Abstract

This paper describes a new architecture of image motion detection. We assume that two photodetectors are located vertically in or behind a diffusing plate on which a moving image is focused. Since the plate is dispersive, the image incident on the plate suffers spatial blurs during the propagation in it. Suppose for simplicity the image is one dimension. Let a depth-dependent OTF of the blur be H(u, z) (u: spatial frequency, z: depth). Let an incident image bef(x - vt) (x: position, v. velocity, t: time). Then, temporal spectra of the brightness observed by the detectors at x = 0 and depth z₁ and z₂ are proportional to F(-ω/v)H(-ω/v,z₁) and F(-ω/v)H(-ωv,z₂), respectively (ω: temporal frequency). By taking their ratio, we obtain H(-ω/v,z₁)/ H(-ω/v,z₂), which is actually a function of a temporal frequency ω but which can also be determined beforehand as a function of a spatial frequency u by using a ratio of the OTFs H(u,z₁)/H(u,z₂). Therefore, matching values of both the functions by varying the velocity v, we can determine an unknown motion of the image. We will describe the theory more precisely and generally. Some experimental results will be shown. A possible application as a smart pixel of an artificial retina will be discussed.

© 1992 Optical Society of America