Activity Recognition with Wearable Sensors on Loose Clothing

Research output: Contribution to journalArticlepeer-review

21 Citations (Scopus)
258 Downloads (Pure)

Abstract

Observing human motion in natural everyday environments (such as the home), has evoked a growing interest in the development of on-body wearable sensing technology. However, wearable sensors suffer from motion artefacts introduced by the non-rigid attachment of sensors to the body, and the prevailing view is that it is necessary to eliminate these artefacts. This paper presents findings that suggest that these artefacts can, in fact, be used to distinguish between similar motions, by exploiting additional information provided by the fabric motion. An experimental study is presented whereby factors of both the motion and the properties of the fabric are analysed in the context of motion similarity. It is seen that while standard rigidly attached sensors have difficultly in distinguishing between similar motions, sensors mounted onto fabric exhibit significant differences (p<0.01). An evaluation of the physical properties of the fabric shows that the stiffness of the material plays a role in this, with a trade-off between additional information and extraneous motion. This effect is evaluated in an online motion classification task, and the use of fabric-mounted sensors demonstrates an increase in prediction accuracy over rigidly attached sensors.
Original languageEnglish
JournalPLOS One
Volume12
Issue number10
DOIs
Publication statusPublished - 4 Oct 2017

Keywords

  • Wearable sensors
  • Fabric sensors
  • Textiles
  • Machine Learning

Fingerprint

Dive into the research topics of 'Activity Recognition with Wearable Sensors on Loose Clothing'. Together they form a unique fingerprint.

Cite this