Abstract

It is well established that the human visual system can sense the speed and direction of motion in a moving retinal image. Computational techniques have recently been proposed which emulate this property of the visual system (e.g., Ref. ¹). We show that by using a specific pattern of connections among directionally selective motion sensors, it is possible to construct a system for extracting 3-D egomotion parameters (heading, rotation, and environmental layout) from the 2-D retinal velocity field generated by motion through a rigid environment. A 3-D motion filter can be constructed for a particular heading direction by connecting a set of 2-D motion sensors which are directed radially outward from the filter position. Any activity in the sensors is summed. When an array of such 3-D filters is used, forward translation produces a peak of activity in the filter which coincides with the direction of heading. This peak is detectable even if the eye/camera rotates during the translation. Rotational filters are constructed in a similar fashion to detect roll, pitch, and yaw.

© 1987 Optical Society of America