Accurate Body Pose Matching for Individuals with Stroke using Siamese Networks

Stroke is one of the major causes of long-term disability in United States. With more than 800,00 people experiencing stroke every year, it is important that efficient means for recovery are presented to support stroke subjects. Exoskeleton and serious game based rehabilitation are some of the state...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:IEEE/ACM Conference on Connected Health: Applications, Systems and Engineering Technologies (Online) s. 177 - 181
Hlavní autoři: Gokhman, Ruslan, Sawdayi, Talya, Khan, Rana, Satyanarayana, Ashwin, Vinjamuri, Ramana, Kadiyala, Sai Praveen
Médium: Konferenční příspěvek
Jazyk:angličtina
Vydáno: IEEE 19.06.2024
Témata:
Accuracy
Exoskeletons
Games
Machine learning
Machine learning algorithms
Pose Matching
Siamese Networks
Stroke (medical condition)
Stroke Rehabilitation
ISSN:2832-2975
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Popis
Shrnutí:Stroke is one of the major causes of long-term disability in United States. With more than 800,00 people experiencing stroke every year, it is important that efficient means for recovery are presented to support stroke subjects. Exoskeleton and serious game based rehabilitation are some of the state-of-art approaches used in the recovery of stroke subjects. Accurate matching of body poses performed by individuals with stroke is essential in understanding the current state of recovery of the subject and plan further rehabilitation. Established machine learning based approaches fall short in accurately matching the poses of stroke subjects with ground truths. In this work, we present algorithms supported by Siamese architectures to effectively identify the poses performed by the subjects. Our proposed framework involves data pre-processing, extraction, building classification models and validating them using a body pose data set of individuals with stroke. On a considered public database, our proposed pose identification models namely, Siamese based LSTM and Siamese based CNN gave 7.8% and 14.2 % better identification accuracy than the traditional LSTM approach.
ISSN:2832-2975
DOI:10.1109/CHASE60773.2024.00030