If OpenSim suitable for IMU motion capture enhancement?

[Yury G.]

Hi there!

I'm working on an IMU-based motion capture system for rehabilitation and investigating whether I can use Simbody/OpenSim to improve current system. Current system is basically a mapping of 10 IMU quaternion orientations directly to a very simple human model in Unity. It suffers from usual IMU problems like gyro drift, magnetic fields, etc., wich sometimes cause Unity skeleton to be invalid (bones may look broken) and whole model looks clunky.

The idea is to build biologically more valid skeletal model of human body (~20 bones and joints) without muscles (maybe add the muscles at some point) and map IMU quaternion data to this model. I want to add constraints on joints, maybe derive some optimization criteria for the whole skeleton or anything else, that may enhance current solution =)

The data-flow can be something like: I have a stream of IMU data, then it "goes through" the opensim/simbody skeletal model (getting enhanced meanwhile), and then I stream it further, i.e. into the Unity.

After searching the forum it looks like OpenSim currently doesn't support IMU sensors out of the box, but it's possible to work with `OSensors` using Simbody api (https://github.com/opensim-org/opensim- ... matics.cpp). So basically the questions are:

1) Can I use opensim to enhance sometimes really bad IMU data by applying model constraints and maybe by some other optimization techniques? Or is it a bad idea, and OpenSim is not suitable for such a task?

2) Is it ok to have, let's say, 10 IMU orientation sensors and fit them to the model with not 10, but more body segments, like 20? Will other 10 body segments, which are not "welded" to IMU sensors, be adjusted by IK as well in order to fit the whole body?

Thanks for your help in advance and thanks for the cool open-source project you're doing!

[jimmy d]

Hi, Yuri-

We have been working on new methods for IMU tracking in OpenSim and are close to recruiting Beta testers-- something it sounds like you would be interested in.

You can sign up for email announcements here. We should be sending out a call for Beta testers in a week or so.

Cheers,
-J

[Yury G.]

Thanks for a prompt reply, James!
Indeed it does sounds interesting - signed up for the announcements!

And going back to my 2 original questions, does it sounds reasonable at all?

[jimmy d]

1) Can I use opensim to enhance sometimes really bad IMU data by applying model constraints and maybe by some other optimization techniques? Or is it a bad idea, and OpenSim is not suitable for such a task?

That is an open question. I think you can improve estimates when you include biological joint constraints, but if you have 'really bad IMU data' I doubt modeling will help. Modeling and simulation rely on high-quality data as inputs, whether it is marker data or IMU data. You should never rely on modeling to fix poor data collection.

2) Is it ok to have, let's say, 10 IMU orientation sensors and fit them to the model with not 10, but more body segments, like 20? Will other 10 body segments, which are not "welded" to IMU sensors, be adjusted by IK as well in order to fit the whole body?

You need one IMU per modeled body to track.

[Hans-Jochen Edenhart]

Hi James and Yuri.
I think i am just working on a very similar project. i am trying to simulate gait by an IMU sensor System and measure the GRFs with a preasure walk way.
a former project simulated a knee bend with IMUs but with the help of a kinect camera (IR).
My task would now be to do this just by IMUs with keeping the fault as small as possible.
So i would like to sign me in too for email announcements.
maybe we ccould stay in contact to unite our efforts.
thanks so far.
Hans-Jochen

[Yury G.]

Sure! If I succeed in creating desired model I'll be happy to share it here or, maybe, create a working example from it and put it in opensim-core sandbox.

[jimmy d]

Hi, If you are interested in the testing OpenSim IMU, please join the opensim-testers email list and email me at jjdunne@stanford.edu.