A Method Of A Robot Vision System For Accurate And Fast Localization Based On Target Features

At present,Intelligence has become one of the key goals to develop robotics,and machine vision is one of the key factors to improve intelligent level.With the development of digital manufacturing technology,quite numbers of autonomous localization of robot technology are applied to the area of automatic manufacturing and digital assembly.And the robot vision positioning tech is the key and important part,which influences the position accuracy of the follow-up work and the rationality of the operation mode.Firstly,the development and application of automatic localization technology and equipment are summarized.On the basis of analyzing the robot technology and visual measuring method,a robotic automatic localization system,which integrates the technology of computer control,robotics and visual measurement,is established in order to realize automatic localization with high efficiency and good quality.Based on the mechanical structures and functions of the robotic automatic localization system,a visual servo based technical proposal of hardware design of control system is proposed.It includes hardware configuration design,control unit design and universe structural design,etc.Secondly,the Eye-in-Hand type robot vision servo system based on HOG feature kernel correlation filtering(KCF)for monocular vision target tracking,and binocular vision localization based on Blob target features,is applied in this paper.Through monocular vision recognition,target tracking and location,we guide the robot to the measurement position above the target quickly,so that the target is in the best measurement range of binocular vision.Then a binocular vision measurement system is used to locate the target accurately,so that the end of the robot is accurately positioned to the job position with the optimal attitude.Because of the relatively low position precision of the robot,the corresponding position and attitude error correction method should be adopted to realize the precise positioning of the robot,thus ensuring the position accuracy of the operation.Based on the analysis of the categories and sources of the robot's error,the compensation of robot positioning error is studied in this paper.Problems such as limited field of view or tracking target loss remain in Eye-in-Hand robotic visual controls.Thus,online measurement cannot be performed within a certain distance and a certain degree,and a closed-loop control cannot be constructed.In this paper,a visual servo combining the closed-loop and open-loop controls was proposed.a feed-forward compensation for relative linear motion was also integrated to improve the positioning accuracy,estimating the error by the data determined by actually moving the path of the robot record by the visual system.Finally,based on related experimental data,it is proved that the localization system can achieve quick target location and enhance the position accuracy effectively.The positioning precision of the robot is improved by the error iterative compensation and the online error feedback.Thereby,the average final position error is reduced from 0.6mm to 0.05 mm,increased by 90%.The average attitude positioning error reduced from 0.2° to 0.02°,increased by 90%.The feedforward compensation model is used to compensate the relative linear motion of the robot,and the average deviation is effectively reduced from 0.54 mm to 0.065 mm.The direction migration of the relative linear motion is negligible.For all groups of experiments,the maximum direction error is 0.012°,and the average is 0.006°.