Smart Annotation Tool for Multi-sensor gait based daily activity data

Beitrag bei einer Tagung
(Originalarbeit)


Details zur Publikation

Autorinnen und Autoren: Martindale C, Roth N, Hannink J, Sprager S, Eskofier B
Jahr der Veröffentlichung: 2018
Tagungsband: 2018 IEEE International Conference on Pervasive Computing and Communications Workshops (PerCom Workshops)
Sprache: Englisch


Abstract

The monitoring of patients within a natural, home environment is important in order to close knowledge gaps in the treatment and care of neurodegenerative diseases, such as quantifying the daily fluctuation of Parkinson’s patients’ symptoms. The combination of machine learning algorithms and wearable sensors for gait analysis is becoming capable of achieving this. However, these algorithms require large, labelled, realistic datasets for training. Most systems used as a ground truth for labelling are restricted to the laboratory environment, as well as being large and expensive. We propose a study design for a realistic activity monitoring dataset, collected with inertial measurement units, pressure insoles and cameras. It is not restricted by a fixed location or capture volume and still enables the labelling of gait phases or, where non-gait movement such as jumping occur: on-the-ground, off-the-ground phases. Additionally, this paper proposes a smart annotation tool which reduces annotation cost by more than 80%. This smart annotation is based on edge detection within the pressure sensor signal. The tool also enables annotators to perform assisted correction of these labels in a post-processing step. This system enables the collection and labelling of large, fairly realistic datasets where 93% of the automatically generated labels are correct and only an additional 10% need to be inserted manually. Our tool and protocol, as a whole, will be useful for efficiently collecting the large datasets needed for training and validation of algorithms capable of cyclic human motion analysis in natural environments.


FAU-Autorinnen und Autoren / FAU-Herausgeberinnen und Herausgeber

Eskofier, Björn Prof. Dr.
Lehrstuhl für Informatik 14 (Maschinelles Lernen und Datenanalytik)
Hannink, Julius
Lehrstuhl für Informatik 14 (Maschinelles Lernen und Datenanalytik)
Martindale, Christine
Lehrstuhl für Informatik 14 (Maschinelles Lernen und Datenanalytik)
Roth, Nils
Lehrstuhl für Informatik 14 (Maschinelles Lernen und Datenanalytik)
Sprager, Sebastijan Dr.
Lehrstuhl für Informatik 5 (Mustererkennung)


Zitierweisen

APA:
Martindale, C., Roth, N., Hannink, J., Sprager, S., & Eskofier, B. (2018). Smart Annotation Tool for Multi-sensor gait based daily activity data. In 2018 IEEE International Conference on Pervasive Computing and Communications Workshops (PerCom Workshops). Athens, GR.

MLA:
Martindale, Christine, et al. "Smart Annotation Tool for Multi-sensor gait based daily activity data." Proceedings of the 2018 IEEE International Conference on Pervasive Computing and Communications Workshops (PerCom Workshops), Athens 2018.

BibTeX: 

Zuletzt aktualisiert 2019-08-04 um 10:34