Research On Uncalibrated Visual Servo Control Method Of Tomato Picking Manipulator

In order to realize the visual servo control of the picking robot,this paper studies the uncalibrated visual servo control method for the 7-Do F picking manipulator of tomato string.Then the paper constructs the visual servo control system,and carries out picking experiments with the uncalibrated visual servo control method.The main research contents are as follows.(1)Kinematics modeling is carried out for the 7-Do F tomato string picking manipulator with the D-H parameter method and the forward and inverse kinematics analysis of the manipulator are conducted.The simulation results show that the pose error of the end-effector in x,y and z directions obtained by the inverse kinematics solution is[0.001,0.011,0.006]~T,which verifies the correctness of the inverse kinematics.The trajectory planning and simulation of the manipulator are carried out.The results show that,under the joint space pentomial trajectory planning method,the motion trajectory of the end-effector is smooth,and the joint angle,angular velocity and angular acceleration change smoothly and there is no mutation point.(2)The point feature is selected as the target feature and the image of target is acquired by the binocular camera located on the end-effector.First of all,the method of particle swarm optimization(PSO)SVM is used to segment tomato fruit clusters.The results show that the method of particle swarm optimization(PSO)is better than that of the traditional SVM.Secondly,a complete binary image of fruit stalk is obtained by filtering,denoising and morphological operation of the segmented image.Finally,the center of mass of fruit stalk is obtained by using the algorithm of computational geometric distance for the connected region.(3)In order to solve the problem of unknown statistical properties of system noise effect on the estimation precision of image Jacobian matrix,this paper puts forward the method of Adaptive Unscented Kalman Filter(AUKF)to estimate the image Jacobin matrix on the basis of the Kalman Filter(KF)and the Unscented Kalman Filter(UKF)method by introducting the adaptive noise statistical estimator.The simulation results show that compared with the visual servo control system constructed by the KF method and the UKF method,the distance error between the final position of the end-effector and the target is reduced by 79.74%and 80.36%respectively,and the response time is reduced by 40%and 20%respectively.(4)To solve the problems of large calculation and singularity of the image Jacobian matrix during the process of visual servo control for the picking robot,this paper puts forward the T-S fuzzy neural network algorithm to replace the image Jacobin matrix and the manipulator Jacobian matrix to reflect hand-eye relation.By collecting the datas of image feature variation and joint angle variation,the T-S fuzzy neural network is trained and tested,and used to predict the joint angles of the picking manipulator.The simulation results show that after 1.9s of the visual servo process,the 3d coordinate error between the end of the manipulator and the target tends to be stable,and the distance error value is 3.42mm.(5)A 7-Do F visual servo system is built for the picking manipulator of tomato strings,and the picking experiments are carried out.The results show that the success rate of the static picking test of the visual servo system based on AUKF method is 90%,the average time is 18s,and the average distance between the end-effector of the manipulator and the target point is 0.008m.Both BP neural network method and T-S fuzzy neural network method are applied to the system.The results showe that the success rate of static picking test of visual servo system based on T-S fuzzy neural network method is 86%,and the average time is 26s,and the average distance between the end-effector of the manipulator and the target point is0.013m.Compared with the BP neural network method,the success rate of the picking test increases by 22.86%,the average time reduces by 23.08%,and the average distance between the end-effector and the target point decreases by 59.38%.