Messages

1

General / Re: [SOLVED] Project points of doll model into original frame

« on: April 22, 2015, 10:51:32 AM »

Dear William,

thank you very much! That was actually the problem, I was considering the standard definition of the projection matrix. Is it written anywhere in the manual that defined the inverse of P?
However, thank you again very much!

Best regards.

P.S. the last page of the 'Agisoft PhotoScan User Manual Standard Edition, Version 1.1' reports the formulas to reproject the 3D points into the image frame.

2

General / Re: Project points of doll model into original frame

« on: April 21, 2015, 04:01:24 PM »

My intent is to manipulate a 3d mesh created with photoscan using MATLAB.
As explained I exported the 3D points of the .obj file and the extrinsic and intrinsic paramters in the .xml file.
I read them manually to avoid a possible error in a script.
I use the formulas reported in photscan manual to reproject the 3D points into the original 2D images used to compute the model, but the reprojection is wrong.

3

General / Re: Howto get world-2-pixel coordinate correctly

« on: April 20, 2015, 12:03:00 PM »

Hi, it seems I am not the only one experiencing this issue...
I've done a ton of tests and checked different file formats, but I could not find a solution. I even tried to use the .fbx file to export both data points and parameters but got no solution...

Here you find another topic I opened:
http://www.agisoft.com/forum/index.php?topic=3646.msg19074#msg19074

In my humble opinion the problem is in how photoscan exports external parameters or eventually 3D points...

4

General / [SOLVED] Project points of doll model into original frame

« on: April 10, 2015, 11:20:41 AM »

I downloaded the data of the doll tutorial, with .obj file and input images included.
I've exported from the photoscan project file the cameras photoscan file .xml.  Using the parameters included in it, I've tried to reproject the points of the mesh into the first image, thus I've copied the intrinsic and extrinsic parameters for that camera and apply the projection function provided in the manual (and the one provided in Caltech vision calibration toolbox) but the the points projection is completely wrong!
 
I think that either there is a bug in Photoscan or the cameras and obj files refers to different world coordinates.
Could someone help me please.

5

General / Re: Explanation of exported camera transformation matrices.

« on: April 09, 2015, 10:14:39 AM »

The transformation should be the [R|t] matrix, indeed the last row is [0 0 0 1] but trying to reproject the points using this transformation they do not match...

EDIT: Multiplying the 3D points by -1 the reprojection it's much better but points now mismatch by a translation offset...

6

General / Re: World coordinates to pixel coordinates with exported parameters

« on: April 09, 2015, 09:56:28 AM »

Hi, I am also having troubles trying to reproject the 3D points of the mesh into the original frames.
Could you explain me how you import the parameters?
Did you just use the R matrix and t vector from the .xml file? Did it work?
I extracted the parameters from .xml file, and apply the procedure to reproject the points that is explained in the manual, but it doesn't work, I've checked every single step, and cannot figure out where may be the problem. But I noticed that the R matrix is different in phi-omega-kappa .txt and photoscan .xml file....

7

General / Re: Field explanation of photoscan .xml camera file

« on: April 08, 2015, 11:33:18 PM »

I give a look at the parameters in the fbx file, the coordinates ot the points of the 3D mesh are the same as in the .obj file, the translation vector also, but the not rotation matrix (or better the angles that define the rotation matrix).

No one ever tried to project a computed 3d mesh into the original frames?

8

General / Re: Field explanation of photoscan .xml camera file

« on: April 07, 2015, 02:47:03 PM »

Therefore the exported coordinates in the .obj file are already in the world frame coordinates, right?

Because I am trying to project the 3D mesh in a corresponding input image of the model using the P = K[R|t] matrix extracted from the camera .xml file but it doesn't project the coordinates correctly.
I've tried to do this test with the data of the doll of the first tutorial...
I even notice that the Q = [R | t] matrices from the photoscan camera .xml and phi kappa omega .txt are different...
Is there a bug in Photoscan or I am missing something?

9

General / Field explanation of photoscan .xml camera file

« on: April 07, 2015, 12:53:46 PM »

What do the fields embedded in the tag <regoin> </region> refer to?
Are the coordinates of the 3d .obj defined in the region space? Should I therefore translate the 3d coordinates according to it?

Who can help me?
Thank you in advance.

10

Bug Reports / Re: Errors exporting the data

« on: April 07, 2015, 11:06:48 AM »

Does anyone ever tried to build a 3D model with photoscan and do a sanity-check projecting the 3D points in the oringinal frames? Did it work? How you exported the data?

EDIT: (SOLVED?) Using the parameters exported in the .xml file and multiplying the 3D points by -1, it makes the reprojection better but points still do not match, they are translated by an off-set...

EDIT: (SOLVED!) in another post the user William reported me that photoscan exports the inverse of the standard notation of the matrix P, i.e. the martix maps the coordinates from the camera frame to the world frame and not vice versa, as usual.
Thus, inverting the matrix P actually solved the problem!

11

Bug Reports / [SOLVED] Errors exporting the data

« on: April 03, 2015, 03:40:18 PM »

I am using Photoscan 1.1.4 64bit for windows.
I need to import the 3d model computed by photoscan into matlab. The model has been computed using pre-calibrated fix parameters.
I exported the model as an .obj file and the extrinsic parameters either in the format phi kappa omega .txt OR photoscan .xml. The extrinsic parameters in this two models differ by a rotation matrix, for instance, the rotation matrix exported in the .xml file is the R matrix of the .txt file rotated by 180° around the x axis.

Moreover, I get a "correct" reprojection of the points only when I use the phi kappa omega .txt file to obtain the parameters, and I multiply the x coordinate of the translation vector by -1 and invert the x-axis on the frame i.e. u = width - u. Or, in other words, when  I multiply the x coordinate of the translation vector and the x focal length by -1.

To reproject the 3D-points I followed the schema reported in the manual and I even tried to apply the function developed into the calibration toolbox from Caltech. The reprojection using these two methods are identical, but they are correct only when I apply that sort of patch.

EDIT: I checked with another model exported from the same set of photos and configuration, this time neither that patch works... This model is smaller and it is not centered in the z axis as the previous one.


Thank you in advance.