Related Experiment Videos
Algorithms for matching 3D line sets
Behzad Kamgar-Parsi1, Behrooz Kamgar-Parsi
1Office of Naval Research, 800 N. Quincy St., Arlington, VA 22217, USA. kamgarb@onr.navy.mil
IEEE Transactions on Pattern Analysis and Machine Intelligence
|October 6, 2004
Summary
This study introduces novel algorithms for matching 3D line sets, addressing previously unsolved cases in computer vision. These methods offer exact solutions, proving convergence and coordinate transform invariance for enhanced scene registration and object recognition.
Area of Science:
- Computer Vision
- Computational Geometry
- 3D Reconstruction
Background:
- Matching sets of lines is fundamental in computer vision for tasks like scene registration and motion estimation.
- Existing algorithms for 3D line set matching have limitations, including handling only specific cases, providing approximate solutions, or lacking coordinate transform invariance.
Purpose of the Study:
- To develop exact algorithms for matching 3D line sets across three cases: finite-finite, finite-infinite, and infinite-infinite.
- To address the unaddressed finite-infinite case and improve upon existing methods for the other two cases.
Main Methods:
- Development of novel algorithms for general 3D line set matching.
- Ensuring algorithms are provably convergent and invariant to coordinate system transformations.
Main Results:
- Exact solutions are provided for all three cases of 3D line set matching (finite-finite, finite-infinite, infinite-infinite).
- The new algorithms demonstrate convergence and invariance to coordinate transforms.
- Experimental validation using synthetic and real 3D image data.
Conclusions:
- The presented algorithms offer a comprehensive and robust solution for 3D line set matching.
- These advancements are expected to improve performance in various computer vision applications.