On Image Photometric Moment-based Uncalibrated Visual Servoing System

Due to the accuracy and robustness of the Image-based uncalibrated Visual Servoing(IBVS),it has been widely used in the field of robotics in recent years.It is a key problem to obtain the Jacobian matrix to express the relationship between the visual information and the motion of manipulator in the field of the IBVS.In this thesis,the representation and calculation of image Jacobian matrix is focused,the process of image feature selection and image Jacobian matrix estimation is studied.The main work is as follows:Firstly,to avoid the rank problem of image Jacobian matrix caused by target loss,the photometric moment is used to describe the image.Based on the global photometric moment,the two-dimensional Gaussian function is used to superimpose the initial photometric moment,which reduces noise interference to image feature extraction.Aiming at the coupling problem of position control and attitude control in the coordinate system of the manipulator,the central moment of the target represents the translation motion on X and Y axis,the geometric moment represents the translation motion on Z axis,and the rotation invariant moment represents the rotation motion on Z axis.A set of image features is designed to decouple the translation and rotation components in the image Jacobian matrix.Secondly,Cubature Kalman Filtering is modified and applied to image Jacobian matrix estimation.To solve the problem that the covariance matrix is not positive definite due to error accumulation in visual servo system.The singular value decomposition of the covariance matrix is performed in the time update and measurement update,avoiding the failure of the servo caused by Cholesky decomposition of the matrix.Aiming at the environmental mutation and observation noise,the elimination factor is introduced into the prediction covariance matrix,and the prediction covariance is adjusted in real time to enhance the tracking and estimation effect of the system.An online estimator of image Jacobian matrix based on Improved CKF is designed.Finally,based on the Robotics and Vision toolbox,the simulation experiment is carried out on the Matlab experiment platform.The Improved CKF algorithm is used to estimate the Jacobian online with the image photometric moment as the observation vector,and the IBVS Simulink model based on photometric moment is established.3-DOF and 4-DOF Eye-in-hand systems are simulated respectively,and the performance of the algorithm is tested under the environment of target occlusion and environmental noise.The results demonstrate that the IBVS system based on image photometric moment designed in this thesis has strong robustness and decoupling characteristics,which can eliminate the coupling between attitude and position in a certain range and promote the development of IBVS system.