Hello,
What is your experience with inertial measurement as an alternative to camera/marker systems for motion capture?
Do you like it?
Why?
What would you say are the advantages/disadvantages?
Is it easier?
Is it less accurate?
There are several systems, see e.g:
https://www.xsens.com/tags/motion-capture/
Regards,
John Rogers
IMU vs. camera/marker mocap
-
Benjamin Michaud
- Posts: 31
- Joined: Mon May 03, 2010 6:35 am
Re: IMU vs. camera/marker mocap
Dear John,
Here at the lab, we actually are using both. As you may expect, there is kind of drawbacks for both approaches.
IMU has the great advantage of being easy to use outside of the lab while being actually quite accurate.
The main problem with them is the drift over time. From my experience (XSENS Awinda), 10 minutes of use is still short enough though to prevent any drift. We tested it on a machine with an arm that was continuously rotating and changing direction, and after 10 minutes, it was still below 2 degrees of difference with real position (estimated via the controller of the machine).
Other problem (which is more biomechanics related) is the size of the units. Compared to reflective markers, it is way larger and heavier. Hence, the subject may experience discomfort and changes its kinematics. But more important, the larger inertia may less accurately tracks the kinematics of the underlying bone you try to track.
It is easy to use if you use the default model proposed by the company, but requires time if you want to build your own kinematic chain. Here, we decided to code the kinematic reconstruction by ourselves using an Extended Kalman Filter. This was challenging, but doable.
For the accuracy, it depends on your need. Vicon system claims precision up to 0.2mm, but it is for the marker tracking. When it comes to kinematic reconstruction, it is much harder to evaluate. For the upper limb (as we do), I think the lost in precision is largely compensate by the wobbling mass effects.
For the systems, we tested two of them: Delsys and XSens. I strongly recommend against Delsys as it is completely inaccurate and after having discussed for a full year and half with them, we decided to drop them. XSens on the other hand was just so good. I have nothing to say against. I know that a team from Universite de Sherbrooke published a paper comparing accuracy and drift, I can't remember the name of the group, but you will find them easily on scholar, and XSens was topping every tests.
Regards!
Here at the lab, we actually are using both. As you may expect, there is kind of drawbacks for both approaches.
IMU has the great advantage of being easy to use outside of the lab while being actually quite accurate.
The main problem with them is the drift over time. From my experience (XSENS Awinda), 10 minutes of use is still short enough though to prevent any drift. We tested it on a machine with an arm that was continuously rotating and changing direction, and after 10 minutes, it was still below 2 degrees of difference with real position (estimated via the controller of the machine).
Other problem (which is more biomechanics related) is the size of the units. Compared to reflective markers, it is way larger and heavier. Hence, the subject may experience discomfort and changes its kinematics. But more important, the larger inertia may less accurately tracks the kinematics of the underlying bone you try to track.
It is easy to use if you use the default model proposed by the company, but requires time if you want to build your own kinematic chain. Here, we decided to code the kinematic reconstruction by ourselves using an Extended Kalman Filter. This was challenging, but doable.
For the accuracy, it depends on your need. Vicon system claims precision up to 0.2mm, but it is for the marker tracking. When it comes to kinematic reconstruction, it is much harder to evaluate. For the upper limb (as we do), I think the lost in precision is largely compensate by the wobbling mass effects.
For the systems, we tested two of them: Delsys and XSens. I strongly recommend against Delsys as it is completely inaccurate and after having discussed for a full year and half with them, we decided to drop them. XSens on the other hand was just so good. I have nothing to say against. I know that a team from Universite de Sherbrooke published a paper comparing accuracy and drift, I can't remember the name of the group, but you will find them easily on scholar, and XSens was topping every tests.
Regards!