The advancement of autonomous vehicle technology relies heavily on sophisticated machine-learning models that facilitate real-time object detection and classification. To address this requirement, we propose VINO_EffiFedAV, a novel federated learning framework specifically designed for autonomous vehicles. By leveraging NVIDIA T4 GPUs and integrating camera sensor data, this framework develops and trains customized YOLOv8 models directly within each vehicle, ensuring localized processing and enhanced data privacy and security. YOLOv8′s superior real-time performance, lightweight and scalable architecture, and improved accuracy in detecting small objects make it an ideal choice. Our approach effectively manages Non-Independent and Identically Distributed (Non-IID) data from diverse environmental conditions through an efficient selective client update mechanism integrated with the Federated Averaging (FedAvg) algorithm. This strategy strategically filters and aggregates client contributions, reducing the impact of outlier data and maintaining a robust global model. The system's versatility and effectiveness are validated across various datasets, including KITTI for IID conditions and Nuimages and Cityscapes for Non-IID scenarios, enhancing the model's ability to generalize across different driving environments. Notably, our framework reduces average communication cost by up to 50 % and 64 % and computational complexity by 72 % and 84 % for IID and Non-IID scenarios. Moreover, consistent latency performance ensures reliable real-time object detection, achieving impressive mean average precision scores of 84.3 % and 61.6 %. These results underscore the potential of VINO_EffiFedAV to enhance object detection systems in autonomous vehicles, contributing to safer and more efficient navigation.