Work at SourceForge, help us to make it a better place! We have an immediate need for a Support Technician in our San Francisco or Denver office.

## Re: [Vxl-users] Problem for beginner - at least I hope

 Re: [Vxl-users] Problem for beginner - at least I hope From: Peter Vanroose - 2004-10-10 19:50:17 ```>> From what I have read so far, I need to compute a homography between planes > and 9 points is too much (4 points would be enough) but I can probably use a > least squares method to improve result whith my 9 points... Good question. In principle, this can be done by using contrib/oxl/mvl/HMatrix2DComputeRobust.h but the interface there is less transparent than what is in core/vnl/algo/vgl_h_matrix_2d_compute_linear.h The former uses any set of match point pairs, the latter uses four point matches. So probably someone should port HMatrix2DComputeRobust to core/Vnl/algo ... -- Peter. ```

 [Vxl-users] Problem for beginner - at least I hope From: Thierry Klein - 2004-10-10 15:33:35 ```Hi, I'm really new at this and this is not my field so I figured out someone could may be help. I am projecting my computer screen with a video projector to a real screen. I project a calibration grid with 9 points to screen and I can measure the position of those 9 points on the real screen (I do this with physical captors by tapping on screen targets). Then I am drawing on physical screen - I capture all coordinates with my captor system - and would like to re-draw this correctly on computer screen. >From what I have read so far, I need to compute a homography between planes and 9 points is too much (4 points would be enough) but I can probably use a least squares method to improve result whith my 9 points... That's it. I think I understood that VXL can do this (or Gandalf). But which one should I use ? I found the doc quite complex in both cases - maybe because it's not my field. Can anybody tell me which package I should use and possibly sketch me some code to compute the transformation and then compute my computer image based on real screen image ? Thanks Thierry ```
 Re: [Vxl-users] Problem for beginner - at least I hope From: Peter Vanroose - 2004-10-10 19:50:17 ```>> From what I have read so far, I need to compute a homography between planes > and 9 points is too much (4 points would be enough) but I can probably use a > least squares method to improve result whith my 9 points... Good question. In principle, this can be done by using contrib/oxl/mvl/HMatrix2DComputeRobust.h but the interface there is less transparent than what is in core/vnl/algo/vgl_h_matrix_2d_compute_linear.h The former uses any set of match point pairs, the latter uses four point matches. So probably someone should port HMatrix2DComputeRobust to core/Vnl/algo ... -- Peter. ```
 Re: [Vxl-users] Problem for beginner - at least I hope From: Amitha Perera - 2004-10-11 12:42:02 ```On Sun 10 Oct 2004, Thierry Klein wrote: > From what I have read so far, I need to compute a homography between planes > and 9 points is too much (4 points would be enough) but I can probably use a > least squares method to improve result whith my 9 points... [...] > Can anybody tell me which package I should use [...] rpl/rrel/rrel_homography2d_est.h can be used for robust homography estimation. (Least squares is an option, of course.) Have a look at the example in rpl/rrel/examples/homography2d_fit.cxx. This file estimates the same homography using multiple robust estimation techniques. If you want to simply get a least squares fit, use rrel_homography2d_est::weighted_least_squares. To convert the homography matrix into a parameter vector and back, use the member functions homog_to_params and params_to_homog. Amitha. ```