An Adaptive Algorithm Switching System for Image Based Object Registration

Most modern day image based object registration systems typically rely on one or two algorithms to perform registration across the entire spectrum of operating conditions. While such systems may perform well in one constrained environment, such as an indoor laboratory, they fail when used in varying conditions, such as outdoors. This paper proposes an adaptive system for image based object registration in unconstrained environments. There are two main components in this system. The first is a set of image metrics such as light levels and frequency as found in a fourier transform from a scene. The second is adaptive relationships between the metrics and a set of algorithms used for object registration, such as hue thresholding, Scale Invariant Feature Transforms and Edge Detection and Segmentation algorithms.