Neural Network-Based Image Moments for Robotic Visual Servoing

This paper applies two Neural Network (NN)-based image features Zhao et al. (2012) to solve the problem of decoupling the rotational velocities around x and y axes of camera frame in robotic visual servoing systems. Based on these two image features and the other four image features used in previous work Chaumette (IEEE Trans. Robot. 20(4):713-723 2004), the interaction matrix has a maximal decoupled structure and thus the singularity of interaction matrix is avoided in Image-Based Visual Servoing (IBVS). The analytical form of depth is given by using classical geometrical primitives and image moment invariants. The IBVS Proportional Derivative (PD) controller is then designed and the stability of the controller is proved by using Lyapunov method. The tracking performance is thus enhanced for a 6 degree-of-freedom (DOF) robotic system. Experimental results on the robotic system are provided to illustrate the effectiveness of the proposed method.