Date of Award

Fall 2008

Project Type

Dissertation

Program or Major

Electrical Engineering

Degree Name

Doctor of Philosophy

First Advisor

Richard A Messner

Abstract

The use of nonuniform image sensors in mobile based computer vision applications can be an effective solution when computational burden is problematic. Nonuniform image sensors are still in their infancy and as such have not been fully investigated for their unique qualities nor have they been extensively applied in practice. In this dissertation a system has been developed that can perform vision tasks in both the far field and the near field. In order to accomplish this, a new and novel image sensor system has been developed. Inspired by the biological aspects of the visual systems found in both falcons and primates, a composite multi-camera sensor was constructed. The sensor provides for expandable visual range, excellent depth of field, and produces a single compact output image based on the log-polar retinal-cortical mapping that occurs in primates. This mapping provides for scale and rotational tolerant processing which, in turn, supports the mitigation of perspective distortion found in strict Cartesian based sensor systems. Furthermore, the scale-tolerant representation of objects moving on trajectories parallel to the sensor's optical axis allows for fast acquisition and tracking of objects moving at high rates of speed. In order to investigate how effective this combination would be for object detection and tracking at both near and far field, the system was tuned for the application of vehicle detection and tracking from a moving platform. Finally, it was shown that the capturing of license plate information in an autonomous fashion could easily be accomplished from the extraction of information contained in the mapped log-polar representation space.

The novel composite log-polar deep-field image sensor opens new horizons for computer vision. This current work demonstrates features that can benefit applications beyond the high-speed vehicle tracking for drivers assistance and license plate capture. Some of the future applications envisioned include obstacle detection for high-speed trains, computer assisted aircraft landing, and computer assisted spacecraft docking.

Download

Share

COinS