Depth-Camera Based Energy Expenditure Estimation System for Physical Activity Using Posture Classification Algorithm

Area of Science:

• Biomechanics
• Computer Vision
• Health Informatics

Background:

• Modern lifestyles often involve insufficient physical activity, impacting overall health.
• Accurate estimation of energy expenditure (EE) is crucial for developing effective exercise plans.
• Existing methods for EE estimation have limitations that this study aims to address.

Purpose of the Study:

• To propose an accurate energy expenditure (EE) estimation model utilizing depth camera data and physical activity classification.
• To determine the optimal number and placement of depth cameras for EE estimation.
• To compare the performance of different machine learning models for EE estimation.

Main Methods:

• Three depth cameras were positioned at side, rear side, and rear views to capture kinematic data.
• Support Vector Machine (SVM) was employed for physical activity classification.
• Linear regression, Multilayer Perceptron (MLP), and Convolutional Neural Network (CNN) models were evaluated for EE estimation.

Main Results:

• Optimal EE estimation with a single depth camera (side view) was achieved using the MLP model (MAE: 0.55, MSE: 0.66, RMSE: 0.81).
• For higher accuracy, two cameras (side and rear views) combined with CNN for light-to-moderate activities and MLP for vigorous activities yielded improved results.
• Specific RMSEs for standing, walking, and running were 0.19, 0.57, and 0.96, respectively, using the optimized camera and model configurations.

Conclusions:

• The study demonstrates that optimal energy expenditure estimation can be achieved by selecting appropriate machine learning models and camera configurations based on accuracy requirements.
• This research is the first to investigate the impact of different camera setups and models on EE estimation accuracy.
• The findings provide a foundation for developing more personalized and effective exercise and lifestyle management strategies.