Improving fit and smoothing motion

Main Forum
#1

Hello AnyBody,

I am working with a minimal marker set based on points that correspond to the Microsoft Kinect skeleton. I have based my script on Full Body Gait in the AMMRV1.4.1 repository. The end goal is to measure the mechanical work done to accelerate and deccelerate the body - and this is done by integrating Pmech vs. time. The results are higher than expected and I think this is due to the motion/marker fit. When watching the model run, it appears to have a jerky motion in which the skeleton/red markers “snap” to the data/blue markers but then “relax” and then the cycle of “snapping” and “relaxing” repeats. I am also dealing with long trials and it is frustrating to have kinematics fail at the very end. I have increased the KinematicTol to 2e-2 to reliably run without error. By reducing the KinematicTol (back to the default of 1e-3) it appears to help this issue, however, the kinematics very rarely complete without an error at this lower tolerance.

Additionally, I have also tried increasing the filtering (by lowering the cut-off frequency of the low-pass filtering in Model Setup) and I have tried using a low-pass butterworth filter upon writting the joint angle outputs to file. Neither of these have drastically changed the results of the work calculation.

Please let me know if you think there is anything else that can be done to improve the fit of motion data, make the fit consistent, or more effectively filter the motion.

Thanks for your help,

Jeremy

#2

Hi Jeremy,

I have a couple of comments:[ol]
[li]I think the tolerance should not be increased, this may potentially cause other problems.[/li][li]Look for the “big” marker errors: If there are marker errors which the model can not reduce due to the configuration and the design variables etc. try to look for ways to reduce the error on these first. The big errors may make it harder for the solver to converge in the end. Additionally if you have a big marker error say for example in the pelvis which can not be made any better because the markers has a locked position, then the error there may cause problems in other areas where you are looking for relative smaller motions.[/li][li]Look for ways to isolate the problem:[LIST=1][/li][li]Try to add joint drivers on parts of your model and remove the markers associated with this body parts.[/li][li]Then run the motion optimization of the model and save the optimized joint angles and dimensions.[/li][li]Run a second optimization where the found motion in step 2 is used as joint drivers and the markers used in step 2 is taken out. Add back the markers removed in step 1 and remove the drivers added in step 1.[/li][/ol]
[/LIST]In this way you run the optimization of the motion in two stages, which is more tedius, but it might help you to localize the problem

Best regards
Søren

#3