different calibration options

[Laura Rathmair]

Hi there!

I've read on some forum posts it is possible to synchronize more than two cameras in the calibration process, given that the checkerboard is upright.

It has come to my attention that a new option might soon be available, which allows multiple cameras to be arranged in a circle, with the checkerboard being placed flat on the floor for calibration so that it is visible to all cameras. I was wondering when this update is going to be published?

Thanks in advance if you guys have any insights.

-Laura

[Matt Petrucci]

Hi Laura,

We do not currently have development plans to add this option in the web application. If you have a pipeline running locally, you could change the checkerBoardMount here to flat: https://github.com/stanfordnmbl/opencap ... in.py#L320. Note, your cameras would all need to see the checkerboard pretty easily. If you the angle is too steep, the calibration will fail.

Hope this helps,

[Laura Rathmair]

Hi Matt,
thank you for the reply- we've reprocessed a couple of sessions locally and some did work, however, serveral times the models came out faulty after rotating them 90 degrees.

For example, the model would walk fine but it was not rotated, so it walked towards the floor- other than that, the movement of the model looked fine. After rotating it, the model suddenly made weird movements with its legs. After rotating it back, the weird movements remained.
When we tried it october, the code main.py from github.com worked fine. The inconsistencies only started about a month ago. When we noticed that a new code was published, we tried with that one, but the results and weird movements stayed the same.

Is my data now lost or is there any way to retreive the original data?

Thanks in advance for your reply!

[Matt Petrucci]

Hi Laura,

Sorry just to ask a few clarification questions:

1. How many cameras were involved in the original data collection in the web app and what orientation was the checkerboard?
2. When you are trying to reprocess with a flat checkerboard, are you increasing the number of cameras that are involved?

Matt

[Laura Rathmair]

Hi Matt,

there were 4 cameras involved, the checkerboard was laying flat on the ground (we used size A2 to make it work, though the cameras were within 5m of the checkerboard). Originally, the code main.py did what we expected and the model then walked upright, but this has not been the case with the last few reprocessed sessions.

The cameras were set up as seen in the photo attached, which now I know is not ideal, as they do not all have a similar capture volume. However, the two more-or-less frontal cameras could see the subject at all times. When excluding the saggital cameras, the model would sometimes walk fine, at other times walk with a forwad lean.
This is a different problem, though. We'd like to focus on why the code rotating the model has the effect that the model turns out shaky.

[Matt Petrucci]

Hi Laura,

Thanks for the added information.

You previous question was about if the data that you had is lost before reprocessing, and just wanted to confirm that data before reprocess was collected in the web application? And if yes, was the calibration done with a flat checkerboard?

Matt

[Laura Rathmair]

Hi Matt,

yes all data was collected in the web application. The calibration was done with a flat checkerboard.
we changed the checkerboardmount to "ground" in the code and later wanted to locally reprocess the trials in order to rotate them, which worked fine initially, but was inconsistent or failed completely for the last few sessions. One of them reported an "error in processing, you may want to re-record this trial" for all trials within one session, others just messed up the model after rotating it, which is why we stopped reprocessing.

best,
Laura