Data Collection and Processing Issues

[Josephine Taylor]

Hi!

I am trying to collect overground walking trials of the subject at self-selected, slow and fast speeds. I am using a 3 camera set-up. The model created appears to be very incorrect when walking! The videos do not appear to be synced with each other.

Looking closer at the videos, the subject begins out of the frame for 1 of 3 cameras. The subject begins walking forward, entering the frame of all cameras. Upon reaching the end of the room, the subject turns around out of the frame of all cameras and walks back within frame of all cameras.

Do you have any suggestions on fixing the issue with synching the cameras? If possible, I would like to keep the same camera set-up.
Could this issue be fixed by reprocessing the data with only 2 cameras? If so, how can this be done?
Is it possible to trim the trials to only include the section the subject is within the frame of all three cameras?

Additionally, when the subject is walking at faster speeds, the trials look even worse. What can I do to improve these?

Thanks for your help.

[Matt Petrucci]

Hi Josephine,

Yes it sounds like you are having a camera sync issue.

See our best practices and the video for setting up overground gait. You can also take a look at the tutorial video for some tips for a walking setup.

But some key points are:
1. Make sure the person is clearly seen by 2 cameras through the entire trial.
2. Make sure the person enters and exists the FOV at the same time.
3. If that doesn't help, have the person do a hand punch in the air when all cameras can see them (this will definitely help if you want to keep a 3 camera setup).

For data you have already collected, you could consider removing the front facing camera (or shorten the videos) and reprocessing. To do this, you would need to re-engineer this code and setup the opencap-core environment on your computer.

Hope that helps,
Matt

[Josephine Taylor]

Hi Matt,

Thanks for your help!

How would you go about shortening the videos? Can I download the videos, trim, and then reprocess? Or, is there another way to go about this?

Also, I am working on reengineering the code you linked. I keep running into issues in the code below, specifically there is an error finding . I think it is because there is no extrinsic file when I download the data. Is this part necessary? Where can I find the extrinsic file?

for trial in trials_tmp:
if 'extrinsics' in trial.lower():
extrinsics_idx = trials_tmp.index(trial)
if 'static' in trial.lower():
static_idx = trials_tmp.index(trial)
session_with_static = True
trials = [trials_tmp[extrinsics_idx]]
if session_with_static:
trials.append(trials_tmp[static_idx])
for trial in trials_tmp:
if ('static' not in trial.lower() and
'extrinsics' not in trial.lower()):
trials.append(trial)
else:
for trial in trials_tmp:
if 'extrinsics' not in trial.lower():
trials.append(trial)

Thanks,
Josephine

[Matt Petrucci]

Hi Josephine,

Yes you would need to download and trim the videos to reprocess.

There should be a cameraIntrinsicsExtrinsics.pickle file for each camera within a session. How are you downloading the data?

[Josephine Taylor]

Hi Matt,

I am downloading the data from the web application by clocking the "DOWNLOAD DATA" button.

The error I keep getting is that the"extrinsics_idx is not defined"

Thanks,
Josephine

[Antoine Falisse]

try reprocessing the calibration file as well: https://github.com/stanfordnmbl/opencap ... ons.py#L69