Towards the Recognition of Human Activities from Smartwatch Data

If we could give every individual the right amount of nourishment and exercise, not too little and not too much, we would have found the safest way to health.

— Hippocrates

Hippocrates of Kos, an ancient Greek physician often referred to as the Father of Medicine, believed that diseases could occur with idleness, excessive exercise, or overpowering food consumption in comparison to exercise. His theory therefore highlighted the importance of physical activity towards a healthy life.

Technology has undoubtedly improved many aspects of our quality of life. Reinforced by the effects of the worldwide pandemic we are currently living in and the measures implemented to control the spread of COVID-19, technology has allowed us to work remotely and to keep in touch with our beloved ones despite the social distancing. Nonetheless, teleworking forced us to set up improvised offices in kitchens or living rooms, avoiding the need to commute daily. This new situation has dramatically reduced our amount of physical exercise.

According to the World Health Organisation (WHO), 60 to 85% of people from developed and developing countries lead sedentary lifestyles. Physical inactivity is a serious problem, as it can lead to major illnesses, such as cardiovascular or pulmonary diseases, cancer, diabetes, depression, or obesity. Thus, although it is insufficiently addressed, physical inactivity is one of the most serious public health concerns. To tackle this issue, we envision the use of systems based on Information Technology (IT) and powered with Artificial Intelligence (AI) to develop new digital tools that can help monitor users’ physical activity. Both human and virtual personal trainers or recommendation systems could then exploit this information to detect deficits in the physical exercise, and to engage users to follow healthier, more dynamic lifestyles.

In the last few years, the field of Human Activity Recognition (HAR) has attracted significant and growing interests due to its broad range of applications in areas such as healthcare, fitness, athletics, or elderly care, to name but a few. Initial works were performed exploring the measurements collected with sensors embedded in smartphones. However, the location of the smartphone is critical to guarantee capturing enough body information. Because of their location in the wrist and the comfort wearing them, smartwatches are a promising device for this task. In this regard, we aim to model human activities from the measurements collected with the sensors embedded in the smartwatch included in the IoT infrastructure of the sustAGE system.

In sustAGE, we are collecting a HAR dataset in which we ask participants to lie, stand, sit, walk, run, climb stairs, and cycle. Analysing the accelerometer measurements in the x-, y-, and z-axes collected with a customised smartwatch app and using AI techniques, we will power the sustAGE system with a functionality to monitor the activities users are doing. This information could enrich the decision-making processes of sustAGE to improve, personalise, and better adjust the physical activity-related recommendations provided by the system to the individual needs and progress of each user. Furthermore, in our attempt to contribute with technology to prevent COVID-19, our dataset also includes washing hands samples, so our models can recognise this activity. This way, the sustAGE system could remind users to wash their hands when they have not done it for a while. If you are curious about the performance of our HAR models, stay tuned to the outcomes of sustAGE