A Memo on How to Use the Levenberg-Marquardt Algorithm
for Refining Camera Calibration Parameters
Pradit Mittrapiyanuruk
Robot Vision Laboratory, Purdue University, West Lafayette, IN, USA
mitrapiy@purdue.edu
This short note briefly explains how to implement the Levenberg-Marquardt (LM)
algorithm in MATLAB for the minimization of the geometric distance given by Equation (10) of the Zhang’s report. For students who are not familiar with the LM method, we suggest you visit the Wikipedia page
For a more elaborate discussion of the LM method, the following references may be useful
- Appendix 6 in the Hartley & Zisserman’s Book
- Section 9.4 in the Chong & Zak’s Book (the official textbook for ECE580)
- K. Madsen et al, “Methods for non-linear least squares problems”, IMM Lecture Note 2004, available on the web at
In the rest of this note, I will first demonstrate how to use the LM method to solve a simple curve fitting problem and then I will give some clues that may help you to apply the LM algorithm for the refinement of the camera calibration parameters in Homework 5.
Example: Curve fitting with the LM algorithm
Problem: Fit a set of point data {xi,yi} (xi=0, 0.1, …, 2*PI) with the function*
f(x) = a*cos(bx)+b*sin(ax)
where a and b are the parameters that we want to estimate by the LM method. The problem of finding the optimum can be stated as follows
argmindTd
a,b
di=[yi f(xi;a,b)]
whered=[d1d2......dn]Tand
*
This example, taken from Wikipedia, is interesting because the complexity of the fitted function would preclude a linear least square solution to the problem.
The Levenberg-Marquardt (LM) algorithm is the most widely used optimization algorithm. ItThe Levenberg-Marquardt (LM) algorithm is the most widely used ...
The Levenberg-Marquardt (LM) algorithm is the most widely used optimization algorithm. ItThe Levenberg-Marquardt (LM) algorithm is the most widely used ...
热门文档