Noise-Robust Image Jacobian Estimation for Visual Servoing Using BP Neural Network and Robust Kalman Filtering

Accurate estimation of the Jacobian matrix in visual servoing systems is crucial, but it is susceptible to interference from factors such as noise. This study proposes an image Jacobian matrix optimization method that combines BP neural network and Robust Kalman filter. First, use BP neural network to model the coordinates of image feature points and output preliminary estimates, and then optimize them through Robust Kalman filtering. Through simulation experiments, it was found that compared to using Robust Kalman Filter alone, this combined method makes the system's linear velocity and angular velocity more stable, reduces image feature errors, significantly improves the stability and performance of visual servoing systems in noisy environments, and provides an effective solution for the robustness of visual servoing systems in practical applications.