A Privacy-Preserving and Efficient Driver Recognition Framework for Sustainable ITS Using DRL and FL

In Intelligent Transportation Systems (ITS), driver recognition presents challenges of data privacy, computational efficiency, and energy consumption. Optimizing energy use in ITS has become crucial with the rise of environmentally conscious Information and Communication Technology (ICT) practices. Therefore, this paper introduces a privacy-preserving and energy-efficient task offloading strategy using Deep Reinforcement Learning (DRL) and Federated Learning (FL) within a network leveraging Smart Traffic Cameras (STCs) for edge computing. Initially, the public transports employ a DRL-based strategy to offload tasks to STCs, optimizing network resources and minimizing energy use. At the second phase, allows private vehicles to train models locally, offloading only model parameters, thus ensuring data privacy and reducing communication energy costs. Finally, aggregates these parameters at a central cloud server, refining a Network-Wide Model (NWM). The proposed framework enhances model performance, preserves privacy, and improves computational efficiency, reducing the carbon footprint of ITS operations. Simulations demonstrate that DRL's Actor-Critic Algorithm (ACA) reduces task latency and energy consumption while FL ensures efficient model training with minimal communication overhead.