Nonlinear Least Squares C/C++ Code for Carpet Evaluation

[Simon666]

Do you know of code (preferably C/C++) to do this for non linear least squares? i found a program called levmar on the internet, but it seems inadequately documented on how to use it and probably not so useful to my purposes.

My problem is evaluation of a carpet. The middle is worn out and it is laser scanned. i have a point cloud of about a couple of hundred thousand pointsand I need to model the following profile through it. Desired to find are the lengths l1, l2 l3, l4 (l5=width of carpet-l1-l2-l3-l4) and the height difference z. I have to find this using Least Squares Fitting. This seems quite non linear (unless maybe around good initial values which could be found) and the partial derivatives would also have to be determined approximatingly and numerically.

[FONT="Courier New"]-------------\                     /---------------     ^
              \-------------------/                     |  z[/FONT]
l1            l2         l3        l4         l5        |

 

[sid_galt]

Well there's galahad but it's in fortran
http://galahad.rl.ac.uk/

You can use it in C/C++ code though I believe.Also there's Ipopt for nonlinear optimization written in C++.
https://projects.coin-or.org/Ipopt

You might also want to check out Numerical Recipes in C
The book is available online (here - http://www.nrbook.com/a/bookcpdf.php) and it provides algorithms for a large number of problems including nonlinear least squares. You can't copy their code without owning the book though I believe.

 

[ravenprp]

Unfortunately, I am not familiar with any specific C/C++ code for nonlinear least squares in carpet evaluation. However, there are various libraries and packages available that offer implementation of this technique in different programming languages, including C/C++. Some popular options are GSL (GNU Scientific Library), NLopt, and Ceres Solver. These libraries offer a wide range of algorithms and functions for nonlinear least squares, and they have detailed documentation and examples to help users understand and use them effectively.

In terms of your specific problem of evaluating a worn-out carpet, it does seem like a nonlinear least squares approach would be appropriate. As you mentioned, finding good initial values for the parameters could greatly improve the convergence of the algorithm. It might be helpful to use some pre-processing techniques, such as smoothing or filtering, to clean up the point cloud before fitting the model. Additionally, there are methods for estimating partial derivatives analytically, which could also improve the accuracy and efficiency of the fitting process.

Overall, I suggest exploring the libraries mentioned above and experimenting with different algorithms and techniques to find the best fit for your specific problem. It may also be helpful to consult with a domain expert or seek assistance from a data analysis professional to guide you in your approach. Best of luck with your project!