Validity of Gap Filling during 3-D data analysis

[Matt Miltenberger]

Hello,

I was wondering about gap filling missing data points during 3-D data analysis. Some of my markers disappeared as I was collecting data at 60 hz, I used gap filling to put in the missing data points which got me wondering about the validity of doing this for an extended amount of frames. Is there a set number of frames or time period that I can feel confident about my data being "filled in" without compromising validity. In addition, would it depend on frequency of data collection and type of movement (precise vs gross motor). Thank you in advance for your help

[John Rasmussen]

Re: Validity of Gap Filling during 3-D data analysis

Hi Matt,

In our experiences, gap filling is not a good option. It is basically an artefact you are introducing into your data. If you have enough markers and sufficiently intelligent software, then you may be able to use redundancy in the marker configuration to to compensate for the missing markers. You can see how we do this in the AnyBody Modeling System here: http://www.anybodytech.com/fileadmin...e/lesson6.html.

Best regards,
John

[Brian Schulz]

Re: Validity of Gap Filling during 3-D data analysis

The number of frames you can confidently gap fill depends on many details, but the primary concern is how well the missing region can be interpolated from the existing data. A gap in a smooth, consistent trajectory- whether during a precise or gross task- is very different from a gap in a jerky, inconsistent trajectory.
The interpolation method also matters quite a bit- filling gaps using linear interpolation induces spikes in differentiated velocity and acceleration while higher-order polynomials or spline fits would be far less likely to do this.
My advice would be to fill as only the smallest gaps possible and to use at least a third-order polynomial or spline fit to do it.
I currently only fill gaps of a single frame using a third-order polynomial to fix very brief marker jumps that I snip out, but I have pretty good data (mostly gait) without many gaps.
Perhaps a good way for BIOMCH-L to help address this question would be for everyone who is willing to share to list the maximum size of gap they are comfortable with filling, what kind of data they are working with, and their preferred method of gap filling.