Research On Image Visual Servo Based On Depth Independence

Robot vision control is a very important direction in robot research.Robots with visual model can obtain more information and have stronger interaction with the real world.The image-based visual servo system does not require camera calibration.The motion control of the manipulator can be accomplished by image information in the camera plane.The control structure is relatively simple,so it has become a research hotspot of the current visual servo system.This paper mainly focuses on the identification of image Jacobian matrix in visual servo system and the mapping relationship between image feature space and manipulator motion space.In this paper,two different control algorithms of visual servo system are proposed.These two algorithms are based on Kalman filter(KF)and neural network respectively.They can omit the image depth information needed in the visual servo system and reduce the additional cost.Based on the above research contents,the innovations of this paper are summarized as follows:(1)Aiming at the instability of disturbance error and noise statistical error of uncalibrated hand-eye system based on Kalman filter algorithm,and the slow convergence speed of the system.In this paper,a hybrid kernel online sequence extreme learning machine(MIXEDKOSELM)based on hybrid kernel and online sequence learning is proposed.The MIXEDKOSELM can correct the error in Kalman filtering algorithm,which greatly improves the performance of image-based visual servo(IBVS)control system.This visual servo control method based on KFMIXEDKOSELM-IBVS does not need camera parameters in the servo process,and is more robust to disturbance errors and noise statistical errors.In order to test the performance of MIXEDKOSELM algorithm,several UCI datasets are selected to validate MIXEDKOSELM algorithm.The experimental results show that the proposed algorithm has good data fitting ability and strong generalization ability.Then the KF-MIXEDKOSELM-IBV proposed in this paper will be validated on PUMA 560 manipulator in MATLAB simulation environment.The experimental results also show that the IBVS control algorithm has good robustness and experimental accuracy can be greatly improved.Compared with the latest algorithms,the trajectory length of the end effector obtained by the proposed algorithm is increased by about 0.02 M and the number of iterations is increased by about 20 times compared with other algorithms.(2)In order to fit image Jacobian matrix with neural network,RVFL algorithm is used to train the neural network.Before the input of RVFL,the ensemble empirical mode decomposition method is added to decompose the complex signal containing noise information into several subsignals,and then these signals are used as the input of RVFL to train the neural network.Because RVFL has been proved to have a very good fitting effect,and EEMD can play a better role in filtering noise in the signal,so the EEMD-RVFL algorithm combining the advantages of both has better performance in data fitting.Considering the advantages of EEMD-RVFL algorithm,it can greatly improve the performance of IBVS as a mapping model between image feature transformation and manipulator joint angle transformation.In EEMD-RVFL-IBVS,the inverse Jacobian matrix of the initial Jacobian matrix is replaced by the inverse Jacobian matrix of the EEMD-RVFL.This method not only takes into account the coordinates of the feature points in the camera plane,but also takes into account the errors of the image feature points.In fact,it expands the dimension of the input vector,so that the input vector has more information.The advantage of this method is that it can better fit the image Jacobian matrix.In order to verify the reliability and robustness of the proposed algorithm,the algorithm is also validated in the MATLAB simulation environment.The experimental results show that EEMD-RVFL-IBVS has better performance,better robustness and faster convergence speed.