A Secure Object Detection Technique for Intelligent Transportation Systems

A hybrid privacy-preserving algorithm integrating Pre-Aggregation Similarity Measurement (PA-SM) and Differential Privacy (DP) is developed to address security and privacy concerns in Federated Learning (FL) for Intelligent Transportation Systems (ITS). This approach effectively protects against both data poisoning-based model replacement and inference attacks, ensuring the integrity of the training process while preserving model performance. Evaluated on the CIFAR-10 and LISA traffic light datasets, the solution demonstrates a robust defense against adversarial attacks with minimal performance degradation, thereby enhancing the security and resilience of autonomous vehicles and ITS infrastructures against potential cyber threats.