform,and the test results are excellent.The improved neural network model has practical application significance.The main research contents of this article include the following aspects:(1)The model structure and optimization method of the convolutional neural network is studied.It analyzes the existing network detection model of target positioning in different architectures and provides a theoretical basis for subsequent research and improvement.(2)Detailed analysis of Corner Net,and based on it,improved methods such as multi-scale feature fusion,central keypoint detection,and target frame regression are added to improve the influence of the in-target features on the original network target intelligent positioning,which can better Position multi-scale and overlapping targets.(3)Research on image enhancement and data set enhancement methods,use the network public data set and self-made data set to train the network,and test the trained model on the edge computing platform.Experimental results show that the improved network comprehensive positioning speed and positioning accuracy results are better and have high practical value.