Research And Development Of Vision-guided Robot Technology For Production Line Picking And Unloading

In today's big industrial era,vision-guided robots have been widely used in practical industrial production.However,as the diversity of the field environment increases and the gradual improvement of the standards need to be met by various projects,these factors undoubtedly increase the pressure on the robot system from the aspects of hardware support and software development.This project takes science and technology projects as the starting point.And we adopt multiple cameras to monitor the actual production line material tray,target placement area and so on.After image preprocessing and making identification for the workpiece,the mechanical arm can autonomously grab the workpiece.On the way,they pass through the calibration device to obtain the accurate position of the workpiece.At the same time,the coordinates of the placement point captured by the discharge area are transmitted to the robot system.Finally the accurate and high-speed pick-and-place material of the entire vision-guided robot is completed.The main work is as follows:Firstly,a multi-camera visual guidance system was constructed.According to the requirements in the actual project,we corresponded the industrial lens based on the resolution of the camera and the position of the actual workpiece.The problem of insufficient light is solved by selecting the correct type of light source and arranging them perfectly.The Modbus and RS232 serial communication protocol are used to connect the vision system and robot execution system,so that the whole visual guidance process can be realized.Secondly,a variety of image preprocessing algorithms are reorganized,dominated by processing speed.A set of algorithm flow that is most suitable for the subject is summarized: the average grayscale operation was performed on the input original image,and then processed image was mean-filtered.Next,to separating the target and the background,the most between-cluster variance threshold segmentation method performed well.At the end,using Canny operator edge detection was to complete the whole process of image preprocessing.In the process of target recognition,different recognition algorithms are adopted according to the environment of each camera located.The hough circle transform was used to detect the circular workpiece,and the minimum perimeter polygon contour method was used to determine the polygon workpiece.Thirdly,for the research of super-resolution reconstruction algorithm,this paper proposed a new reconstruction algorithm model.In order to solve the shortcomings of the traditional multi-camera registration reconstruction algorithm,such as time-consuming,poor reconstruction quality and great dependence on the environment.This project does the super-resolution reconstruction work on the single-phase machine,builds a lightweight neural network model to ensure the reconstruction speed.At the mean time,we optimize the kernel of each network layer to obtain the improvement of reconstruction quality.In this project,the image with the resolution of 430*321 was reconstructed to the image with the resolution of 1280*964 to meet the precision requirements of the project,thus reducing the cost of purchasing the camera..