| Industrial robots are the most representative production equipment in advanced manufacturing systems,and they are increasingly being used in modern manufacturing processes to efficiently perform industrial production tasks in a variety of complex environments instead of manually.Traditional industrial robots that perform repetitive tasks according to pre-programmed procedures are no longer able to adapt to the complex and diverse demands of current production process.As vision sensors can provide industrial robots with more complete information about the environment and have the advantage of non-contact measurement,vision perception has gradually become the most important perception capability of intelligent industrial robots.Therefore,equipping industrial robots with vision system is a major means of enhancing the intelligence of industrial robots and is of significant utility in complex and diverse production tasks.This paper fully describes the structure of the vision system of industrial robots and the principle of realizing the functions of each part,analyzes the current status of research and application of vision technology for industrial robots at home and abroad,and on this basis,carries out research on the key technologies of vision systems for intelligent industrial robots,and explores the intelligent application of different types of industrial robots in combination with different vision task requirements.The main research conducted in this dissertation is as follows:1.Research on the imaging principles of industrial robot vision systems.Establishing an accurate mapping relationship between the real world and image pixels is a prerequisite and guarantee for subsequent industrial robots to perform vision tasks accurately.The projection process of world coordinates to digital image pixels in real scenes is explained by describing the transformation process between four coordinate systems,and the internal and external parameter matrices of the camera are clarified by establishing a linear imaging model.Besides,the types of lens distortions in the actual imaging process are introduced,and a non-linear model closer to the real imaging process is established by introducing distortion coefficients.The camera calibration can determine the internal and external parameters of the camera for each imaging process.The solution method and the optimization process of the camera parameters are introduced in detail through the Zhang Zhengyou calibration method,and the camera parameters of different models of industrial cameras are obtained by using Zhang Zhengyou calibration method.2.Research on the application of high-speed sorting of multiple randomly distributed targets with the parallel robot and 2D vision system,and a 2D robot vision system-based sorting experimental platform is designed and built.The mapping of target objects from the vision task space to the parallel robot operation space is completed by establishing a user coordinate system,and the image de-duplication algorithm is used to achieve the vision information acquisition of dynamic targets on the conveyor belt.Stable edge feature is selected as the basis for judging different types of the experimental target,and the traditional Canny edge detection operator and the RCF edge detection algorithm based on deep convolutional neural networks are optimized respectively to analyze and compare the performance of edge extraction.A matching model based on the distance between edge and centre-of-mass is proposed for the rotation target matching process.The result of the experiment shows that the parallel robot end can successfully grasp the dynamic target on the conveyor belt,proving the correctness of hand-eye transformation model,and the de-duplication algorithm effectively eliminates the duplicate vision information;the robot end correctly places the target object according to the fixed pose and category,which proves the accuracy of matching process and 2D pose calculation of the target.3.Research on the application of 2.5D vision system to guide the precise positioning of six-degree-of-freedom articulated industrial robots.In the context of single event effects testing of aerospace devices,an automation irradiation experiment platform is designed and built to address the requirements of single event effects testing and the shortcomings of the current test process.The 2.5D vision system is used to collect information about the irradiation position of the target,and the matching point pairs obtained from the feature matching process are statistically analyzed.Combined with the target saliency principle,a feature matching method based on the difference of regional distribution is proposed to complete the instance target search in complex background.The hand-eye relationship model between the industrial robot and the vision system is continuously optimized by giving an error weight allocation method to ensure the accuracy of the irradiation position.Finally,the automation irradiation experiment platform accurately completes the irradiation process that demonstrates the effectiveness of the proposed method.4.Research on the application of industrial robots for the grasping of spatially arbitrary positional 3D target,and an experimental platform for 3D target grasping is built with binocular vision system.The accurate grasping of spatially arbitrary positional targets by industrial robots relies on the 6-DOF pose estimation of spatial targets by the 3D vision system.3D point cloud data is generated by matched pairs of parallax image.In order to estimate the 6-DOF pose of target,a pose estimation method based on global structure feature constraints is proposed to quantify the global features which describe the target structure as constraints in the point cloud reconstruction process.The experimental result shows that robot can grasp arbitrarily placed target with the corresponding pose,which proves the feasibility of the proposed method.The application environments of industrial robots combined with 2D,2.5D and 3D vision system are constructed respectively,and the key technologies such as hand-eye calibration of different types of industrial robots,instance search under complex background and 6-DOF pose estimation are explored in this paper.The research of this paper covers a wide range of applications of robots combined with vision systems which can provide a reference for future research on intelligence of industrial robot. |
PDF Full Text Request