Physical Activity Energy Expenditure Monitoring using COSMED K5, IMU, and HR

This dataset has been used in our manuscript entitled "PM-EKF: A Physiological Model-Based Extended Kalman Filter for Daily-Life Physical Activity Energy Expenditure Estimation". A total of 10 participants (30% female) were recruited for this study. Inclusion criteria were: (1) aged between 18 and 60; (2) have a Body Mass Index (BMI) lower than 40 kg/m2; (3) free of cardiovascular diseases, respiratory diseases, metabolic disorders; (4) not being pregnant for female participants; (5) free of physical disabilities that impact daily living. The study was ethically approved by the Ethics Committee of Computer \& Information Science of the University of Twente (File no. 230728). Informed consent was obtained from all individual participants included in the study. Static data, including age, sex, height, and weight, were collected. Body composition was estimated using a Bioelectrical Impedance Analysis (BIA) scale (Omron BF511). Activity Duration [s] Sitting resting 300 Sitting reading 300 3Standing still 180 Surveying (on a laptop) Not fixed Emptying dishwasher Not fixed Mopping Not fixed Stacking shelves with books Not fixed Treadmill (3 km/h) 300 Treadmill (5 km/h) 300 Climbing stairs (5 times) Not fixed Cycling at 125 watt 300 Each data collection session commenced with a 30-minute quiet rest in the supine position to estimate rest metabolic rate (RMR). A series of activities of daily living (ADL) followed this, as presented in the table above. Most activities were performed for at least 5 minutes to reach steady-state energy expenditure. For emptying the dishwasher, mopping, stacking shelves, climbing stairs, and surveying, the participants were instructed to execute the activities at their own pace, which made the duration of these activities variable. Each participant performed all activities in a randomized order to prevent the introduction of bias in the dataset. The randomization of the performed activity order and the variable duration of certain activities, as mentioned before, help to simulate real situations of daily living. IMU data was collected at 30 Hz at three body locations, i.e., left thigh, right thigh, and pelvis, using Movella Xsens DOT. Single-lead Electrocardiogram (ECG) data was collected using CardioBan Kit at 80 Hz. Breath-by-breath respiratory data, serving as ground truth, was collected using COSMED K5. Activity labels were manually annotated by the researchers based on video recordings from five fixed cameras installed throughout the eHealth House using the OMNIA COSMED software. To estimate the RMR from O2 consumption and CO2 production, the first 5 minutes of the RMR recording were discarded to remove transient effects following the onset of rest. The mean volumes of O2 and CO2 computed from the remaining period were taken as the participant’s RMR. Total energy expenditure (TEE) measured during ADL comprises both RMR and PAEE. Therefore, the RMR-contributed O2 and CO2 volumes were subtracted from the TEE to obtain the activity-related gas-exchange signals (PAEE = TEE – RMR). For multi-modal signal synchronization, the breath-by-breath O2 and CO2 volume data were resampled to 1 Hz and smoothed using a first-order Savitzky–Golay filter (20 s window) to reduce noise and mitigate artifacts from talking.