Federated Learning for Driver Status Monitoring

We apply a novel concept for distributed learning to the problem of driver status monitoring. The main benefit is that only local, in-vehicle training data is used, thus privacy sensitive pictures of the driver do not leave the vehicle. We show the challenges of this application, in particular in the distribution of the data and apply different, recent techniques of federated learning. In result, we show for our data set that federated learning can achieve almost same performance as classical learning, where all data is collected and processed in a single learning applications.