Abstract

Stabilization of gaze is a fundamental requirement of an active visual system for at least two reasons: i) to increase the robustness of dynamic visual measures during observer's motion; ii) to provide a reference with respect to the environment [Ballard and Brown, 1992]. The aim of this paper is to address the former issue by investigating the role of integration of visuo-inertial information in gaze stabilization. The rationale comes from observations of how the stabilization problem is solved in biological systems and experimental results based on an artificial visual system equipped with space-variant visual sensors and an inertial sensor are presented. In particular the following issues are discussed: i) the relations between eye-head geometry, xation distance and stabilization performance; ii) the computational requirements of the visuo-inertial stabilization approach compared to a visual stabilization approach; iii) the evaluation of performance of the visuo-inertial strategy in a real-time monocular stabilization task. Experiments are performed to quantitatively describe the performance of the system with respect to different choices of the principal parameters. The results show that the integrated approach is indeed valuable: it makes use of visual computational resources more efficiently, extends the range of motions or external disturbances the system can effectively deal with, and reduces system complexity.

Citations