Design Of Conveyor Belt Sorting System Based On Machine Vision

With the economic transformation of the country and the adjustment of industrial structure,intelligent manufacturing has become an unstoppable trend of industrial development.It is becoming more and more common for machines to replace workers in production.Traditional robots often use off-line programming to fix the motion trajectory,and perform some simple and repetitive tasks with poor flexibility.When combined with machine vision technology,the robot has an "eye" that can observe the environment and can replace the human eye to perform some complicated and high-precision work.In industrial production,sorting products on conveyor belts is a very important part.Due to the complicated situation of sorting operations,the labor intensity is high and the cost of manual sorting is high.It is easy to be affected by subjective consciousness.In response to the appeal issue,this thesis combined with robotics and vision technology to design a sorting system to replace the manual sorting task.There are two main core problems that the sorting system needs to solve.One is the problem of target recognition.The target on the conveyor belt moves at a fast speed,with many types and scattered distribution.And the difference between different types of targets is small.High efficiency identification algorithms must be used to meet the sorting real-time requirements.The second problem is how to track and target crawl.The same target will appear repeatedly in the image sequence,and we need to design the algorithm to mark it out.While tracking a single moving target,the position of other targets entering the grabbing area is constantly changing.An effective tracking strategy should be designed to ensure that each target can be tracked.This thesis provides detailed principle analysis and experimental demonstration for solving two problems of appeal.The main work content is as follows:(1)The construction and calibration of the sorting system was completed.The sorting robot uses the EPSON LS6-602 robot.Industrial cameras use Basler's acA1600-20 gm camera.The whole system was calibrated to obtain the transformation relationship between the robot base coordinate system and the tool coordinate system,the angle and running speed of the conveyor belt,and the conversion relationship between the image pixels coordinates and the tool coordinates.(2)The system quickly identifies multiple targets in the image.The deficiencies in the speed and stability of the gray-scale matching algorithm in the recognition of various types of targets are analyzed.Therefore,the ORB algorithm for quickly detecting and matching feature points is adopted.In order to solve the problem that the performance of the ORB algorithm is degraded when the rotation angle is large,the system calculates the contour centroid of the target and the template separately,and rotates both contour centroids to the horizontal direction simultaneously,eliminating a part of the angle and then matching.Select the template type with the largest number of matching points as the recognition result.(3)Designed a time-based target duplication algorithm and a conveyor tracking crawling strategy.Calculate the system time of each target in the image that leaves the camera's field of view,and bind it to the target's properties.Compares the time attribute of all targets of the current image with the previous frame image,and marks the target as a duplicate target if the difference is within the threshold range.The system first plans the robot tracking track.Calculate the difference between the current system time and the target time attribute,and get the time when the target enters the grab area to estimate the time required for the robot to track.Calculates the coordinates of the target grab point based on the conveyor speed.The coordinates of the target grab point are calculated based on the conveyor speed and tracking time.(4)Use multi-threading technology to implement the producer and consumer models to coordinate the target recognition task and track the crawl task simultaneously.A thread is responsible for identifying,and the recognition result is stored in the public queue according to the time attribute of the target from small to large.Another thread is responsible for tracking,taking data from the result queue,and sending the result data to the robot via TCP communication.In order to avoid the data being lost or blocked,the tracking thread only sends one data to the robot at a time,and then waits for the robot to capture the completion signal before sending the next data.Finally,the previous work was integrated,and the above algorithm flow was implemented based on Opencv and QT platform programming.It accurately sorts the targets on the conveyor belt continuously.