search

JARVIS-MoCap / JARVIS-AcquisitionTool

AcquisitionTool to record multi-camera recordings for the JARVIS 3D Markerless Pose Estimation Toolbox
https://jarvis-mocap.github.io/jarvis-docs/
GNU Lesser General Public License v2.1
19 stars 3 forks source link

Missing triggerdata.csv #19

Open harshk95 opened 1 year ago

harshk95 commented 1 year ago

Hi Timo, We have been using the JARVIS Acquisition tool for some time now (6x FLIR Blackfly as we mentioned in issues in the past) and were able to use it successfully. However, with some recent recordings, we have an aberrant behaviour of the tool. It appears that the number of frames in the video does not match the number of frames from the metdata.csv and neither do either of these match the number of frames that we measure using exposure active pulses from the cameras. The main thing that appears to be predictive of this behaviour was a missing triggerdata.csv file in the recordings that had that and they all are from a single day of recordings. The other strange thing we have not observed before is that in the metadata file, the cameras do not start from frame 0, rather they all start from a different number and all end at a different number (using frame ID). Running it again now does not reproduce this issue. We have attached below an example metadata file. We were wondering if given this behavior you could help us diagnose the problem, help us avoid it in the future and if there was some possibility to salvage the current recordings.

metadata.csv

Thanks a lot.

padok commented 1 year ago

Hey there,

I'm not Timo, but maybe I can help you out. Right now, it's normal that the metadata file ends a bit earlier than the cameras, this is caused by the way of how we sync them up at the moment (not quite optimally). Sometimes the metadata file closes before the cameras are done recording frames, and this causes a small misalignment between the metadata file frame numbers and the video files. Check the frame count of the video files to make sure that they are complete. We know about this and we're working on fixing it.

Your file is pretty weird, though. The start or first frames normally should be in sync, but it looks like the cameras got some noise or something triggered them randomly before you started the acquisition in Jarvis. It could be a problem with your wiring or your microcontroller or your storage medium. Are you using a SSD? It is required, because recording with multiple cameras is causing quite an amount of random writes and HDD are not up to that task.

Let me know if you can reproduce it again.

Louis

Antorithms commented 1 year ago

Hi Louis, In our experience so far, we do not experience that the metadata file ends before the cameras and is quite precise. We have not been able to reprodice the glitch that we reported above and it seems to have affected recordings done on that particular day. It does appear there is some degree of randomness in the triggering and the number of frames in each camera are way off and do not seem to be in a manner that we can find the spurious frames from the metadata file. We use M2 SSDs for acquisition and never had this trouble in many recordings in the past or since we reported that issue.

Thanks.

padok commented 1 year ago

Hey Antorithms,

What do you mean by randomness? Is you setup working as expected? If you experience frame drops, and you are sure it's not related to your hard disk:

  1. What USB-Card are you using? Do you use a one USB-Controller per camera configuration?
  2. Are you useing different USB cables than the ones supplied by Flir?
  3. Have you set the recommended USB configuration (linux)?
  4. Did you verify your camera configurations?

Cheers Louis