singe camera case in correspondences

[alexlib]

In previous versions a single camera case was treated in correspondences. at least a necessary rt_is. file was written after the targets are converted to flat coordinates for the following tracking.

present version of correspondences in liboptv does not treat this case.

[alexlib]

@yosefm maybe pbi has a solution the single-camera case? I think we need to add to correspondences.c this part:

  /* -------------if only one cam and 2D--------- */ //by Beat Lüthi June 2007
  if(cpar->num_cams == 1){
	  if(res_name[0]==0){
          sprintf (res_name, "rt_is");
	  }
	 fp1 = fopen (res_name, "w");
		fprintf (fp1, "%4d\n", num[0]);
	  for (i=0; i<num[0]; i++){
          o = epi_mm_2D (geo[0][i].x,geo[0][i].y,
		      Ex[0], I[0],  G[0], *(cpar->mm), vpar,
		      &X,&Y,&Z);
          pix[0][geo[0][i].pnr].tnr=i;
		  fprintf (fp1, "%4d", i+1);
		  fprintf (fp1, " %9.3f %9.3f %9.3f", X, Y, Z);
          fprintf (fp1, " %4d", geo[0][i].pnr);
          fprintf (fp1, " %4d", -1);
          fprintf (fp1, " %4d", -1);
          fprintf (fp1, " %4d\n", -1);
	  }
	  fclose (fp1);
	  match1=num[0];
  }

and concatenate match1 with other matches of pairs, triplets and quadruplets?

from https://github.com/OpenPTV/openptv-python/blob/cc690f1fd0da7a01914fdb43f18d62a2cb9aff00/src_c/correspondences.c#L67

[yosefm]

Even in 3dptv, as I remmember, the 1-camera case is special. It is not added to other correspondences. If you do 1-cam, you don't need correspondences per se. You just build up the correspondence lists in a trivial way, as this code does. However, I highly recommend that you do not directly write to the rt_is file as this piece does. Build up the data structures and then call the writing function only in the same place that it's called for the real correspondences. That would be in the UI (e.g. Python), if anywhere.

…

On Fri, Mar 30, 2018 at 9:12 AM, Alex Liberzon ***@***.***> wrote: @yosefm <https://github.com/yosefm> maybe pbi has a solution the single-camera case? I think we need to add to correspondences.c this part: /* -------------if only one cam and 2D--------- */ //by Beat Lüthi June 2007 if(cpar->num_cams == 1){ if(res_name[0]==0){ sprintf (res_name, "rt_is"); } fp1 = fopen (res_name, "w"); fprintf (fp1, "%4d\n", num[0]); for (i=0; i<num[0]; i++){ o = epi_mm_2D (geo[0][i].x,geo[0][i].y, Ex[0], I[0], G[0], *(cpar->mm), vpar, &X,&Y,&Z); pix[0][geo[0][i].pnr].tnr=i; fprintf (fp1, "%4d", i+1); fprintf (fp1, " %9.3f %9.3f %9.3f", X, Y, Z); fprintf (fp1, " %4d", geo[0][i].pnr); fprintf (fp1, " %4d", -1); fprintf (fp1, " %4d", -1); fprintf (fp1, " %4d\n", -1); } fclose (fp1); match1=num[0]; } and concatenate match1 with other matches of pairs, triplets and quadruplets? from https://github.com/OpenPTV/openptv-python/blob/ cc690f1fd0da7a01914fdb43f18d62a2cb9aff00/src_c/correspondences.c#L67 — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#118 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAEljuPEF-p3U6AU-axznfmGV9lUbL5sks5tjczdgaJpZM4KTi6N> .

[alexlib]

i am not sure, but maybe it's better to use epipolar.pyx instead of correspondes.c and introduce it there? I could probably bind in epipolar.pyx the epi_mm_2D that exists in epi.c and use it to fill the data structures?

[alexlib]

@yosefm can you please take a look at the first attempt to get single camera solution through correspondences.pyx and epi_mm_2D binding? I'm stuck with the question on how to set up Frame 3D positions? I found in tracking_framebuf.pyx the positions () but not set_positions. What do you suggest? I'd appreciate a review of this approach. Would two if statements be Pythonic enough?

alexlib@edfcee2

[yosefm]

I'll take a look over the weekend.

…

On Tue, Apr 3, 2018 at 12:07 AM, Alex Liberzon ***@***.***> wrote: @yosefm <https://github.com/yosefm> can you please take a look at the first attempt to get single camera solution through correspondences.pyx and epi_mm_2D binding? I'm stuck with the question on how to set up Frame 3D positions? I found in tracking_framebuf.pyx the positions () but not set_positions. What do you suggest? I'd appreciate a review of this approach. Would two if statements be Pythonic enough? <alexlib@edfcee2> — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#118 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAEljr1ebcHqKzDZJJWBS2QGGfBfS2Noks5tkpMUgaJpZM4KTi6N> .

[yosefm]

The approach looks fine (apart from code quality comments I left on the commit). But I don't know where you want to set_position or put those of statements. Please give me some direction as line number, example code, or something else I can find.

[alexlib]

Left you a comment on that line. from epi_mm_2d we get pos a 3d vector. How do I fill this value into the appropriate frm structure. I want to follow your advice to fill frame and then write rt_is file from it using the standard methods. The only workaround for a single camera is this dummy correspondence and rewriting of frame structure.

[alexlib]

@yosefm please help me to proceed - creation of sorted_pos is still not the solution as I'll have to use point_position that is basically using ray_tracing that does not have a epi_mm_2D option,

https://github.com/yosefm/pbi/blob/master/ptv/sequence.py#L93

openptv/py_bind/optv/orientation.pyx

Line 70 in 7e2a62f

def point_positions(np.ndarray[ndim=3, dtype=pos_t] targets,

openptv/liboptv/src/orientation.c

Line 87 in 88ef4b7

double point_position(vec2d targets[], int num_cams, mm_np *multimed_pars,

Maybe I can use some dirty trick by using skew_midpoint twice with the same camera?

openptv/liboptv/src/orientation.c

Line 39 in 88ef4b7

double skew_midpoint(vec3d vert1, vec3d direct1, vec3d vert2, vec3d direct2,

[alexlib]

We decided to create a generalized point_positions in orientation.pyx that will include if ... else statement and separate the 3D point_positions case from 2D (single camera case). Before that we also create sorted_pos and sorted_corresp arrays in correspondences.pyx for the single camera case.
Todo:

1. in correspondences.pyx create a dummy function that creates the structures
2. create a function in orientation.pyx
3. expose epi_mm_2d
4. create single camera test