TY - GEN
T1 - Virtual IMU Data Augmentation by Spring-Joint Model for Motion Exercises Recognition without Using Real Data
AU - Xia, Chengshuo
AU - Sugiura, Yuta
N1 - Funding Information:
Thiswork was supported by JST PRESTOGrant Number JPMJPR2134
Funding Information:
This work was supported by JST PRESTO Grant Number JPMJPR2134.
Publisher Copyright:
© 2022 ACM.
PY - 2022/9/11
Y1 - 2022/9/11
N2 - A conventional motion exercises recognition system only tracks designated motion types, and it enables users cannot use a customized system according to personal needs. The virtual IMU data provides a new opportunity to reduce the cost of training datasets and flexibly design the activity recognition system using online resources. To better design a user-customized motion exercises recognition system using virtual IMU data, this paper proposes a virtual IMU sensor module with a spring-joint model to augment the virtual acceleration signal from the limited online 2D video. The original virtual acceleration signal is extended with data from different acceleration distributions generated by the spring-joint model and used to train a motion exercises recognition system. The proposed method can design a classifier for three motions with limited video resources, showing an average accuracy of 85.5 on the real motion data of seven individuals.
AB - A conventional motion exercises recognition system only tracks designated motion types, and it enables users cannot use a customized system according to personal needs. The virtual IMU data provides a new opportunity to reduce the cost of training datasets and flexibly design the activity recognition system using online resources. To better design a user-customized motion exercises recognition system using virtual IMU data, this paper proposes a virtual IMU sensor module with a spring-joint model to augment the virtual acceleration signal from the limited online 2D video. The original virtual acceleration signal is extended with data from different acceleration distributions generated by the spring-joint model and used to train a motion exercises recognition system. The proposed method can design a classifier for three motions with limited video resources, showing an average accuracy of 85.5 on the real motion data of seven individuals.
KW - Data Augmentation
KW - Motion Exercises Recognition
KW - Virtual IMU
UR - http://www.scopus.com/inward/record.url?scp=85145881131&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85145881131&partnerID=8YFLogxK
U2 - 10.1145/3544794.3558460
DO - 10.1145/3544794.3558460
M3 - Conference contribution
AN - SCOPUS:85145881131
T3 - Proceedings - International Symposium on Wearable Computers, ISWC
SP - 79
EP - 83
BT - ISWC 2022 - Proceedings of the 2022 ACM International Symposium on Wearable Computers
PB - Association for Computing Machinery
T2 - 2022 ACM International Symposium on Wearable Computers, ISWC 2022
Y2 - 11 September 2022 through 15 September 2022
ER -