| Research Objectives:
The research is aimed at developing techniques for registration of color and range images. In many remote robotic tasks involving tele-operation, operator performance is enhanced by integrating information acquired from various sensors, such as color, infrared, and laser range cameras. A major difficulty in designing such a multi-sensor system is the development of methods for combining the various data sources into one world model for examination by the operator. By registering different sensing modalities with LADAR range images, visualization techniques can be employed for sensor integration. This research investigates techniques for registration of color and range images using a stereo-based method. |
 |
Methodology and Results:
A novel stereo-based method for registering color and LADAR images has been developed. Corresponding
features in perspective-projected LADAR intensity images and images acquired
using a color camera are employed to calculate depth using triangulation, given
a rough initial guess of the relative sensor positions. The sum-squared-error of
the stereo and LADAR range values is then calculated, and a downhill simplex
algorithm is iteratively applied to refine the rotation and translation
estimates while rejecting outliers. Finally, the rotation and translation values
are used along with the LADAR range data to re-project the color image to the
viewpoint of the scanner. The registered color image that results is then used
as a texture map to create a photo-realistic, color model of the scene.
This work was conducted by Mark D. Elstrom while at IRIS lab under the supervision of P. Smith (Thesis Chair) and M. A. Abidi. This work was supported by DOE's University Research Program in Robotics under grant DOE-DE-FG02-86NE37968.
- Papers
M. D. Elstrom, P. W. Smith, and M. A. Abidi, "Stereo-Based Registration of LADAR and Color Imagery", SPIE Conference on Intelligent Robots and Computer Vision XVII: Algorithms, Techniques, and Active Vision, 3522, pp. 343-354.