Paper Contents
Abstract
Autonomous vehicles (AVs) depend on accurate object detection for safe navigation, yet their performance can significantly degrade under low-light conditions, posing serious challenges for real-world deployment. Traditional centralized machine learning approaches require large-scale datasets, but they often face limitations related to data privacy, latency, and scalability. To overcome these issues, this research proposes a novel framework that integrates Federated Learning (FL) with YOLOv5a cutting-edge object detection modelto enhance detection performance in low-visibility environments.The proposed federated architecture enables collaborative training across multiple AVs and edge devices without sharing raw data, thereby preserving privacy and reducing communication overhead. YOLOv5 is further optimized for real-time applications through adaptive data augmentation, transfer learning, and model pruning, ensuring robust performance in nighttime and low-light scenarios.Experimental results show that our federated YOLOv5 model outperforms traditional centralized methods, achieving up to a 25% improvement in detection accuracy and a 35% reduction in false positives under low-light conditions. The system also maintains high computational efficiency and supports real-time inference, making it well-suited for large-scale deployment in AV networks.
Copyright
Copyright © 2025 Dr. Ranga Swamy Sirisati. This is an open access article distributed under the Creative Commons Attribution License.
