Hi,

the order of my 4 Sony cameras is different in my CalibrationRecord compared to the motion takes of an actor.


How can I match the order of the cameras?

The problem is that there is no way to fit the pose. If you got it right on one cam, the actor is rotated up to 180° on another cam.

If only the camera order differs, then you may edit the scene file saved from the calibration project. The file is in XML format, so you simply reorder <Camera> nodes according to your action video and load the scene file to an action project.
If actual camera positions changed between recording calibration and action, you need to reshoot.

I swaped the camera tags and loaded the adjusted scene file. The values of the cameras updated as aspected. But the video source is still the same, which still results in the same problem that I described inside the inital post.
I added a picture showing the camera order of the calibration and the action take.

Why do you expect that video source should change?
I see that in between calibration and action video, cameras #2 and #3 are swapped. Thus, you need to:

• Process calibration video.
• Ensure the results are good. Besides looking at final status (Failed/Good/Perfect), it is recommended to look at computed camera positions from the above to quickly check whether geometry matches actual.
• Save scene file from calibration project.
• Swap camera nodes #2 and #3 in scene file.
• Load edited scene file to an action project. Now camera positions should match the video.

That's all I did, but unfortunately it doesn't work out at all.

Like I said, after swapping the nodes, the scene file seems to be correct. But everything else and of course the problem remains.

Can you provide your videos (both calibration and action) for review? Use some file-sharing service for that purpose.

Thanks for your affords.

I uploaded the calibration video, calibration project and an action record here:
https://drive.google.com/file/d/0BzzjWB ... sp=sharing

Hi,

I've checked your videos.
Calibration went OK, except that at first cameras were upside down, but after marking the ground points things came to normal. I do not know the actual camera heights, so I've chosen the scale that the lowest camera #2 is at 1.3m.
Saved scene file, swapped <Camera> nodes #2 and #3 and loaded the scene into a newly created action project. By node I mean the whole text between <Camera> opening tag and following </Camera> closing tag, as shown in the screenshot. Maybe you misunderstood that part and swapped only opening <Camera> tags? That would explain why you did not succeed.

And I can say that actor's position is consistent in all the cameras. However, there are other problems with tracking caused mainly by poor choice of actor's clothing.

• Light colors always produce more issues with shadows, so avoid them unless you have nearly ideal uniform lighting. This is especially true for shoes. In your video, almost all colors on the actress are too light.
• Clothing should differ in colors from the surroundings, otherwise the program can hardly distinguish where is the body and where is the background. Pants are a bad example in your video.
• Wear those clothes that can be properly modelled by the program. E.g. pants rather than shorts, T-shirt over long-sleeves rather than vest over half-length sleeves.

Thanks for your detailed reply.

That's all I did too.


I will try again.


I hope to get a descent setup with Kinect Ones in the future, because the setup with these cameras have been a pain in the ...