Research On Online Estimation Of Image Jacobian Matrix For Uncalibrated Robot

Robot visual servoing is the fusion of image processing, robot kinematics, robot dynamics, control theory, computer technology and real-time computation, etc, and a main branch in the frontier of computer vision. Robot visual servoing is also a fundamental subject with important theoretical research sense and extensive industrial application prospect. According to the format of vision feedback signals, robot visual servoing can be divided into two basic architectures, position-based visual servoing (PBVS) and image-based visual servoing (IBVS). Because of the structral simplicity and robustness to calibratin error of the system, image-based visual servoing has been largely applied in the real system. Image Jacobian Matrix (IJM) has always been used in IBVS as the hand-eye relation of robot, which is an important element of the system. Traditionally, calibration parameters (both interior and exterior) of camera, together with the IJM model are involved in the solving of IJM. Indeed, inevitable error, sensibility to the environment, and unfeasible implementation of calibration algorithm baffled it's use in real systems. Therefore, it's very meaningful to investigate how to gather precisely the solving of IJM in an uncalibrated environment. Aiming at the phenomenon, this thesis investigate several works, particularly listed as follows:First of all, we review now availible methods in uncalibrated solving of IJM, summarize their advantages and disadvantages respectively. Then, we introduce an relatively ideal approach, Kalman Filter based online-estimation of IJM, in detail and verify the validity of this approach according simulation and experiment.As we must linearize rough a real system and assume the system noise to be gaussian distribution, we introduce Particle Filter (PF) into IJM estimation. First, we introduce the PF from evolution aspect, and then verify the validity of PF with real linear-non-gaussian system. At last, we use PF to estimate IJM, and realize precisely tracking of robot to an mobile object, and analyse some arising problems when use PF in IJM estimation.Aiming at the non-gaussian distribution of a real system which could not be captured, we introduce Robust Information Filter (RIF) to estimate IJM. We first introduce the estimation principle and the robustness of RIF to the system noise, and then verify the advantage of RIF to KF with real linear-bounded-noise system. At last, we use RIF to estimate IJM both with simulation and experiment, and realize precisely tracking of robot to an mobile object.