An Invariant Landmark Object Recognition Method based on Adaptive Receptive Field Model of Primate's VisualCortex

There is a clear need for an effective method of invariant landmark recognition and place determination in the development of autonomous guidance of robotics vehicle. Due to the change of perspective of objects viewed from moving vehicle, invariance is a critical requirement in landmark recognition. Neurobiological research in 80s has found that the receptive fields in primate's visual cortex compensate dynamical changes of retinal images of objects by its own dynamical warping. This suggests an straightforward and efficient approach to invariant object recognition. Motivated by the computational strategy of biological vision system, CompuSensor Technology (CST) Corporation has developed a novel method compensates image scale, rotation, and translation variance by a direct affine transform of Gabor basis function controlled by a Lie group processor. We propose to develop a method of incorporating CST's scheme to an artificial intelligence system for landmark recognition and place determination with invariance regarding to rotation, scale, translation. The method of using a straightforward geometric compensation controlled by a Lie group processor overcomes shortcomings of current state of the art object recognition methods: It greatly reduces redundant template storage and matching in terms of views; It avoids the multiple response problem of the Fourier-Mellin transform method.