Visual space perception during motion in depth: A computational and psychophysical approach to calibration problems

Abstract

This thesis leads to new insights into the visual perception of the 3D space during in-depth translation. Under large-field vision, this type of motion leads to apparent distortions of the visual space that we analyze theoretically and psychophysically. Our psychophysical results support the theoretical view that vision alone leads to ambiguities and errors in depth estimation in this case. However, when the motion is self-generated by the observer (rather than by the object), the 3D visual perception is improved and allows partially lifting of visual ambiguities in slant or absolute depth estimates. Finally, we discover that in-depth motion presents strong compensatory advantages, as compared to lateral motion, for the judgments of absolute distances. These results establish the need to carefully choose the type of motion in calibrating a mobile robot, depending on various factors such as the ability to stabilize motion and whether the system is equipped with non-visual information about self-motion.