Related Experiment Video
Updated: Dec 31, 2025
08:05
Design and Analysis for Fall Detection System Simplification
Published on: April 6, 2020
11.1K
Automatic Classification of Squat Posture Using Inertial Sensors: Deep Learning Approach.
Jaehyun Lee1, Hyosung Joo1, Junglyeon Lee2
1Interdisciplinary Program of Medical and Biological Engineering, University of Ulsan, 93, Daehak-ro, Nam-gu, Ulsan 44610, Korea.
Sensors (Basel, Switzerland)
|January 16, 2020
Summary
Deep learning significantly improves squat exercise form classification accuracy using wearable sensors compared to conventional methods. Optimal single-sensor placement for self-fitness is the right thigh.
Area of Science:
- Biomechanics and Exercise Science
- Machine Learning in Sports Technology
- Wearable Sensor Applications
Background:
- Inexperienced exercisers risk injury during core exercises like squats due to poor form without expert coaching.
- Current wearable sensor and algorithm implementations for exercise classification lack sufficient accuracy.
- Accurate, accessible methods for monitoring exercise form are needed to prevent injuries.
Purpose of the Study:
- To compare the squat posture classification performance of deep learning versus conventional machine learning.
- To determine the optimal placement of inertial measurement units (IMUs) for accurate exercise classification.
- To evaluate the feasibility of using minimal sensors for self-fitness applications.
Main Methods:
- Collected accelerometer and gyroscope data from 39 participants using five IMUs (thighs, calves, lumbar).
- Participants performed correct and incorrect squat variations.
- Compared classification accuracies of deep learning and conventional machine learning using one, two, or five IMUs.
Main Results:
- Deep learning achieved 91.7% accuracy with five IMUs, significantly outperforming conventional machine learning (75.4%).
- Using two IMUs, deep learning on the right thigh and calf yielded 88.7% accuracy.
- A single IMU on the right thigh achieved 80.9% accuracy with deep learning, outperforming conventional methods (58.7%).
Conclusions:
- Deep learning models offer superior accuracy for squat posture classification compared to conventional machine learning.
- A single IMU on the right thigh represents the most practical and accurate solution for self-fitness squat monitoring.
- These findings can inform the development of accessible injury prevention tools for exercisers.
