Motion Estimation from Range Image Pairs

Open Access
Natale, Donald John
Graduate Program:
Electrical Engineering
Master of Science
Document Type:
Master Thesis
Date of Defense:
November 18, 2008
Committee Members:
  • Richard G Jenkins, Thesis Advisor
  • Richard Laurence Tutwiler, Thesis Advisor
  • cartesian elevation map
  • flash LADAR
  • camera transformation
  • perspective projection
As rapidly as new sensor technologies become available to researchers, the usefulness of their integration into intelligent systems is explored. High resolution 3D video cameras are becoming available to researchers for the first time. These cameras give us the ability to measure 3D geometry at video frame rates. The goal of this work is to estimate the 6 degrees of freedom motion transformation of a moving 3D camera between range image pairs. The intended application of this ability lies in autonomous navigation, where the motion of a vehicle could potentially be computed without the use of external sensors or baselines. This would give machines the ability to navigate much in the same way humans do, using visual cues to make running estimates of their motion. Motion estimation techniques which use intensity images or stereo vision systems have not yet reached the levels of precision needed for generalized application in obstacle avoidance, crash prevention, or high speed navigation applications. Previous incarnations of 3D range cameras either did not operate at high enough frame rates, or were not of high enough resolution to be a practical solution to the problem of machine vision in autonomous navigation [2]. Starting with a formulation of a popular optic flow algorithm which had been modified by Horn and Harris for potential use with range imagery [5], a robust algorithm will be developed and tested with real data acquired with a novel high resolution 3D flash LADAR (LAser Detection And Ranging) video camera.