OPTIMAL NON-LINEAR SEARCH METHOD FOR CAMERA CALIBRATION |
|---|
| Tomislav Pribanić, Mario Cifrek, Stanko Tonković |
- Abstract:
- Extracting three-dimensional information from two-dimensional image coordinates acquired by video cameras is a essential problem in computer vision. Prerequisite is a camera calibration procedure. For the purpose of camera calibration both linear and iterative techniques have been developed. The iterative methods utilize non-linear camera models, i.e. non-linear search technique. Three different popular non-linear search techniques (Newton method, Gauss-Newton method, Levenberg-Marquardt method) were chosen and compared in order to find out which one gave the best results for the purpose of camera calibration and 3D reconstruction. Levenberg-Marquardt method has been proven to be superior which was in accordance to theoretical expectation of its features. It exercised robustness and convergent sets of camera model parameters in all experimental trials. Moreover, it justified intuitive expectation of improving reconstruction accuracy by augmenting linear camera model with additional non-linear lens distortion parameters.
- Keywords:
- Levenberg-Marquardt, camera calibration
- Download:
- IMEKO-TC4-2002-159.pdf
- DOI:
- -