Research On Object Localization In Complex Industrial Environment And Visual Servoing Technology Under Spatial Constraint

Visual servoing technology is a feedback control method for robotic arm,which could improve the flexibility and robustness of the robotic arm by introducing visual feedback information.Therefore,visual servoing technology is also known as an important method to realize the intelligentization of robotic arm.With the increasing complexity of working environment and the increasing number of application scenarios for robotic arm,higher requirements have been presented for the comprehensive performance of the visual servoing system.This paper takes eye-to-hand system as the object,and makes a research aiming at the poor positioning stability of the visual positioning system in the complex industrial environment and the spatial constraint problem in the visual servoing,to achieve the purpose of efficient and precise control of the robotic arm without calibration.The main contents of the paper are as follows:1.Improve the positioning accuracy of the vision system in a complex industrial environment.With the advent of intelligent manufacturing,the working environment of intelligent robotic arm is becoming more and more complex,a higher level of requirement is presented for the visual positioning in the visual servo system.A target location method which is suitable for complex industrial environment is researched in the first part of this paper.(1)Aiming at the failure of existing traditional positioning methods in complex industrial environments,a combination of the CCD imaging model,the attenuation model of target reflected light and the imaging model of ambient light is made to analyze the root causes of image degradation from the micro perspective,and a composite positioning method based on the YOLO-V3 and salient feature point extraction method is proposed.Firstly,the YOLO-Mobile Net model is proposed and the YOLO-Mobile Net model is used to achieve real-time acquisition of the ROI in the degraded image.Secondly,the salient feature points on the object are extracted in a small range through traditional image processing algorithms such as image morphology operations,Hough detection,and least square fitting.Finally,a binocular positioning system is built,and the pose information of the obejct in space is obtained by sparse reconstruction.(2)Aiming at the problem of the conflict between computing resources and positioning accuracy in the classic U-net model when processing high-resolution images,an object positioning method based on secondary segmentation is proposed.Firstly,the classical model is optimized by using the designed composite pooling module and complex up-sampling module,and an improved U-net is constructed.Then two Im-Unet models are built to realize the coarse positioning of the object and the precise extraction of the contour respectively.In this way,the conflict between computing resources and positioning accuracy can be effectively resolved through the repeated use of limited computing resources.Finally,a data set was established to verify the effectiveness and positioning accuracy of the target location method based on secondary segmentation.2.Research on visual servoing control method under spatial constraint.In order to make the visual servoing technology be applied to more scenes,it is necessary to deal with the constraint brought by scenes.In the second part,on the application background of assembly,this paper researches how to deal with the space constraint brought by the assembly process in the visual servoing technology.(1)Aiming at the problem of the failure of traditional trajectory planning methods under space constraint,the root cause of the problem is explained detailedly by introducing the concept of assembly nodes.In order to effectively solve the space constraint problem,a visual servoing control method based on perspective transformation is proposed.Firstly,the homography matrix is used to establish a two-dimensional virtual image plane and obtain the features of the assembly nodes in the plane.Secondly,by tracking the target feature in the virtual image plane,the robotic arm can move along the desired assembly trajectory.Finally,in order to enhance the sensitivity of the control system to the space motion of the robotic arm,three different visual features are combined into a composite visual feature,and an ADRC controller is established on this basis.The result of simulation and verification experiments shows that the visual servoing control method based on perspective transformation can solve the spatial constraint problem.(2)Aiming at the problem of lack of robustness of the docking trajectory planning method based on pose extraction,a visual servoing control method based on vanishing point vector is proposed.Firstly,a vanishing point vector that can reflect the assembly nodes is designed in the image plane and the effectiveness of the vanishing point vector is proved.Secondly,by tracking the vanishing point vector in the image plane,the robotic arm can move the spatial feature point to the assembly node without calibration.Finally,by gradually reducing the vanishing point vector modulus,the robotic arm can move the spatial feature point to the object position.On this basis,a model predictive controller is designed to further restrict the movement speed and range of the robotic arm.The Simulation and verification experiment results show that the visual servoing control method based on vanishing point vector can solve the space constraint problem.At last,combined with the research results of this paper,an assembly-based system is developed and the system's hardware and software architecture and the realization method of the core controller are explained.The feasibility and effectiveness of the theories,methods and technologies proposed in this paper are verified by carrying out assembly experiments on simulated workpieces and positioning the targets in real industrial environments.