<title>Rapid target tracking algorithm using Gabor representation of target signature and Lie derivatives</title>

Abstract

Conventional approach to object tracking through image sequence either use area correlation or use extraction of contrast edges and other features from the target image, taken as the target signature. However it is often difficult to reliably extract the target signature when the target and background appearance, as well as target/background polarity and contrast, change drastically over the course of the image sequence, and to maintain high accuracy of track in real-time. One of the major neurobiological discoveries in the last two decades is that the what processes, which determine the identity of an object, are segregated from the where processes, which determine the spatial location and motion of an object. Biological visual motion tracking does not require continuously detecting the target signature. It simply maintains the target signature throughout the image sequence via an adaptive process of the receptive fields of neurons. Formulated as a pure where process that maintains the Gabor representation of the target surface signature, we reduce the tracking process to the analytical computation of affine transformations of the surface signature through the image sequence. The great simplicity, high accuracy, and robustness of this analytical pure where process demonstrated the power of the biological computation strategy.