Epipolar Geometry for Humanoid Robotic Heads

Stereoscopic vision is a capability that supports the ability of robots to interact with visually complex environments. Epipolar geometry captures the projective relationship between the cameras in a stereo vision system, assisting in the reconstruction of three-dimensional information. However, a basic problem arises for robots with active vision systems whose cameras move with respect to each other: the epipolar geometry changes with this motion. Such problems are especially noticeable in work with humanoid robots, whose cameras move in order to emulate human gaze behavior. We develop an epipolar kinematic model that solves this problem by building a kinematic model based on the optical properties of a stereo vision system. We show how such a model can be used in order to update the epipolar geometry for the head of a humanoid robot.