Human Activity Recognition on Embedded Devices: CNN-Based Walking Detection and Step Counting

Recognizing human activities purely from movement patterns has many applications in healthcare and fitness tracking. Although machine learning models that achieve high accuracy have been developed, their use in resource-constrained embedded devices remains challenging. This paper reviews the state of the art in Human Activity Recognition (HAR) and presents a CNN designed to detect walking trained on a large, publicly available, free-living dataset. The model is also used together with an open source step counting algorithm in order to measure its accuracy on unseen data.