| Summary: | 碩士 === 國立臺北大學 === 資訊工程學系 === 105 === Monitoring the energy expenditure (EE) is a crucial factor for people tracking their physical activity (PA). People can avoid the obesity and reduce the risk of chronic diseases by monitoring their EEs. In this study, a depth camera-based system for EE estimation of PA in gym is proposed. Most of the previous studies utilize inertial measurement unit (IMU) to estimate EE. Compared to the IMU-based systems, the proposal system can provide a more convenient way to monitor subjects’ treadmill workouts in gym without wearing any devices. In this study, 21 participants were recruited to join the experiment. The subjects’ skeletal data acquired from depth camera was along with the oxygen consumption simultaneously obtained from K4b2 to establish an EE predictive model. In order to obtain a robust EE estimation model, this study adopted to place three depth cameras in side, rear side, and rear views. By comparing the different predictive models and different camera locations, the results show that the multilayer perceptron (MLP) is the best predictive model in this approach, and placing camera in rear view can obtain the best performance in EE estimation. The measured and predicted metabolic equivalent of tasks (METs) are strongly positive correlated r = 0.93, and the coefficient of determination r2 = 0.86. The mean absolute error (MAE) = 0.61, the mean squared error (MSE) = 0.67, and the root mean squared error (RMSE) = 0.76. According to the results, this research can provide a handy and reliable system for monitoring user’s EE of performing the treadmill workouts.
|
|---|
