Resumen
Daily physical activity is one of the key factors to improve health and support the prevention of a variety of chronic diseases e.g., hypertension, depression or acute events such as strokes. Self-monitoring by the patients has shown to improve adherence to care and thus leads to general improvement of health conditions. However, precise physical activity detection and quantification may involve heavy or expensive sensors and often-complex computations. Other types of bottlenecks, such as proprietary algorithms or machine learning methods, which often do not meet requirements of medical use cases due to a lack in transparency and requested levels of accuracy and robustness, motivated the work presented here. In this paper, we propose the adaptation of the Euclidean norm minus one method, which was already demonstrated as relevant for physical activity intensity discrimination. The main modification stands in the use of a gravity estimate to countervail imprecise sensors, which allows using the metric with low sampled wrist accelerometer data, collected with off-the-shelf smartwatches in daily live environments. As proof of concept, the proposed algorithm was evaluated on a reference data set acquired on healthy subjects. The method shows the ability to discriminate between low, moderate, and high intensity activities.