Federated Learning Platform for Secure Object Recognition in Connected and Autonomous Vehicles

Integrating smart technologies in vehicles has brought rise to connected and autonomous vehicles (CAVs). One of the services impacted by this paradigm shift is driving assistance. Most systems use learning-based approaches such as object recognition to improve transportation quality. So, these services require exchanging and sharing data along the CAV network, which raises security issues. Within this work is a federated learning (FL)-based platform as a step towards secure CAV systems. It uses FL to secure client data locally and alleviate pressure on CAV servers and services against targeted attacks. A testbed evaluates the platform's feasibility in preserving the integrity of its implemented classifier while keeping a level of security of its training data. According to experimental results, the FL-based implementation can maintain the integrity of the classifier's accuracy even after introducing its distributive scheme. Also, security evaluations present the benefits of FL reinforcing network security and improving client data privacy. Based on the results, the proposed platform proves its feasibility as an integrity-preserving and secure option for object recognition-based driving assistance services in CAVs.