• Son Luu

MoCap Lab 3

Updated: Nov 14, 2019


Class: Motion Capture

Instructor: Kat Sullivan - Todd Bryant



Lab 3 Objectives: Record a team member in mocap. Review the data for any gaps, and fix it. Take screenshots of the data, before and after (or give us the filepath to review), and export fbx files for retargeting.


Process

1. Last week during Lab 2, we didn't capture T-pose of the character in the video. Thus, as part of Lab 3, we had to re-record the video with T-pose included for we move on to data leaning.


This step included the process of dressing the subject with motion capture suit and 37 sensors.


Below, Dana was our subject for motion capture.



2. We began to record a series of different movements into different files.


We decided to create various body movements from simple to complex and observe how the data turns out to be.



Take 1: walking in circle




Take 2: walking diagonally




Take 3: running




Take 4: rolling on the floor





Take 5: free movements




Take 6: interacting with an object



Take 7: interacting with another object




3. Before data cleaning, we made copies of all the takes.


4. Data cleaning process


The first take achieved 99% capturing. As the movements became more complex, we started to see more and more unlabelled dots.


5. Exploring and learning techniques




On the right hand panel, clicking on Labeling will reveal the level of data capturing - attached to a percentage that represents how much labeling has been captured.


If it doesn't have a number, then it's 100% captured. Otherwise there's a number that indicates how much has been captured.


In this case, on the body part is highlighted in red has 99% data captured. Then, you can zoom into the missing or unlabeled part to relabel.




In this image, the unlabeled part can be fixed using Interpolation function in the Editing Tool. Interpolation basically bridges any gap in the motion that was not captured based on a movement prediction model, whether it be a Linear, or Cubic, or Pattern-based, Constant, or Model-based. Small gaps can easily fixed by using this function.










Lessons Learned:

1. Remember to save a copy of the original data file

2. When in doubt, save the data files as FBX files.

3. Make sure Export Skeleton is on when saving.

4. Many unlabeled were ghost points and could just be deleted.

5. Navigation to the unlabeled points can be time consuming. Even using "Find Next" will also requires a little more navigation by searching on the timeline and zooming into the targeted track (the track with the missing or unlabeled points).