Known cameral positions

[Karlsc]

If I have known inter-camera geometry, is there some way to tell 3D Flow about this, and will that aid in image construction? Specifically what I'm trying to do is create volumetric Video. So I have cameras at fixed distances from each other that cover the scene so that no "voxel" has less than 3 cameras viewing it. I have a total of 9 cameras spread over a roughly 12m x 12m space (a sailboat), not all in the same plane.

When I do standard image import, some of the images are not recognized as overlapping because much of the surface is "glossy" and what changes or object edges.

[Andrea Alessi]

Hi Karlsc,

if you have a fixed rig you should first create a "calibration" dataset, and then you can use the same internals and externals in other projects. The best way to do so is via XMP files that can be loaded quickly.

You can find the tutorial here: https://www.3dflow.net/technology/do...flow-tutorial/

[Karlsc]

Thanks, Does that work in Zephyr Lite as well?

[Karlsc]

Looking at the directions though, I'm not sure it will work. Because what ive found previously is that even with the guaranteed camera overlap..the import photos stage ignores some of the cameras

[Andrea Alessi]

HI Karlsc,

zephyr lite does not allow the loading photos with known parameters so you can not proceed down the XMP workflow, so you will need Pro or Aerial. Let me know if you-d like a trial

[Karlsc]

The problem Andrea is that when I try the mininmal Fixed cameras (one image per camera) it FAILS to construct a 3D model. My thought was that if I could tell the constructor the relative distances of the cameras, it could then use that to better assess the overlapping points of interst and properly map them in 3D space.

In this case I had 8 cameras encompassing an overlapping 360 degree view of the object

that was why I had the question earlier of what is the minimum number of cameras and overlaps. I've attached the images so you can see what I'm working with

[Andrea Alessi]

Given the input images quality and quantity i am not entirely sure it is possible to obtain a good 3D model

8 cameras is a very small number and it's critical images are crisp enough. If you can share the dataset with me i can try processing it for you

If possible however i would try to use better camera sensors.

[Karlsc]

these are 4k images for the most part but pulled from video. I cannot upload the other 3 images for some reason I'll try again

[Karlsc]

[Karlsc]

using these same cameras in video mode with auto iextraction, I've had the software autoidentify as few as 5 images and construct at least something from that

which is why my question is about the minimum number of cameras needed

[Karlsc]

Alternatively, if the camera positions are known and fixed - that would help the math in identifying overlaps non?

[Andrea Alessi]

the general rule is that you need to see the same point you want to be reconstructed from at least 3 images, taken with a different position and rotation. Obviously, the better the images the better the result.

using different type of cameras is not going to help when there are very few images. If the camera position is know (and you need to calibrate the setup once first) you can use the XMP setup (pro/aerial only), however, if the number of camera is super-low, the subsequent results are still not going to be excellent, unfortunately.

Can you share the original input video ? I may be able to help you more if i have the original input.

[Karlsc]

>>

and you need to calibrate the setup once first

<<

how is that done in an open fixed location? I know how to calibrate a single camera but that doesn't help when they are separated by at least 5 meters.
the camera layout I have is set up so that any point has at least 4 cameras on it

there is no single original video. there are 8 separate videos each fairly heavy since it took some time to set up the cameras initially. I've given up pulling frames using the built in tool you guys have. It doesn't work very well: for example on a source I know to be 30fps I can only set to pull 25 FPS..

so I've moved to using FFMPEG and pulling a frame per frame. I know this means that the I-frames are being calculated and synthesized but when I inspect by hand I damned if I can tell which frame is a b-frame and which an I-frame

[Andrea Alessi]

If your cameras are fixed, take a good dataset and export the camera internals and externals. You'll need Pro or Aerial in order to load photos from known parameters.

I'll be happy to help if you can send me the original data you are working with.