In this project we are investigating the processes by which humans visually
detect changes in their environment and how to apply this knowledge to
building robot vision systems. Psychophysical experiments done with collaborators
at the Nissan Cambridge Basic Research Center have shown that people can
only detect changes in their visual field if they have allocated visual
attention to the aspect of the scene that is changing. If this attentional
allocation is disrupted, by a flash of light or a blanking of the image,
then otherwise easily seen changes become very difficult to detect. Changes
to items in the visual scene which are not attended to cannot be detected.
These experiments lead to a model of the human visual system wherein very
little direct information about the environment is stored at any given
time. This suggests that real-time robot vision systems may be achieved
by likewise retaining only sparse descriptions, and updating these descriptions
only when scene changes are detected. We are currently implementing such
a system in our lab at CIM.