Enhancing Privacy and Performance in V2P Systems through Decentralized Federated Learning
Rihab Hmaied, Takoua Kefi, Ryma Abassi · 2025
In spite of the substantial development that faced Intelligent Transportation Systems (ITS) and, particularly, Vehicle-to-Pedestrian (V2P) communication systems, they still confront several challenges. These challenges are not only related to performance optimization, but also to privacy preservation. In fact, traditional centralized machine learning approaches for V2P exacerbate this privacy risk (1) by exposing the pedestrian data to privacy risks such as unauthorized tracking and data misuse, and (2) by intermingling with Noisy Background Information (NBI) real data.This paper presents RingFL-V2P, a decentralized Federated Learning framework that protects pedestrian data during model training. Instead of relying on a central server, RingFL-V2P uses a ring structure. Each device trains the model locally and then passes the updated model to the next device. This method keeps personal information safe while allowing devices to learn together.A comparative evaluation was carried on NuScenes-KITTI datasets in order to scrutinize the effectiveness of the proposed approach in the context of V2P communication systems. Results highlight that RingFL-V2P maintains competitive performance while ensuring privacy preservation.