Urban Fatigue Driving Prediction With Federated Learning

Fatigue driving results in a great damage to road safety. Therefore, monitoring the fatigue driving is essential to protect the traffic participants. In reality, fatigue driving behavior on highways is simply defined by driving time, while the measurement of fatigue driving in cities is not clear. It is difficult to monitor fatigue driving in urban areas in real time. In this paper, we propose a clear criterion for determining urban fatigue driving behavior. The criterion integrates the driver’s current driving status and objective factors on the road. To process a large number of continuous vehicle trajectories in real time, we propose a distributed paradigm based on a cluster of servers. In addition, we use federal learning in our experiments for fatigue driving prediction while protecting user privacy. Finally, we confirm the performance of our proposal in real data published by DiDi.