This work proposes to combine wearables and the use of artificial intelligence techniques to detect manual mobility patterns caused by brushing teeth. Specifically, the article describes and evaluate a system based on convolutional neural networks, able to identify brushing gestures from samples of few seconds of inertial accelerometry signals gathered by wrist-worn devices. The architecture is systematically trained and validated with the signals provided by different public databases which were collected when different experimental subjects executed different manual actions. The results show the effectiveness of the detector, since it reaches a sensitivity and specificity greater than 95% when applied to discriminate brushing from other hand actions. In addition, the system is re-trained and assessed with the real-life samples captured by a smartwatch, where the neural model could be implemented to operate and produce real-time decisions.
Buscar en Dimension