Abstract
The increasing complexity of urban environments, characterised by diverse mix of vehicles, and vulnerable road users such as pedestrians, and cyclists, poses significant safety and efficiency challenges for modern transportation. Intelligent Transportation Systems (ITS) aim to mitigate these issues, but the existing centralised models for road user detection face critical limitations regarding data privacy, latency, and scalability. Federated Learning (FL) has emerged as a promising decentralized paradigm to overcome these challenges by training models collaboratively without sharing raw data. This review followed the systematic literature review (SLR) methodology, with a comprehensive search of academic and scholarly databases and systematically analysed the current state of research on the application of Federated Learning in ITS for detecting mixed road users. Also examined in this work is the evolution of intelligent transport systems from centralized to decentralized and edge-based detection architectures, analyses the principles and advantages of Federated Learning, and explore the key scholarly debates surrounding its implementation, including the trade-offs between privacy and performance, data heterogeneity, and the gap between simulation and real-world deployment. This review highlights significant research gaps in the application of Federated Learning specifically for the task detection and classifications of vulnerable road user (VRU) in a mixed-user transport setting. The potential of Federated learning to enhance road safety through timely alert mechanism was explored, and the recommendations included that future research should consider among other things the creation of federated learning frameworks that are able to detect and classify different road users in a mixed road setting.