Abstract

A novel implementation of a hybrid optical/electronic recognization system is described. The main concept of the system is an optical mapping of the Fourier transform of an input scene into logpolar coordinates. In the mapped plane both positions, rotations, and sizes are converted into shifts by logpolar mapping. By adding the intensity of the mapped image in two directions, two 1-D signals appear representing an angular and radial image representation of the original image. The 1-D signals are obtained with a cylindrical lens having destroyed the coherence of the mapped image with a rotating ground glass disk. The 1-D signal is detected with a 1024-point linear CCD array and fed to an electronic correlator where the recognition, orientation, and position measurements take place. The advantage of the electronic correlator vs an optical correlator is mainly flexibility. Optimal filter design for a collection of objects to be recognized can relatively easily be done. The performance of the presented system with respect to speed, capacity, and accuracy depends on the data for the input/output devices and on the speed of the electronic correlator. With the technology today it is estimated that one object out of 50–100 possible objects can be recognized with an angular accuracy of <5° and a relative positional accuracy of 5–10% within 0.5 s.

© 1988 Optical Society of America