eprintid: 32550
rev_number: 11
eprint_status: archive
userid: 7123
dir: disk0/00/03/25/50
datestamp: 2023-01-11 14:07:04
lastmod: 2023-01-31 09:09:09
status_changed: 2023-01-11 14:07:04
type: doctoralThesis
metadata_visibility: show
creators_name: Große Sundrup, Jonas Christopher
title: Classification of Human Motion with Applications in Gesture Recognition and Treatment of Major Depressive Disorder
divisions: i-110001
adv_faculty: af-11
abstract: In this thesis we are investigating the question of classification of human motion from three different perspectives:
Firstly, we develop a methodology that allows the assessment of human muscle signatures to differentiate between gestures they originate from under consideration of natural shifts of those signatures in time with each motion.
We can show that our proposed method can leverage Dynamic Time Warping to successfully classify hand gestures based on muscle signals with an accuracy corresponding to or exceeding the current state of the art.
In addition to that, we can show that our proposed method allows insight into the underlying mechanisms this classification results are based on and why they are so accurate.
We can show that this insight allows for the derivation of intelligent improvements and modifications to the method to improve accuracy even further or to adapt it to specific circumstances.
Secondly, we investigate the automated, but understandable derivation of classification procedures for the differentiation between motor patterns inside and outside of intervals of stressful rumination for patients diagnosed with major depressive disorder.
We can not only demonstrate that an automated tailoring of a classification procedure is possible, but also that an appropriate splitting between parts of the methodology allows for an assessment of impact for each involved component, yielding the option of intelligent choice for the underlying problem.
As a consequence, we can propose a method that is tailored specifically to the underlying structure of the foundational patient data that was obtained within the project and is capable of yielding a lightweight, optimized classification procedure that is also suited for medical applications due to its understandability.
Thirdly, we investigate the question of optimal sensor location, to derive an optimal sensor layout for specific or comparable applications under consideration of application-specific constraints and requirements.
We propose a method based on an exemplary reconstruction problem that allows the automated assessment of quality of a specific sensor layout.
This assessment is then shown to be useful as a basis on which an automated procedure can derive the optimal sensor layout for a specific application.
In addition to that, we can show that this method is capable to quantify the sensitivity of the underlying problem to properties of the sensor hardware in use, allowing for a quantitative assessment of the necessity of constraints on the sensor hardware used.
date: 2022
id_scheme: DOI
id_number: 10.11588/heidok.00032550
ppn_swb: 1832749460
own_urn: urn:nbn:de:bsz:16-heidok-325504
date_accepted: 2022-12-19
advisor: HASH(0x55fc36c05400)
language: eng
bibsort: GROSSESUNDCLASSIFICA2022
full_text_status: public
place_of_pub: Heidelberg
citation:   Große Sundrup, Jonas Christopher  (2022) Classification of Human Motion with Applications in Gesture Recognition and Treatment of Major Depressive Disorder.  [Dissertation]     
document_url: https://archiv.ub.uni-heidelberg.de/volltextserver/32550/1/dissertation_jonas_grosse_sundrup_digital.pdf