Low-Power Sound-Based User Activity Recognition

Tiny computers and sensors integrated into our clothes are likely to have a major impact on our life in the future. By monitoring our physical condition and activities they will allow to provide us assistance with everyday tasks or warn us of dangerous situations: They will recognize that we are running to catch a train and might tell us that we don't need to hurry because the train is delayed. Or they will follow the work steps of a maintenance worker and warn him in case he is about to make a hazardous mistake. In this work, we show that with the help of an acoustic analysis of the environment, we can recognize everday tasks, be it typing on a keyboard, brushing teeth or closing a door. Furthermore, we show that the signal processing algorithms can be designed that these sensors operate for months or years on a miniature battery.