Macquarie Motion Capture Manual

Understanding Vicon Nexus

Home
Intro
6 Steps
1. System Prep
2. Subjects
3. Capture
4. Gap Filling
5. Output
6. Analysis
Terminology
System Specs
About & Contact
Cleaning Up Your Data
This page deals with issues of missing and messy data. In an ideal capture session, your captured trials will be
reconstructed perfectly, with no gaps or missing markers. In reality, markers disappear with some frequency.
How often this occurs will depend on a multitude of factors. At the most basic level, data quality is dependent
upon: the number and placement of motion cameras, speed of subject movement, and frequency of marker
occlusion.
Gap lling and data cleaning are the bane of motion capture operators. It is far quicker to spend 20 minutes
optimising the capture volume before a session starts, than it is to spend hours, days, or even months cleaning
up bad data. To optimise your system requires an understanding of how it works. Thus, this section concludes
with a few Tips on Getting Clean Data.
There are ve general types of data issues: 1) markers disappearing momentarily, 2) markers becoming
unlabeled (gray), and then becoming labeled again (yellow/coloured), 3) markers being incorrectly labeled after
disappearing momentarily, 4) markers disappearing and then reappearing only to stay unlabeled, and 5) markers
moving erratically at times. The rst section, Gap Filling, will address the issues 1 and 2. The second section,
Missing Data, will address issues 3 and 4. We have no solution to the 5
th
issue.
Gap Filling
This section covers the rst two types of missing data. For this section, a new capture trial was conducted, with
the explicit aim of creating messy data.
The trial should be loaded and reconstructed. 1.
The rst step in data cleaning is to examine the entire trial by eye, and mentally identify the various
problems. Click the play button on the Time Bar.
2.
Nexus does a reasonable job of automatically identifying data issues, and will often pick up small gaps
imperceivable to humans when the trial is replayed at normal speed. Under the Tools pane, click the
Label/Edit button, and examine the Gap Filling section. Figure 1 illustrates the Gap Filling section of the
Label/Edit pane for our messy data. This section lists instances of missing data. The Trajectory column tells
us which marker went missing, while the #Gaps column tells us how many times during that trial that
marker went missing.
3.
Left-click on one of the marker names (Intermediate in Figure 1). The Time Bar will automatically jump to
the rst instance where a marker gap occurred. The blue squiggly lines don't relate to missing data, but are
instead a history line of marker motion throughout the trial.
4.
Notice when we click on dierent marker names in the Gap Filling list, the information in the Range
textboxes below this changes. The range numbers indicate the number of frames before and after the
5.
current frame where the marker has gone missing. In Figure 1, the marker went missing 1 frame before the
current frame, and continued to be missing for 2 more frames. Thus, Nexus jumps to the middle of the gap,
not the rst frame. For example, if you had a marker missing for 201 frames, you would see -100, 100. If it
went missing for 20 frames, you would see -9, 10. By taking the absolute value of these, adding them
together, and adding 1, you know how many frames the marker went missing for. In the rst gap instance
for the Intermediate marker, the gap was 4 frames (1 + 2 + 1).
It is at this point that the art of motion capture comes into play. The size of the gap will depend on what gap
lling algorithm you use, how the result will look, and whether an alternate method will be needed. There
are however two general approaches: 1) small gaps should be lled with the Spline Fill algorithm. 2) large
gaps should be lled with the Pattern Fill algorithm.
6.
The spline ll method is an automatic method, which essentially extrapolates the missing trajectory based
on the last known, and rst reappearing coordinates. This method will be the primary form of gap lling you
will use, and can generally be used for gap instances of around <=60 frames. The larger the gap, to more
likely the spline ll will give an erratic result. The spline ll algorithm is also extremely susceptible to erratic
motion occurring the last few frames before the marker gap. This is because the spline ll extrapolates the
unknown trajectory based on where it was going before it disappeared. Unfortunately, a marker often
disappears due to erratic motion (sometimes a result of the system itself, not the subject).
7.
The Pattern Fill algorithm is used in cases where the Spline algorithm fails. For this method, you pick
another marker which you think exhibits a similar motion to the missing marker. Nexus then generates a
trajectory based on the selected marker. For example, if we were dealing with two markers on a xed bone,
and one went missing, we would select the nonmissing marker. This method works best when we are
dealing with body symmetric markers (e.g., lip corners). We will now run through an example of Pattern Fill.
8.
A slightly larger gap for the intermediate marker is selected. Next, note down the frame at which the gap
occurred. The current frame number is displayed on the Time Bar, below the word General. In Figure 1, the
current frame is 246, in Figure 2, it is 627. Click the Pick Source button, and select the Proximal marker, as
shown in Figure 2. One the proximal marker has been selected, the Fill button will be enabled. Click the Fill
button.
9.
You should now review the results of the ll operation. Do this by clicking the Play button, paying close
attention when you approach the frames of missing data. If the gap algorithm produced erratic results, you
can undo the operation by clicking control-z, or clicking the third left-most icon button under the le menu
(a u-turn icon). You can redo this undone operation by hitting control-y, or the fourth left-most icon button
(the other u-turn icon).
10.
If you are happy with the gap ll, click Save. 11.
We will now examine the second 12.
Figure 1. The Gap Filling section of the Label/Edit pane show numerous gaps in our trial.
Figure 2. Pattern Fill algorithm is used for larger gaps.
Missing Data
This section covers the third and fourth type of missing data. The third is when after a gap, markers are labelled
incorrectly. The fourth is when a marker goes missing, and then comes back as gray. Nexus will not report the
rst issue (as is has done the labelling), and often the gap in the fourth case. Instead, you will identify these
issues when you rst play the reconstructed trial from the Time Bar.
We know of only two solutions which can be applied to these issues. The rst is neat, the second is not. 1.
The rst solution involves manually relabelling the marker(s) in question. 2.
First, go to the place in the Time Bar where the gap issue has occurred. Under Tools pane, click the
Label/Edit icon button. Under the Manual Labeling section, the set of template markers is listed. Click the
rst marker name. The mouse will turn into a cross hair. Select the appropriate marker, going through the
entire list. When done, replay the entire trial.
3.
If this operation is successful, Nexus will have relabelled the marker without unlabelling/mislabelling the
same marker elsewhere in the trial.
4.
If the operation is successful, save the trial. If it is not, undo the operation. 5.
The second solution involves xing the outputted data le after the rest of the gaps have been lled. 6.
Sometimes Nexus is unable to automatically relabel a marker due to an error in marker reporting. This
occurs because for a split second, Nexus thinks that two unique markers exist at very similar points in
space. The data le for this instance is illustrated in Figure 3 (we will cover data les later in Analysis). A copy
of this le can be found in the resources directory, which accompanies this manual.
7.
The data lists ve markers: Metacarpal, Proximal, Intermediate, Distal, and *1. Notice how marker *1 has no
data points up until frame 657. At this exact frame, Distal has its last frame. Thus, Nexus is reporting the
same marker in two distinct locations. We x this issue by simply cutting the remaining data points from *1
marker, and pasting them into the Distal gap. This solution works because we can see what has happened.
However, in more complex templates, with multiple markers disappearing, this solution may not work.
8.
Figure 3. Data le when manual relabelling fails.
Tips on Getting Clean Data
As mentioned at the beginning of this section, the quality of captured data is largely dependent upon three
things: cameras, subject speed, and marker occlusion. We will address each of these areas separately
respectively.
Camera Numbers and Placement
The most important factor in obtaining clean data is the motion capture cameras. How many cameras you have,
and their placement with respect to the motion being observed, will determine the quality of your data. The
simplest method to increase data quality is to buy more cameras. However, given the cost of motion systems,
such a strategy is not always feasible. Thus, if you have limited cameras, how you place the cameras in relation to
the motion being captured is vital.
In order for a marker to be visualised, it must be viewable by at least two cameras. However, this minimum
requirement is often insucient. In such an environment, markers will frequently disappear due to: occlusion,
quick movements, or system errors. Thus, it is advisable to have markers viewable by three or more cameras if
possible. In the Macquarie setup, which began with facial analysis, four cameras were placed in a tight array,
spaced 2m from the subject. This setup produced high-quality data. You may consider using a similar
arrangement if examining a localised body part.
Processor Settings and Subject Speed
Vicon is designed to capture a whole range of movement, from the very slow (turtle) to the very quick (golf
swing). Because of this, Nexus has a variety of settings which allow trade-os between data quality and subject
speed. These settings can be found under the Resources pane, System tab, clicking on the Local Vicon System
tree node, in the properties subpane. Specically: marker movement speed, label model rigidity, and
quality/speed. If your model is slow moving, with relatively xed markers, you will want to set these values low. A
description of these and settings can be found in Managing Local Vicon System of the Vicon system manual.
Subject Tips
Occlusion is the single biggest problem of marker-based motion capture systems. It is because of this issue that
large, multi-camera setups are needed. When considering the size of typical markers (3mm, 1cm), the smallest
crease in clothing, fold of skin, shape of the body (e.g., lip curl), or loose hair, is enough to cause occlusion, and
therefore missing data. There however a number of simple steps you can take to minimise these and other
issues.
If capturing the body, ask subjects to wear tight tting, minimal clothes (e.g., micra). Markers also stick
better to skin than to clothing. Also, the model will be far more rigid (clothing is loose and dynamic, but skin
is xed to the body).
1.
Ask subjects to tie their hair back. 2.
If capturing facial data, ask female subjects to minimise the amount of make-up, and ask male subjects to
shave. These issues eect how well markers will stick to the face. If a marker falls o for any reason, you will
need to recalibrate the subject.
3.
Consider the shape of your subjects body, and vary marker placement accordingly. For example, if subject's
lower lip has a pronounced curl, place the marker a lower to avoid occlusion by the lip. This will often
happen if the subject tilts their head down momentarily.
4.
Ask subjects to remove all jewelry and shiny objects. 5.
Ask subjects to minimise non-related movements during the capture (e.g., licking lips, scrunching up their
nose). This will reduce the frequency of movement spikes in the data.
6.
This Section:
Cleaning Up Your Data
Gap Filling
Missing Data
Tips on Getting Clean Data
Written by Steven R. Livingstone. Designed by Free CSS Templates, Thanks to Dubai Villas