Abstract
A moving observer can determine her direction of motion and the relative distance to surfaces from the pattern of motion on her retina. Rotations due to the observer's eye or head movements can interfere with the direct computation of heading and depth from velocity magnitudes alone. Subtraction of neighboring image velocities removes the confounding rotation information, allowing a simpler computation of both heading and depth. A physiologically based model for computing heading (Royden, 1997) performs a motion subtraction using motion-opponent operators based on physiological properties of cells in the Middle Temporal visual area (MT). This model computes heading well in the presence of rotations. Here, we test the model's ability to signal the change in relative depth between surfaces at different distances from the observer. We simulated motion toward a frontoparallel plane, 1000 cm from the observer, with a smaller wall in front at a distance of 400 cm. Simulated observer motion consisted of translation toward the scene combined with rotation of 0 or 5 deg/sec about the vertical axis. In all cases, the motion-opponent operators gave a maximum response in the locations that coincided with the positions of the edges between surfaces, irrespective of the amount of rotation. In contrast, in the rotation condition, the response magnitude of pure direction selective cells varied across the frontoparallel plane, eliminating the direct correlation between response magnitude and relative distance. When the simulated distance of the front wall was varied between 100 and 600 cm, the response of the motion-opponent operators at the depth edges was proportional to the difference between the inverse depths of the two surfaces. In simulations of two walls, the motion-opponent operators signaled the positions of all depth edges in the scene. These results suggest that motion-opponency can be used for simultaneous computation of heading and relative depth.