Comparing DeepLabCut's gait recognition performance to a marker-based motion capture system
Keywords
Loading...
Authors
Issue Date
2021-06-18
Language
en
Document type
Journal Title
Journal ISSN
Volume Title
Publisher
Title
ISSN
Volume
Issue
Startpage
Endpage
DOI
Abstract
In the eld of motor control, kinematic analysis is often performed using
optical motion capture systems. These techniques usually use a camera sys-
tem combined with markers attached to bony landmarks on the participant's
body in order to obtain kinematic data. There are several drawbacks to this
method. Markers can be intrusive or distracting, motion capture systems are
costly and time consuming to work with, and skilled operators are required
to use the system correctly. Markerless methods o er an alternative way to
collect kinematic data with several advantages: they are cheap, easy to work
with and they can be used outside of a laboratory setting. DeepLabCut is
one such markerless method. The goal is to determine whether DeepLabCut
is accurate enough to be a suitable alternative to a marker-based system.
This work compares the performance of a deep learning-based markerless
motion tracking algorithm to that of a traditional optical motion capture
system. I measured various evaluation metrics using the euclidean distance
between both systems, using the motion capture measurement as ground
truth. This work nds that DeepLabCut is a suitable alternative to marker-
based motion capture systems.
Description
Citation
Faculty
Faculteit der Sociale Wetenschappen