Nahum Kiryati presents Adaptive Color Structured Light.
On 1998-07-22 - 10:30:00
ADAPTIVE COLOR STRUCTURED LIGHT
Nahum Kiryati
Department of Electrical Engineering
Technion - Israel Institute of Technology
Haifa 32000, Israel
nk@ee.technion.ac.il
Structured light range imaging is reliable, precise and in industrial use.
In principle, a plane of light is projected onto the scene. The 3-D position
of all visible points on the illuminated stripe is obtained by triangulation.
To obtain dense reconstruction without scanning, hundreds of light
planes are simultaneously projected. A standard technique for light plane
identification is based on Gray coding their indexes. The projection patterns
are the bit-planes of the Gray code. The scene must remain static within
the acquisition period, which is proportional to the number of projection
patterns used. Our goal is therefore to minimize the number of projection
patterns while maximizing accuracy and robustness.
A methodology for the optimal design of projection patterns is presented.
It draws on an analogy between the design of projection patterns and the
design of signals for digital communication. The design of K projection
patterns for a system with L light planes is equivalent to the placement
of L points in a K-dimensional space subject to constraints. Optimal design
in the MSE sense leads to an intractable multi-parameter global optimization
problem. Attractive suboptimal solutions are derived from families of
K-dimensional space filling curves. Significant gains with respect to the
Gray code method can be achieved by matching the coding to the noise level.
This requires the projection patterns to be redesigned per scene.
We developed a novel structured light system. The projection patterns are
adapted online to the characteristics of the scene. Color is used for light
plane labeling. Unlike previous color structured light techniques, few
assumptions on smoothness and color neutrality of the scene are necessary.
The use of color expands the code space and provides excellent robustness
and accuracy with very few projection patterns. Experimental results will
be shown.
Joint work with D. Caspi, E. Horn and J. Shamir.
Nahum Kiryati
Department of Electrical Engineering
Technion - Israel Institute of Technology
Haifa 32000, Israel
nk@ee.technion.ac.il
Structured light range imaging is reliable, precise and in industrial use.
In principle, a plane of light is projected onto the scene. The 3-D position
of all visible points on the illuminated stripe is obtained by triangulation.
To obtain dense reconstruction without scanning, hundreds of light
planes are simultaneously projected. A standard technique for light plane
identification is based on Gray coding their indexes. The projection patterns
are the bit-planes of the Gray code. The scene must remain static within
the acquisition period, which is proportional to the number of projection
patterns used. Our goal is therefore to minimize the number of projection
patterns while maximizing accuracy and robustness.
A methodology for the optimal design of projection patterns is presented.
It draws on an analogy between the design of projection patterns and the
design of signals for digital communication. The design of K projection
patterns for a system with L light planes is equivalent to the placement
of L points in a K-dimensional space subject to constraints. Optimal design
in the MSE sense leads to an intractable multi-parameter global optimization
problem. Attractive suboptimal solutions are derived from families of
K-dimensional space filling curves. Significant gains with respect to the
Gray code method can be achieved by matching the coding to the noise level.
This requires the projection patterns to be redesigned per scene.
We developed a novel structured light system. The projection patterns are
adapted online to the characteristics of the scene. Color is used for light
plane labeling. Unlike previous color structured light techniques, few
assumptions on smoothness and color neutrality of the scene are necessary.
The use of color expands the code space and provides excellent robustness
and accuracy with very few projection patterns. Experimental results will
be shown.
Joint work with D. Caspi, E. Horn and J. Shamir.