The Research On Algorithm Of Vehicle Door Solder Joint Recognition Based On Machine Vision

At present,welding robots have been widely used in automobile production line in China,however,most of the traditional welding robots use the method of manual teaching to plan the welding path.When the product is upgraded,the welding path needs to be re-planned,besides,the teaching robot cannot adjust the path in real time according to the change of operating conditions.In consequence,the traditional automobile welding production line is in the disadvantage of low production efficiency and poor production quality.In response to the above problems,this paper introduces a method of real-time recognition of solder joints based on machine vision.This article takes the automobile door for example,mainly researches the recognition of solder joints based on machine vision.Firstly,it briefly introduces the theory of machine vision and the composition of industrial machine vision system.In the same time,it selects appropriate light source,camera and optical lens to obtain the image of automobile door solder joints according to actual working condition.Secondly,this article carries out a series of process procedures for the image of automobile door solder joints,including image graying,filtering,sharping and edge extraction.In the process of edge extraction,it introduces the method of wavelet maximum,in addition to some common edge extraction operators such as Canny.And then,it extracts the position of solder joints by a modified random Hough transform algorithm which can detect circle.Thirdly,this article extracts the characteristics of solder joint image to distinguish the disturbing information such as round hole.It extracts the texture feature of solder joint and round hole based on gray level co-occurrence matrix and local binary patterns separately.The experiment shows that both of them have a good distinction between solder joint and round hole.The texture feature based on gray level co-occurrence matrix is ultimately chosen for its lower feature vector dimension.At last,this article designs three classifiers to identify solder joint and round hole.One of them based on BP(Back Propagation)neural network,the other based on support vector machine whose parameter is optimized by grid searching,and the last one based on support vector machine whose parameter is optimized by genetic algorithm.Comparing the accuracy of three classifiers,we can know the classifier based on GA-SVM has the best performance,and it can recognize the solder joints accurately.