Research On Optimization Algorithms Of Uncalibrated Visual Servo Systems

Based on feedback control of visual information, robot visual servo control is higher intelligence and flexibility. In the visual servo system, the accurate calibration of camera model and robot model is more complicated, and the precision is easily affected by external environment, therefore uncalibrated visual servo system is becoming the current research focus. Uncalibrated visual servo method does not depend on parameters estimation precision. In a rough estimation of the parameters value, even under arbitrary parameters values, it can ensure the effective implementation of the servo task. On the basis of the visual servo control research status, this thesis compares the performance of different types of visual servo schemes, and focuses on the robot joint variables in recursive relations, involving a variety of optimization problems.In the first part, the thesis summarizes the research status of visual servo control, and points out some key problems in the field. Then it discusses the classification of different visual servo control scheme, and analyzes their characteristics. In particular, it compares the stability, robustness, sensitivity and dynamic performance in Cartesian space and image space and other aspects of position-based and image-based visual servo schemes.In the second part, for uncalibrated visual servo tracking task of moving targets, this thesis establishes an error performance function between robot end-effector and target characteristics. By Taylor approximation and minimization, it gets strict recursive relationship of the robot joint variables. The relationship contains a mapping between the joint variables of the image plane (i.e., the image Jacobian matrix), and the Hessian matrix of the target performance, which the function expansion requires.Firstly, for the joint variable recursive relationship itself, this thesis introduces the visibility constraint factor to ensure that the End-effector has been in vision, and introduces a speed cost function to ensure the smoothness of servo tracking. More rigorous extremum theory derivation is the innovation point.Secondly, for the image Jacobian matrix in the joint variables recursive relations, the secant method is adopted. A more robust affine model is established for the mapping relationship of joint variables to the image plane. Based on this model and the Broyden rank one correction rules, the recursion relation of image Jacobian matrix is obtained. Further, by introducing a forgetting factor, the convergence of the latter track is improved. Secant method application, based on affine model of the End-effector, is one of the innovation points in this thesis.Thirdly, for the Hessian matrix of the recursion relation, the global Hessian matrix inverse is estimated using the optimization algorithm, by means of dynamic BFGS method and dynamic Broyden’s family correction method. It greatly reduces computational expense of the existing literature, which considered separately residual items and local Hessian matrix inverse cases. At the same time, the singularity problem in the process of optimization is solved. Here, the dynamic BFGS method and dynamic Broyden’s family correction are the innovation points in this thesis.Finally, a three-DOF manipulator visual tracking system is built with Matlab Robotools toolbox. Visual servo optimization algorithm is simulated. The simulation results show good tracking performance of the research methods.