Abstract

L. Itti, Automatic Eye and Head Animation for Animats, From Animals to Animats 8, Proceedings of the Eighth International Conference on the Simulation of Autonomous Behavior, Santa Monica, California, Jul 2004.

Abstract: We propose a computational model for the automatic animation of the eyes and head of virtual or physical avatars. Given any input in the form of video streams, the model finds the most salient (interesting) locations in the agent's visual environment and directs its gaze towards them. The computation of visual salience at the basis of the model relies on a neurobiological model of visual processing along the occipito-parietal stream in the primate brain. The relative contributions of eyes and head towards a given gaze shift are then computed from a gaze decomposition model derived from behavioral recordings in Rhesus monkeys. Finally, the dynamics of eye and head movements are calibrated against behavioral recordings from human subjects. The model autonomously gazes towards locations also gazed to by human observers watching the same video inputs, in a highly significant manner.

Note: Plenary lecture

Themes: Computer Vision, Model of Bottom-Up Saliency-Based Visual Attention, Computational Modeling