Research On Industrial Robot Positing And Grabbing Technology And System Integration Based On RGB-D Camera

With the development of industrial technology,industrial robots are widely used in machining,spraying,assembly,handling and other fields.Vision location is the premise for the robot to achieve target adaptive crawling.Image processing technology is used to achieve segmentation and positioning of the target,and so as to guide the robot to achieve the target's autonomous crawling.Research on the vision-based target positioning and grabbing technology of industrial robots not only solves the problem of unreliable target positioning of industrial robots in traditional applications,but also ensures that the targets can be properly captured in different postures.Therefore,this thesis makes use of RGB-D camera to directly obtain the advantages of spatial 3D information.An industrial robot robot's target attitude and grab system based on RGB-D camera is designed and completed,which has important practical value and practical significance.Firstly,according to the requirements of the target attitude and positioning system,the “eye-to-hand” vision system is selected as the overall framework design of the system.For the problem of kinematics and inverse kinematics in kinematical control of the 6R robots,this thesis uses MATLAB robotic toolbox to carry on the modeling simulation analysis,provides the theory foundation for the subsequent inverse kinematics solution.At the same time,the PC software design for image processing was completed,which laid the foundation for the final system integration.Secondly,there are a lot of hollows in the original depth image information obtained by RGB-D camera,which lead to low target positioning accuracy.An improved depth image repairation algorithm for joint depth image RGB-D camera is proposed.The algorithm uses the method of hollow neighborhood variance value threshold to classify hollows into in-plane hollows and edge hollows.These two kinds of hollows are repaired using plane consistency and color similarity respectively.The depth image repair contrast experiment shows that the repairation algorithm in this paper can effectively repair hollows while maintaining the sharpness of the depth image.The target 3D positioning experiment results show that repaired depth image improves the positioning accuracy of target positioning.Furthermore,for the problem that the target is over-segmented or under-segmented at the connection with the support surface,the overall 3D point cloud of the target is segmented using a method combining the depth threshold and the vector constraint.In the segmentation of the 3D point cloud of the target primary view,an improved K-means cluster segmentation algorithm is proposed for the traditional K-means clustering algorithm with poor stability and low precision.The algorithm uses density parameter method to determine the initial center of 3D point cloud clustering.The RANSAC algorithm is used to fit the 3D point cloud of the target's primary viewing plane,and the attitude of the target is located.Through spatial three-dimensional point cloud clustering segmentation experiments,it is proved that the improved K-means cluster segmentation algorithm improves the segmentation accuracy and stability.Finally,in the VS2012 environment,the MATLAB language is used to implement the design of the upper computer image processing software and obverse and inverse kinematics calculate,and the 6R robot controller is used to complete the integration of the visual positioning and grabbing system.On this system,the target line-down experiment and goal-positioning grabbing line experiment are designed and completed.The experimental results prove the feasibility and practicability of the target positioning method based on depth image information and the design of the system plan.