Enhancing Traffic Light Detection and Classification through Federated Learning

Traffic light detection stands as a pivotal challenge in the realm of connected and automated driving systems and intelligent traffic management. Traffic light detection lies in the real-time and accurate recognition of traffic lights and their states ’go’, ’warning’, ’stop’, ’goLeft’, ’warningLeft’, ’stopLeft’, even under diverse environmental conditions. This paper introduces a synthetic approach for intelligent traffic light detection and classification, namely FL-TLDC, that synergizes federated learning with Fast Region-based Convolutional Neural Networks (Faster R-CNN). The proposed collaborative and decentralised approach allows the collective improvement of traffic light detection algorithms without the need to share sensitive or proprietary data. Simulation results demonstrate that the proposed analysis achieves a high detection performance and provides fast responses.