A Novel Edge-Cloud Cooperative System Utilizing Swin-YOLOv11 for Autonomous Vehicle

Autonomous vehicles require real-time, accurate, and computationally efficient object detection systems to ensure safe navigation. This paper introduces Swin-YOLOv11, an advanced object detection model integrated into an edgecloud collaborative framework designed to optimize both accuracy and computational efficiency for autonomous driving applications. Unlike conventional YOLO architectures, Swin- YOLOv11 replaces the C3K2 module within the YOLOv11 backbone with Swin Transformer blocks, enhancing the model's ability to capture long-range dependencies through hierarchical feature extraction and self-attention mechanisms. Experimental evaluations demonstrate that Swin-YOLOv11 outperforms existing models, including YOLOv8, YOLOv10, and YOLOv11 (C3K2), achieving superior precision, recall, and a mean Average Precision (mAP50) of 51.2% while minimizing computational loss to 1.21. Additionally, the model is optimized for deployment in resource-constrained environments through techniques such as pruning, quantization, and knowledge distillation. The results highlight Swin-YOLOv11 as a robust and efficient solution for real-time object detection in autonomous vehicles, offering improved detection accuracy and operational efficiency in dynamic driving environments.