Research On Sorting And Automatic Clamping Of Disordered Grafting Clamps

During the process of grafting in agriculture,watermelons and eggplants require one grafting clip per plant,and the usage of grafting clips is very high.In modern automated grafting machines,the sorting of grafting clips still require rigid vibrating discs or manual participation,and the overall automation level of grafting machines needs to be improved.To address this issue,a new grafting clip with a tubular structure for automated grafting has been designed.The position and posture of the grafting clip in 3D space is analyzed and calculated through 2D image analysis and 3D point cloud modeling and analysis.Based on the posture parameters of the grafting clip,the mechanical arm is controlled to grasp the grafting clip,achieving automatic grasping of scattered and unordered stacked grafting clips,which is beneficial for promoting flexible automation of grafting machines and improving the overall automation level of grafting machines.The research work in this paper focuses on the following areas:(1)Construction and calibration of visual acquisition system.A grafting clip with a tubular structure suitable for automation scenarios was designed.An image acquisition system consisting of a line laser and an RGB camera based on linear ball guide rails was developed,which can perform two-dimensional image acquisition and three-dimensional point cloud modeling.The camera parameters of the system were calibrated using the Zhang Zhengyou method,and the light plane of the line laser was calibrated using the RANSAC method and the least squares method.Experimental results showed that the average error of the repeat positioning of the guide rail slider was less than 0.016 mm,with a mean square deviation of less than 0.039 mm.The average sliding speed of the slider was 27.7cm/s,with a mean square deviation of 1.03cm/s,meeting the requirements for guide rail precision and speed.The reprojection error of camera calibration was less than 0.185 pixels,and the calibration error of the laser plane was less than 0.27 mm,meeting the calibration accuracy requirements.(2)Two-dimensional image analysis of grafting clip.In the material platform,the grafting clip is mostly in a scattered and independent state,and in a few cases,it is in a disordered heap state.In order to quickly calculate the position and orientation angle of the grafting clip in a scattered and independent state,two-dimensional images of the material platform were first captured and analyzed.Under the backlight environment,the contour of the grafting clip was extracted,and the Hu moment matching values of the contour were used to classify the grafting clip in the scattered and independent state.Based on the angle characteristics of the grafting clip contour,the pixel set of the grafting point was analyzed,and the non-maximum suppression method was used to obtain the pixel coordinates of the grafting point.The orientation angle of the grafting clip was calculated based on the cross-product relationship between the centroid of the grafting clip contour and the vector composed of the grafting point.Experimental results showed that compared with manual labeling,the average pixel positioning error of the grafting point in a scattered and independent state was 2.17 pixels,with a mean square deviation of 1.05 pixels,and the average error of the orientation angle of the grafting clip was 0.96°,with a mean square deviation of 0.62°,meeting the accuracy requirements.(3)Point cloud modeling and analysis of disordered grafting clamps.A method based on connected domains for extracting laser centers is proposed.The clamp is modeled based on the line-plane principle.The point cloud of the platform is segmented out by using a hight threshold segmentation method.Based on the normal vector features of the grafting clamp point cloud,a conditional Euclidean clustering method is used to segment multiple independent grafting clamp point clouds.The FPFH descriptor is combined with the RANSAC method to register the segmented grafting clamp point clouds with the template grafting clamp point cloud.The poses of multiple disordered stacked grafting clamps in the point cloud are calculated based on the rotation matrix and translation vector obtained by point cloud registration.Experimental results show that the average deviation of point cloud registration is 0.587 mm,and the mean square deviation is 0.352 mm,which meets the accuracy requirements.(4)Design of the gripper end and control of the robotic arm movement.A new gripper end effector based on a screw slider structure was designed,and both forward and inverse kinematics analysis were performed on a 6R arm.Grafting clip grabbing and releasing experiments were conducted.The experimental results showed that the success rate of grabbing scattered and independent grafting clips was 89.0%,and the success rate of grabbing unordered stacked grafting clips was 80.5%.Since the releasing operation is less complex,the releasing was successful in all 127 experiments.