Visual Servo System Based on Cubature Kalman Filter and Backpropagation Neural Network

摘要

This paper attempts to enhance the estimation accuracy of image Jacobian matrix, despite the impacts from unknown statistical features of process noise and measurement noise. For this purpose, the singular value decomposition aided Cubature Kalman filter (SVDCKF) was coupled with a noise compensator for visual servo systems based on the backpropagation neural network (BPNN). The proposed algorithm improves the accuracy of the Kalman filter and compensates for the two noise covariances through the nonlinear approximation function of the BPNN. The proposed Jacobian matrix estimation algorithm was subjected to a binocular stereo visual servoing experiment on the Motoman UP6 robot experimental platform. The experimental results show that the proposed algorithm is capable of completing servo positioning.