Project Title:

Log-Hough Transform Feature Detector

09.05-8988B
901695
Log-Hough Transform Feature Detector

Abstract:

Prototype hardware will be designed for executing the log-Hough transform on imagery
at video frame rates. This transform, recently developed by the principal investigator,
is three orders of magnitude more efficient than the traditional Cartesian Hough
transform. The Hough transform is a powerful technique to identify robustly extended
linear features despite the presence of gaps and noise. Examples range from Space
Station structural elements to orbital debris tracks. Space applications include
autonomous, visual docking and robotic constructions. Ground support applications
include high-speed, automatic recognition of orbital debris tracks in star field
imagery. Preliminary feasibility studies indicate that orbital debris processing
can be achieved at rates of 33 ms per image using hardware hosted in the backplane
of a single IBM PC/AT compatible computer. NASA has applied log-polar coordinate
mapping in the design of efficient visual tracking and docking systems. The addition
of the log-Hough transform to these systems enhances their capabilities to much broader
applications in robotic vision. Linear features are a cornerstone of pattern recognition.
The ability to recognize them quickly in compact hardware is a powerful attribute
which will benefit robot vision systems generally.
Application could occur in robotic vision for manipulation, assembly, docking, navigation,
and processing satellite imagery (e.g., road finding).
robot vision, Hough transform, log-polar sensors, polar exponential grids, remappers