Primitive Parallax and Parallax Primitives

This paper describes an implementation of parallax depth perception designed to be a contributing unit to a distributed perceptual intelligence. The work was done in the context of the Cog project, a large-scale eeort to develop human or at least infant intelligence from the bottom up. Our implementation uses fast-approximation \good enough" region mapping and relevant motion detection over multiple observations to build and revise an approximate depth map. This map is intended to be an input for higher-level perception tasks such as object discrimination and attention to change. In turn, our model of parallax is also dependent on preexisting attention mechanisms that guarantee the ability to follow a target and notice independent motion.