Squat Angle Assessment Through Tracking Body Movements

Mohd Asyraf Zulkifley, Nur Ayuni Mohamed, Nuraisyah Hani Zulkifley

Research output: Contribution to journalArticle

Abstract

Squat exercise is frequently used in physiotherapy rehabilitation for stroke patients. In the early stage of rehabilitation, patients are urged to avoid performing any deep squat as the strains on tendon and ligament are much higher compared to the half-squat exercise. Therefore, it is important for patients to be aware of their squat depth. One of the ways to measure squat depth is by using a wearable device which adds unnecessary weight to the patients and makes them feel uncomfortable. Thus, we propose a single camera system that captures video from the frontal view to measure the squat angle continuously according to the number of frames per second. The system will provide knee angle measurements for every frame taken based on a combined approach of deep learning tracking and deep belief networks regressor. The proposed system requires just a bounding box input of the whole test subject taken at an upright position, which will later be the input to a convolutional neural networks-based tracker. Both the head and upper body parts of the exerciser will be tracked independently. The resultant tracked points will be normalized with the test subject height to find the ratio of height to the corresponding points. The ratio features are then will be the input to multiple deep belief networks' regressors to predict the knee angle. The mean of ratio features will be used to segregate the input frame into its respective regressor. The experimental results show that the system produces the lowest mean error angle of 8.64° based on the setup of five regressors with each of them consists of five hidden layers. Hence, it is suitable to be implemented in a squat angle monitoring system to notify the patients of their squat angle depth.

Original languageEnglish
Article number8686074
Pages (from-to)48635-48644
Number of pages10
JournalIEEE Access
Volume7
DOIs
Publication statusPublished - 1 Jan 2019

Fingerprint

Bayesian networks
Patient rehabilitation
Physical therapy
Ligaments
Tendons
Angle measurement
Cameras
Neural networks
Monitoring
Deep learning

Keywords

  • Deep belief networks regressor
  • Physiotherapy monitoring
  • Squat angle analysis
  • Visual object tracking

ASJC Scopus subject areas

  • Computer Science(all)
  • Materials Science(all)
  • Engineering(all)

Cite this

Squat Angle Assessment Through Tracking Body Movements. / Zulkifley, Mohd Asyraf; Mohamed, Nur Ayuni; Zulkifley, Nuraisyah Hani.

In: IEEE Access, Vol. 7, 8686074, 01.01.2019, p. 48635-48644.

Research output: Contribution to journalArticle

Zulkifley, Mohd Asyraf ; Mohamed, Nur Ayuni ; Zulkifley, Nuraisyah Hani. / Squat Angle Assessment Through Tracking Body Movements. In: IEEE Access. 2019 ; Vol. 7. pp. 48635-48644.
@article{7fea892b39014eefaae2531624d135e9,
title = "Squat Angle Assessment Through Tracking Body Movements",
abstract = "Squat exercise is frequently used in physiotherapy rehabilitation for stroke patients. In the early stage of rehabilitation, patients are urged to avoid performing any deep squat as the strains on tendon and ligament are much higher compared to the half-squat exercise. Therefore, it is important for patients to be aware of their squat depth. One of the ways to measure squat depth is by using a wearable device which adds unnecessary weight to the patients and makes them feel uncomfortable. Thus, we propose a single camera system that captures video from the frontal view to measure the squat angle continuously according to the number of frames per second. The system will provide knee angle measurements for every frame taken based on a combined approach of deep learning tracking and deep belief networks regressor. The proposed system requires just a bounding box input of the whole test subject taken at an upright position, which will later be the input to a convolutional neural networks-based tracker. Both the head and upper body parts of the exerciser will be tracked independently. The resultant tracked points will be normalized with the test subject height to find the ratio of height to the corresponding points. The ratio features are then will be the input to multiple deep belief networks' regressors to predict the knee angle. The mean of ratio features will be used to segregate the input frame into its respective regressor. The experimental results show that the system produces the lowest mean error angle of 8.64° based on the setup of five regressors with each of them consists of five hidden layers. Hence, it is suitable to be implemented in a squat angle monitoring system to notify the patients of their squat angle depth.",
keywords = "Deep belief networks regressor, Physiotherapy monitoring, Squat angle analysis, Visual object tracking",
author = "Zulkifley, {Mohd Asyraf} and Mohamed, {Nur Ayuni} and Zulkifley, {Nuraisyah Hani}",
year = "2019",
month = "1",
day = "1",
doi = "10.1109/ACCESS.2019.2910297",
language = "English",
volume = "7",
pages = "48635--48644",
journal = "IEEE Access",
issn = "2169-3536",
publisher = "Institute of Electrical and Electronics Engineers Inc.",

}

TY - JOUR

T1 - Squat Angle Assessment Through Tracking Body Movements

AU - Zulkifley, Mohd Asyraf

AU - Mohamed, Nur Ayuni

AU - Zulkifley, Nuraisyah Hani

PY - 2019/1/1

Y1 - 2019/1/1

N2 - Squat exercise is frequently used in physiotherapy rehabilitation for stroke patients. In the early stage of rehabilitation, patients are urged to avoid performing any deep squat as the strains on tendon and ligament are much higher compared to the half-squat exercise. Therefore, it is important for patients to be aware of their squat depth. One of the ways to measure squat depth is by using a wearable device which adds unnecessary weight to the patients and makes them feel uncomfortable. Thus, we propose a single camera system that captures video from the frontal view to measure the squat angle continuously according to the number of frames per second. The system will provide knee angle measurements for every frame taken based on a combined approach of deep learning tracking and deep belief networks regressor. The proposed system requires just a bounding box input of the whole test subject taken at an upright position, which will later be the input to a convolutional neural networks-based tracker. Both the head and upper body parts of the exerciser will be tracked independently. The resultant tracked points will be normalized with the test subject height to find the ratio of height to the corresponding points. The ratio features are then will be the input to multiple deep belief networks' regressors to predict the knee angle. The mean of ratio features will be used to segregate the input frame into its respective regressor. The experimental results show that the system produces the lowest mean error angle of 8.64° based on the setup of five regressors with each of them consists of five hidden layers. Hence, it is suitable to be implemented in a squat angle monitoring system to notify the patients of their squat angle depth.

AB - Squat exercise is frequently used in physiotherapy rehabilitation for stroke patients. In the early stage of rehabilitation, patients are urged to avoid performing any deep squat as the strains on tendon and ligament are much higher compared to the half-squat exercise. Therefore, it is important for patients to be aware of their squat depth. One of the ways to measure squat depth is by using a wearable device which adds unnecessary weight to the patients and makes them feel uncomfortable. Thus, we propose a single camera system that captures video from the frontal view to measure the squat angle continuously according to the number of frames per second. The system will provide knee angle measurements for every frame taken based on a combined approach of deep learning tracking and deep belief networks regressor. The proposed system requires just a bounding box input of the whole test subject taken at an upright position, which will later be the input to a convolutional neural networks-based tracker. Both the head and upper body parts of the exerciser will be tracked independently. The resultant tracked points will be normalized with the test subject height to find the ratio of height to the corresponding points. The ratio features are then will be the input to multiple deep belief networks' regressors to predict the knee angle. The mean of ratio features will be used to segregate the input frame into its respective regressor. The experimental results show that the system produces the lowest mean error angle of 8.64° based on the setup of five regressors with each of them consists of five hidden layers. Hence, it is suitable to be implemented in a squat angle monitoring system to notify the patients of their squat angle depth.

KW - Deep belief networks regressor

KW - Physiotherapy monitoring

KW - Squat angle analysis

KW - Visual object tracking

UR - http://www.scopus.com/inward/record.url?scp=85065100653&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85065100653&partnerID=8YFLogxK

U2 - 10.1109/ACCESS.2019.2910297

DO - 10.1109/ACCESS.2019.2910297

M3 - Article

VL - 7

SP - 48635

EP - 48644

JO - IEEE Access

JF - IEEE Access

SN - 2169-3536

M1 - 8686074

ER -