Singular Value Decomposition Aided Cubature Kalman Filter with Neural Network in Uncalibrated Binocular Stereo Visual Servoing System

Abstract

In this paper, a method for estimating the image Jacobi matrix in an uncalibrated binocular stereo visual servoing system is presented. The method is based on the singular value decomposition aided Cubature Kalman filter with neural network (NNSVDCKF). First of all, the parameter expression of image Jacobi matrix is obtained by analysis of model of eye-in-hand binocular vision. Then, in the case of the unknown camera parameters, the singular value decomposition aided Cubature Kalman filter (SVDCKF) algorithm is used to estimate the Jacobi matrix, and the neural network (NN) as the noise compensator which compensates the process noise and measurement noise. To demonstrate the effectiveness and practicality of the proposed algorithm, an uncalibrated binocular visual servoing control system based on 6DOF robot PUMA560 is established. Finally, the three algorithms of NNSVDCKF, SVDCKF, and standard Kalman filter (KF) are compared in the uncalibrated servoing system. The simulation results show that the NNSVDCKF algorithm improves the dynamic performance of the system greatly, and has robustness to the constraint in field-of-view (FOV) of the camera.