The Research On Automobile Door Panel Welding Spot Recognition And Positioning System Based On Convolutional Neural Network And Binocular Vision

The National Development Program "Made in China 2025" pointed out that it is necessary to organize the research of intelligent production lines to improve the engineering and industrialization of intelligent manufacturing.In the research of intelligent ultrasonic welding production lines for automobile door panels,the identification of automobile door panel welding points is the basis of welding production.Only by accurately identifying the location of the welding point,multi-objective optimal path planning and collaborative control of multiple robots can be performed.There are many kinds of solder joints distributed on the door panel of the car.They are small,and there are a lot of interference items similar to the solder joints.All solder joints need to be accurately identified and classified and it's importance to prevent the occurrence of misidentification.The traditional method based on machine vision technology is prone to misrecognition of solder joints,while the detection efficiency is low and it need to take a few seconds to process.Besides once the production line runs faster,its recognition effect will become very poor;and this method cannot perform solder joint Classification,once the car door panel model changes,it needs to be readjusted.Therefore,in view of the above problems,there is an urgent need to find an efficient and feasible method for identifying solder joints.Convolutional neural networks can automatically extract the depth features of objects,and realize accurate recognition and classification of objects in complex scenes,which is not prone to misidentification.At the same time,its detection efficiency is very high,generally only takes milliseconds.The characteristics of the convolutional neural network match the requirements of the identification task of the solder joint of the car door panel,so it is undoubtedly a feasible method to apply the convolutional neural network technology to the identification of the solder joint.In this paper,the characteristics of the solder joint recognition task are analyzed in depth,and it propose a solder joint recognition method based on YOLO(You Only Look Once).The main research contents are as follows:(1)The improved YOLOv3 algorithm is used to identify solder joints.Firstly,this paper analyze the principle of YOLO algorithm,then according to the characteristics of solder joints,it appropriately improve the YOLOv3 algorithm by the network structure and loss function.Finally,the improved YOLOv3 network is used to realize the accurate identification of the welding points of the car door panel in real time.(2)The three-dimensional positioning of welding points is realized on the basis of identifying welding points on automobile door panels by using binocular stereo vision theory.First,the calibration theory of the binocular system is studied and calibration experiments are carried out.Then combined with the identification information of the solder joint,the semiglobal stereo matching technology is used to realize the matching of the solder joint on the left and right images.(3)Using the parallax method to locate the solder joint,the three-dimensional coordinates of the solder joint are obtained.The experimental results show that based on the YOLO algorithm and binocular stereo vision technology,the identification and three-dimensional positioning of automobile door panel welding points can be effectively achieved.