Anyone know a solution for misaligned origin at calibration?

[Jennifer Perry]

I have some data collected in cortex from about a year ago that looks to have an incorrect lab origin. We think this because the force plate data appear to be shifted about 2 or 3 inched along the lab y-axis. (COP during gait is lateral of foot on right stance and medial of foot on left stance for all subjects). Does anyone know of a solution for this? Seems to have been an issue for about 3 months worth of data (that no one currently in the lab collected).

Thank you!

[Ton van den Bogert]

Re: Anyone know a solution for misaligned origin at calibration?

Jennifer,

If you do inverse dynamic analysis, you usually get residual loads (3d force and moment) at the top of the model or at the pelvis. Opensim gives you this information, and other software systems probably also. The residual moments will be large when the forceplate coordinate system is shifted relative to the motion capture coordinate system. So you could do an optimization (or trial and error), applying a constant shift to the force plate horizontally until the RMS value of the residual moments (over time) is minimized.

A variation of this idea was first described by Art Kuo in 1998: https://www.ncbi.nlm.nih.gov/pubmed/9675694.

Ton van den Bogert

[Robert D. Catena]

Re: Anyone know a solution for misaligned origin at calibration?

Jennifer,

We had the same problem, but fortunately only for one testing (not 3 months!). Happened because the Lframe pins were hanging off the side of a portable forceplate. Easy to do if you are in a rush. We should probably talk to MAC about this design issue. We now have a lab policy that if we are using the portable forceplates, that the Lframe position is double checked by another person before calibration because this can be such a devastating issue to your data. Here is what we did to correct the data:

First, you need to make sure your suspicions are real. What I did was took 3 testings very close together where the cameras should have been in the exact same location. One testing was the bad data, and the other 2 testing before and after that testing, but with different calibrations. I compared the camera positions in the room between the bad calibration and good calibrations. You can do this by going to tools->settings->cameras->configuration. In this box it will show you the highlighted cameras position and angle relative to your origin. From this I confirmed the data were indeed misaligned, and also by how much in linear and angular directions!

Second, I think there are two options here:
(1) You could possibly move the forceplate data in cortex to realign with the marker data in the settings->forceplate configuration. Easy to do if the change is just linear, but if it is also rotational, then you have to do some figuring out of the rotational matrix offset, and then you would have to rerun the calibration on all the data. I did not feel totally comfortable with this, and it would take more time, so I did not choose this option.
(2) Instead, cortex (I'm using v5.5.01579) also allows you to shift all the marker data in space a specified amount. In the postprocessing frame, under postprocessing tools on the left side of the screen there is an "X" tab. Click on that and you will see where you can make "Global Marker Data Adjustment". I just took how far my cameras were off between good and bad calibrations and applied those numbers here to shift the marker data.

If you have any other questions, feel free to email me directly at robert.catena@wsu.edu

[Edmund Cramp]

Re: Anyone know a solution for misaligned origin at calibration?

This is easy to correct if your data is stored in the C3D format, or can be exported to C3D and then exported to whatever format you are using - I've done this many times for customers. If you know the correct force plate location then you can just edit the C3D file FORCE_PLATFORM:CORNERS parameters to correct the force plate locations, save the file and the problem is solved. If you have a lot of data with the same problem then you can use a program like the C3Deditor to fix all the files automatically.

[Ian Hoelker]

Re: Anyone know a solution for misaligned origin at calibration?

Hello Jennifer,

You’re not alone out there. It is not uncommon to have collected data with misaligned force plates or other experimental setup parameters such as the location of joint centers & landmarks, sensitivity matrices, etc. To assist our clients in resolving these issues we added the capability to modify setup parameters in post processing and resave the data with the edited parameters. If the error occurred in other collection software, importing C3D files into The MotionMonitor’s C3D Model Builder (https://www.innsport.com/application...l-builder.aspx) would permit a batch edit on all of your files. These changes would then automatically carry through to the definition of your biomechanical model, kinematic and kinetic analyses as well as ensemble averages and standard deviations.

Ian Hoelker
www.TheMotionMonitor.com

[Matthew J. Major]

Re: Anyone know a solution for misaligned origin at calibration?

Hello Jennifer,

My laboratory also experienced this problem and if from the same source it stemmed from an issue with recommendations proposed by the manufacturer (MAC) regarding calibration of the system prior to data collection. We also discovered this error through an obvious drift in the forceplate-generated COP data. The manufacturer was able to provide a Cortex post-processing patch to correct these errors and I would recommend contacting them for this solution.

Matty

[Emily Schaefer]

Re: Anyone know a solution for misaligned origin at calibration?

Hi Jennifer,

Motion Analysis has a solution for this and is happy to help. Please feel free to contact Customer Support at support@motionanalysis.com.

Emily Schaefer
Motion Analysis Customer Support

[Edmund Cramp]

Re: Anyone know a solution for misaligned origin at calibration?

It seems that there are lots of ways of fixing the problem, but if you were to use a device like the MTD-3 (a force plate "Mechanical Testing Device") on a regular basis before collecting trial datasets then you would be able to avoid the problem. C-Motion sells the MTD with a software application (Caltester) that makes it very simple to use - the MTD is a simple (but carefully made) metal rod with five retroreflective markers (it also supports active marker systems) that allows you to perform a quick test by applying a force vector to your plate while your 3D system records the physical vector. If the two vectors align then you are good to go. If they are not aligned then you have a problem and can fix it prior to collecting the data.

If you use an MTD to check that the force plate and laboratory coordinate systems are aligned on a regular basis than you'll never have this problem.

[Jennifer Perry]

Re: Anyone know a solution for misaligned origin at calibration?

Thank you all for your help! We will explore some of the options given here. I never thought of using the residuals to work backwards towards an optimized solution.

[Tomislav Pribanic]

Re: Anyone know a solution for misaligned origin at calibration?

It seems to me that you are dealing here with a classical problem of 3D point cloud registration , a well-studied task in computer vison and image processing. In brief, it provides you with a rotation matrix R and a translation vector T which align two point clouds, each of them initially expressed in the different coordinate system. In the case of so called fine registration (when you have rather good initial solution/pretty much aligned sets) perhaps the most famous algorithm is Iterative closet points (ICP) In a (slightly) different context, for biomechanical analysis, you may read about it here (or in more detail in Challis, J.H. (1995). A procedure for determining rigid body transformation parameters. J. Biomechanics 28, 733-737).

Not aware about the details of your problem and it's setting, but let’s say that you already have or can still somehow generate at least three pairs of correspondent 3D points before and after the change of your laboratory coordinate system. Than you are good to go in finding (R, T) relating parameters between two point clouds that you have generated before and after the coordinate system movement. I think you can use matlab if you do not have your own ICP routine to actually compute R&T.

For completeness, the real challenge in 3D registration is to actually determine correspondent point pairs automatically. In your case, for a one-time effort and just few points to be found, doing it manually for few points is just fine.
If you would like to know more about theory behind:
J. Salvi, C. Matabosch, D. Fofi, J. Forest, A review of recent range image registration methods with accuracy evaluation, Image Vision Computing 25 (5) (2007) 578–596.
And of course things get even fuzzier in case registration of non-rigid point sets:
G. K. L. Tam, Z. Q. Cheng, Y. K. Lai, F. C. Langbein, Y. Liu, D. Marshall, R. R. Martin, X. F. Sun, P. L. Rosin, Registration of 3D point clouds and meshes: A survey from rigid to nonrigid, Transactions on Visualization and Computer Graphics 19 (7) (2013) 1199–1217.

Good luck, Tomo.

[Allan Carman]

Re: Anyone know a solution for misaligned origin at calibration?

A quick and easy way to check force plate alignment and calibration prior to testing is to place a calibration weight on the plate with a retro-reflective marker placed in the center. The 3D system reconstructs the 3D point and displays the force vector and magnitude based on plate calibration info. You can quickly check force magnitude, zeros and CoP alignment between the 3D and plate coordinate systems in the collection software. There was one case where when testing several points on the same plate that the CoP error increased (distance between 3D point and displayed force vector) the further the weight was placed from the origin of the plate, turned out there was in incorrect calibration coefficient (sensitivity) entered into the software.