Iterative DLT

INVESTIGATORS

Dr. Yisong Chen, Mr. Jiewei Sun, et al.

 

KEYWORDS:

Homography, DLT, Iterative Optimization, error analysis.

BRIEF DESCRIPTION

An iterative direct linear transformation (iDLT) algorithm is proposed to estimate the planar homography. Via iterative linear optimization we achieve reliable and accurate estimate of perspective transformation induced by given source and observed points. During iteration we progressively do uneven homogeneous coordinate adjustment and gradually construct the optimal formulation that minimizes the difference between the algebraic error and the geometric error. The algorithm overcomes the instability of previous approaches and is able to steadily converge to the global minimum of the geometric error at low computational cost despite erroneous initialization. The efficiency, robustness and precision of the algorithm are demonstrated by adequate experimental and error analysis results.

Performance comparisons between iDLT and several popular algorithms for 10 representative trials are given as follows. It is obvious that iDLT algorithm is robust to error-prone initialization and achieves fast and reliable geometric error minimization.

Here is our ICPR2010 report slides. And here is a very simple matlab demo code. Try it!

In later work we discovered that iDLT is a special case of Sampson's iterative optimization algorithm published in 1982. We are so grateful to Prof. Kenichi Kanatani for pointing out this and for his beneficial discussions

FUNDING AGENCY

N/A

PUBLICATIONS

1. Yisong Chen and Jiewei Sun, Minimizing geometric distance by iterative linear optimization, ICPR2010 .