Computation Efficient ECG Classification on Resource Constrained Devices

Wearable sensors and the plethora of Internet of Things devices are revolutionizing several aspects of everyday lives. In this domain, health monitoring applications are raising interest, thanks to their ability to track the vital parameters of the user wearing the device, and recognizing in advance potential issues health. Most of these solutions often require an internet connection to offload the data to an edge server, although this may not always be present, or use highly complex models which do not fit on constrained wearable devices. In this paper we propose a novel algorithm which tracks simple features in the ECG signal locally to the wearable device, with a lower memory footprint and computation resources needed compared to other proposal. Our extensive performance evaluation and comparison with the state of the art confirms the viability of our approach, as our proposal achieves more than 99% in accuracy on average.