TY - JOUR
T1 - Fall detection from a manual wheelchair
T2 - preliminary findings based on accelerometers using machine learning techniques
AU - Abou, Libak
AU - Fliflet, Alexander
AU - Presti, Peter
AU - Sosnoff, Jacob J.
AU - Mahajan, Harshal P.
AU - Frechette, Mikaela L.
AU - Rice, Laura A.
N1 - Publisher Copyright:
© 2023 RESNA.
PY - 2023
Y1 - 2023
N2 - Automated fall detection devices for individuals who use wheelchairs to minimize the consequences of falls are lacking. This study aimed to develop and train a fall detection algorithm to differentiate falls from wheelchair mobility activities using machine learning techniques. Thirty, healthy, ambulatory, young adults simulated falls from a wheelchair and performed other wheelchair-related mobility activities in a laboratory. Neural Network classifiers were used to train the algorithm developed based on data retrieved from accelerometers mounted at the participant’s wrist, chest, and head. Results indicate excellent accuracy to differentiate between falls and wheelchair mobility activities. The sensors mounted at the wrist, chest, and head presented with an accuracy of 100%, 96.9%, and 94.8%, respectively, using data from 258 falls and 220 wheelchair mobility activities. This pilot study indicates that a fall detection algorithm developed in a laboratory setting based on fall accelerometer patterns can accurately differentiate wheelchair-related falls and wheelchair mobility activities. This algorithm should be integrated into a wrist-worn devices and tested among individuals who use a wheelchair in the community.
AB - Automated fall detection devices for individuals who use wheelchairs to minimize the consequences of falls are lacking. This study aimed to develop and train a fall detection algorithm to differentiate falls from wheelchair mobility activities using machine learning techniques. Thirty, healthy, ambulatory, young adults simulated falls from a wheelchair and performed other wheelchair-related mobility activities in a laboratory. Neural Network classifiers were used to train the algorithm developed based on data retrieved from accelerometers mounted at the participant’s wrist, chest, and head. Results indicate excellent accuracy to differentiate between falls and wheelchair mobility activities. The sensors mounted at the wrist, chest, and head presented with an accuracy of 100%, 96.9%, and 94.8%, respectively, using data from 258 falls and 220 wheelchair mobility activities. This pilot study indicates that a fall detection algorithm developed in a laboratory setting based on fall accelerometer patterns can accurately differentiate wheelchair-related falls and wheelchair mobility activities. This algorithm should be integrated into a wrist-worn devices and tested among individuals who use a wheelchair in the community.
KW - accidental falls
KW - activity recognition
KW - fall detection
KW - wearable sensor
KW - wheelchair
UR - http://www.scopus.com/inward/record.url?scp=85149426491&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85149426491&partnerID=8YFLogxK
U2 - 10.1080/10400435.2023.2177775
DO - 10.1080/10400435.2023.2177775
M3 - Article
C2 - 36749900
AN - SCOPUS:85149426491
SN - 1040-0435
VL - 35
SP - 523
EP - 531
JO - Assistive Technology
JF - Assistive Technology
IS - 6
ER -