cycling

[Reza Kakavand]

Great work!

I was wondering if you could let me know why the results for cycling are decent but not fantastic. I know you use OpenPose and HRNet for pose detection, OpenSim for kinematics and then dynamic models for kinetics. I don't understand which part is causing the inaccuracy for cycling and what could be done to get fantastic results. Thank you in advance for your response!

Regards,
Reza

Reza Kakavand, PhD student,
Department of Mechanical and Manufacturing Engineering,
University of Calgary,

[Antoine Falisse]

Hi Reza,

When is the last time you tried cycling?
We recently pushed an update and did some testing with cycling (search for cycling here: https://www.opencap.ai/news). It looked fine.

Best,
Antoine

[Reza Kakavand]

Hey Antoine
I haven't tried yet myself, but in OpenCap Q&A it was mentioned that:
"Q. Do you think the system will work on stationary cycling?
A. The current release of OpenCap was not optimized for cycling, and preliminary results are
decent, but not fantastic. We are working on improving cycling results in the future."

[Reza Kakavand]

Nice!
I didn't know about the new update. I see now, Thanks!

[Reza Kakavand]

Do we still need to validate the cycling kinematics and kinetics with marker MoCap before using it for a large scale population or it's already been done?

[Antoine Falisse]

We did not do any kinematic validation yet, and I am not aware of anyone working on that at the moment. FYI getting kinetics from cycling using OpenCap is currently not possible (since we have no information about what power is delivered).

Best,
Antoine

[Reza Kakavand]

Thanks Antoine, sounds good!

[Reza Kakavand]

Hi Antoine and Scott
We're trying to develop a stable method to capture our cycling activities.
We have used OpenCap with 2 to 5 phones and the motion we get is often not as expected. Activities like walking, squat, and running work smoothly. We will test 7 phones soon, but I don't know if that's the solution.
Would you please take a look at the sessions below and let us know if you have any suggestions?
321ecaa8-3af6-4e05-b29c-ba50a49d6ca1
1c00647d-2fbd-4f5f-946a-e49c351692c1
7ff1f291-0f9e-4d5d-91a5-4040db23b22d
ecd35018-f599-43b5-bb93-0d54c5e15861

[Reza Kakavand]

It seems the below trials with 4 cameras are stable. Cameras were positioned 40 degrees apart as discussed in the OpenCap/troubleshooting
8079104b-cc27-44aa-a460-a5e4422258d1
3c18d8aa-9f8d-4aee-92f2-a388f4443956

[Scott Uhlrich]

here's a setup that worked well for us with 3 cameras. https://app.opencap.ai/session/6cb1cdf9 ... cc86bd5449

It depends how much occlusion there is. At least 2 cameras need to see every marker at all times. I recommend using the HRNet pose estimation model as well.