Robust gaze-steering of an active vision system under uncertainties in estimated parameters

Gaze-steering is often used to broaden viewing range of an active vision system. Gaze-steering procedures are usually based on estimated parameters such as image positions, image velocities, depths, camera calibration parameters or etc. However there might be uncertainties in these estimated parameters because of measurement noises, estimation errors or etc. In this case, robust gaze-steering cannot be guaranteed. To compensate for such problem, this paper proposes a gaze-steering method based on LMI (Linear Matrix Inequality). In this proposed method, we first propose a PD (Proportional Derivative) control scheme on the unit sphere that does not use depth parameters. This proposed PD control scheme can avoid uncertainties in the estimated depths and camera calibration parameters, and inconveniences in their estimation process such as use of auxiliary feature points and high computational cost in their nonlinear estimation process. And then the control gain of the proposed PD control scheme on the unit sphere is designed using LMI such that the designed control is robust against uncertainties in the other estimated parameters such as image position, image velocities or etc. Simulation results show the proposed method compensates for uncertainties in the estimated parameters better than a contemporary linear method, and more steadily with time than a contemporary nonlinear method. The proposed method takes less computational cost than the contemporary nonlinear method. Though the proposed method takes more computational cost than the linear method, it is fast enough for real-time processing.