Metabolic energy expenditure estimation using a position-agnostic wearable sensor system

This paper presents an energy expenditure estimation method that uses a wearable accelerometer sensor, but does not rely on a priori knowledge about the location of the sensor. The sensor can be worn at any of three pre-defined locations, namely, right wrist, right thigh and right ankle. It is shown that once the system is trained, the proposed mechanism can perform sensor location detection in run-time within a period as short as 4.2 seconds. A sensor-position-specific energy expenditure estimation model is then applied. Experiments were carried out on 25 healthy subjects, and 14 activities with diverse intensities were included. It is demonstrated that the sensor position can be detected at an accuracy of 99.68%, and the Root-Mean-Square-Error for energy expenditure estimation is 1.79 METs (Metabolic Equivalent of Task).