3D Point Cloud or Mesh Data

[Francis_P_]

i'm looking for a solution for capturing faces and i don't know if it's possible with realsense.

my requirements are follow:  

• mesh resolution <1mm
• processing can be up to 30 sec
• usage of multiple cameras prefered 

i have seen that there some projects for creating point clouds but i don't know how the resolution is. are there some more information's?

[samontab]

What I'm going to say applies to the RealSense camera as well as any similar device that uses projected IR patterns (such as the Kinect v1 for example).

- The mesh resolution will vary depending on how far away your object is. It is possible to get a rough estimate on the maximum resolution of the RealSense given it's specs, but I haven't done those calculations so far...

- 30 sec seems long enough to create a full mesh, but there will be a trade off between resolution and time of processing.

- Keeping a constant stream of multiple cameras would only work if you are viewing different areas on each one, but for stitching the views together into a single mesh you would need to have some overlap, which will cause interferences (you are projecting different IR patterns into the same area). You could turn the cameras on and off to eliminate the interferences, but it may get tricky to sync all of it.

[Francis_P_]

thank you for the information.

well the distance will be in the 20-30cm range so i hope that the resolution could be below <1mm.

so when there ist the possability to capture with multible realsense camera, i must have the controll of the pattern generation and image captureing. because i need only the front face (ear to ear) my target would be 3 realsens cameras. so i should get a good overlap. 

what i dont know is how much 3d points do i get per fps. my target would be 200k - 500k 3d points per realsense camera. 

well must see what the future brings :-)

[samontab]

Well, the absolute maximum number of 3D points you can get out of the RealSense in theory is 640 x 480 x 60 = 18,432,000 3D points per second.

In reality though, the number will be lower than that since depth information will not be available for many of those points. Also, you may want to sync it with the RGB camera, which only runs at 30Hz... In any way, I think you should be fine with those requirements...

[Francis_P_]

thanks a lot you for the information.... well so in theory it should work :-) 

it's time for the sdk....