Font Size: a A A

Design And Implementation Of Visual Sorting Robot System

Posted on:2020-05-31Degree:MasterType:Thesis
Country:ChinaCandidate:W ZhangFull Text:PDF
GTID:2438330575960170Subject:Control Engineering
Abstract/Summary:PDF Full Text Request
The application of machine vision technology to industrial robots has become an important way of industrial intelligence.The main breakthrough point of machine vision technology for industrial robots lies in the intelligent classification and positioning of the target.Aimed at this situation,this design adopts convolutional neural network as the main technical means of target classification and orientation,the use of image connected domain centroid extraction and camera space coordinate transformation principle as the target positioning of the auxiliary method to improve the existing industrial controller,realized the intelligent sorting function of wedges with different letters.The system is mainly composed of mechanical control,target identification and target positioning.In the mechanical control part,TCP/IP communication is used to realize the communication between the upper computer and Galil-B140-M industrial controller.Galil-B140-M further controls the X-Y-Z three-axis mobile platform to realize the movement of the manipulator arm.The serial communication mode was adopted to realize the communication between the upper computer and STM32F103ZET6 single-chip microcomputer,and STM32F103ZET6 further controlled the opening and closing of the mechanical claw.Target recognition part compared with multiple classification methods of determination by convolution neural network to simulate the workpiece,which can identify the letters on the way to recognize speed,high recognition rate,the workpiece is put no requirement and no need to manually extract and establish the feature library,etc,through the study of migration training convolution neural network to overcome the image of less difficult.In the target positioning part,firstly,a series of processes such as grayscale,binari-zation,edge detection and centroid extraction of the connected domain are carried out on the workpiece on the conveyor belt to obtain the pixel centroid coordinates of the target.On this basis,the pixel centroid coordinates are converted into spatial coordinates through the principle of camera spatial coordinate transformation.This design completes the integration of the hardware part and the above three software parts,and provides three ways to sort the objects with letters randomly.The results of this paper have a good reference value for the research and development of similar systems.
Keywords/Search Tags:Target classification, Image processing, Convolution neural network
PDF Full Text Request
Related items