Research On Object Recognition System Based On Machine Vision

In recent years,with the continuous popularization of industrial automation,more and more traditional manufacturing industries have begun to transform into mechanical automation production.A few days ago in a paper towel manufacturing enterprise in Quanzhou City,the plastic cover-opening and pasting work for wet tissue packaging was mainly accomplished through manual coordination and teaching robots.In the production process of accessories for plastic wire rope fasteners in a hardware manufacturing enterprise in Quanzhou,the gates shearing of plastic molded parts is also done manually.Both of the above companies are facing problems such as low efficiency and high cost of artificial production.However,the use of a general-purpose target recognition system on the market still faces problems such as high costs,low identification accuracy of specific components,and low accuracy.In view of the problems faced by the above two companies,this paper analyzes the production processes and products of the two companies,and studies a large number of machine vision and image processing algorithms to design a single target recognition algorithm and a multi-target recognition algorithm.On a Windows computer with 2.95GHz Intel i5 as the hardware condition,a 800 x 600 resolution test pattern was identified.The performance of the single target recognition algorithm achieved an average angular error of 0.20 degrees,an average positioning error of 1.43 pixels,and an average time of 20.78 milliseconds.The multi-target performance achieved a recognition success rate of 100%,an average angular error of 0.08 degrees,an average positioning error of 1.37 pixels,and an average time of 571 milliseconds.This paper also completes the machine vision system applicable to actual industrial production based on two kinds of recognition algorithms.Among them,according to the production environment of the two companies to customize the system hardware,and the design of a friendly interface PC software can effectively run with the robot control system.The system performs recognition experiments on a 2.95GHz Intel i5-based Windows PC with a resolution of 756 x 580 cameras.The single-target recognition rate is 98.89%,the average time is 50.66 milliseconds.The multi-target recognition rate is 99.17%,it takes 1011 milliseconds.