Daily life activities such as working and shopping may cause people to carry overloaded bags, frequently borne in an incorrect way (e.g. only on one shoulder, asymmetrically worn). When these activities alter the gait, back pain incidents can occur. Critical conditions can be monitored taking advantage from a wearable assistant, extracting contextual information by on-body acceleration signals. By acquiring data on trunk, limb and foot during gait, we are able to detect five walking tasks on loaded conditions: two-straps backpack carried on shoulders, backpack carried with a single strap on right and left shoulder, bag carried with the right and left hand. Seven subjects participated walking at self-selected speed on a treadmill carrying a load between 10-12% of their body weight. Subjects repeated each task for five times over three weeks. We classified the activities for a single user by use of KNN, naïve Bayes and SVM classifiers. KNN achieved the best recognition accuracy of 96.7% for day dependent classifier training. The sensors placement, which resulted to be different along consecutive days, affects performance evaluation: a +3° rotation on the coronal plane decreases the accuracy to 76.0%

Wearable assistant for load monitoring: Recognition of on-body load placement from gait alterations

Elisabetta Farella;
2010-01-01

Abstract

Daily life activities such as working and shopping may cause people to carry overloaded bags, frequently borne in an incorrect way (e.g. only on one shoulder, asymmetrically worn). When these activities alter the gait, back pain incidents can occur. Critical conditions can be monitored taking advantage from a wearable assistant, extracting contextual information by on-body acceleration signals. By acquiring data on trunk, limb and foot during gait, we are able to detect five walking tasks on loaded conditions: two-straps backpack carried on shoulders, backpack carried with a single strap on right and left shoulder, bag carried with the right and left hand. Seven subjects participated walking at self-selected speed on a treadmill carrying a load between 10-12% of their body weight. Subjects repeated each task for five times over three weeks. We classified the activities for a single user by use of KNN, naïve Bayes and SVM classifiers. KNN achieved the best recognition accuracy of 96.7% for day dependent classifier training. The sensors placement, which resulted to be different along consecutive days, affects performance evaluation: a +3° rotation on the coronal plane decreases the accuracy to 76.0%
978-963-9799-89-9
File in questo prodotto:
Non ci sono file associati a questo prodotto.

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11582/329154
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
social impact