Application Of Machine Vision On Sorting System Of Industrial Robots

Due to the need to increase productivity and reduce labor costs,the scale of robot market in China has been the first in the world for many years.The use of industrial robots instead of artificially engaged in tedious,repetitive and dangerous operations has become an important measure for companies to increase their work efficiency and productivity.The visual sorting of industrial robots is an important application branch in the field of robotics.It has direct engineering application significance,and at the same time it involves many technologies such as optics,digital image processing technology and data communication technology.The efficiency and quality of visual sorting for industrial robots often depends on the accuracy of the visual recognition of the target object and the accuracy of positioning.In this paper,effective methods for object recognition in two different working situations are proposed.The main research work has the following points:(1)A vision sorting system scheme was designed,including a hardware system and a software system.The visual sorting system work platform consisting of industrial robots,image acquisition equipment and industrial control computers was completed;In the Microsoft Visual Studio 2010 development environment and based on OpenCV machine vision library,the digital image processing program and MFC-based PC software interface design were completed.The software system is mainly divided into three major components: control module,visual module and communication module.(2)For the recognition of rule-shaped objects under simple conditions,the limitations of traditional recognition methods through experiments are analyzed.On this basis,a target recognition method based on feature fusion is proposed in this paper.By this method,the three features of the contour length,area and rectangularity of the connected region contours of the target object in the image are extracted and form the feature vector,and finally the visual classification of the target object is accomplished based on the idea of k-nearest neighbor(kNN)classification.There are two main differences from the traditional algorithm.One is that in this paper,the similarity measure is calculated by assigning each feature a corresponding weight value instead of the Euclidean distance or Manhattan distance usually adopted in traditional kNN.In addition,it is not necessary to try to set an appropriate value for k in this paper.The algorithm proposed in this paper overcomes the shortcomings of the traditional methods,and achieves the visual recognition of the target object by merging multiple features.(3)For the recognition of irregularly shaped objects under complex conditions,the performances of three feature extraction and matching algorithms(SIFT,SURF and ORB)are analyzed in this paper.In the images of this paper,the numbers of feature points extracted by SIFT and SURF are small.Therefore,in this paper,based on the extraction of feature points in ORB algorithm,an improved algorithm of false match elimination is designed.The improved algorithm eliminates the mismatched feature point pairs and has better performance.