Deep Learning and IoT for Personalized Health Tracking

In this work, an innovative loT system for long-term personalized monitoring of the activities performed by a person at home is proposed. The system integrates a Wi-Fi wearable sensor and feature extraction techniques to give information on a number of activities with the aim to infer abnormal behaviors. The approach presented has been conceived to be extended to systems requiring multiple wearable sensors giving information in a personalized manner. The activity classification has been performed with a relatively small training set. This result is interesting because it shows the possibility to implement, quite easily, different HAR systems calibrated on different classes of problems for age groups of people. The presented system architecture exploits on-board Wi-Fi connectivity and cloud computing to ensure a constantly update of the network with new training sets when users are added. To this purpose every data sample acquired by the sensor is transferred to the cloud. The system architecture designed open the door to an alternative approach that could take advantage on the use of FPGA technologies for the implementation of complex signal processing systems to produce.