Autonomous vehicles rely on real-time and efficient object detection systems to ensure safe and reliable navigation. Our paper presents Swin-YOLOv11, an enhanced object detection model integrated within an edge- cloud cooperative framework to improve accuracy and computational efficiency in autonomous driving scenarios. Unlike conventional YOLO models, Swin-YOLOv11 incorporates Swin Transformer blocks in place of the C3K2 module, leveraging hierarchical feature extraction and self-attention mechanisms to enhance long-range dependency modeling. Experimental results demonstrate that Swin-YOLOv11 surpasses YOLOv8, YOLOv10, and YOLOv11 (C3K2), achieving improved precision, recall, and mean average precision (mAP50 of 51.2 %), while reducing computational overhead (loss of 1.21). Additionally, the model is optimized for edge deployment through pruning, quantization, and knowledge distillation, ensuring efficient performance in resource-constrained environments. A dynamic task offloading strategy is also introduced to balance computational loads between edge devices and cloud resources, enhancing adaptability in real-world conditions. Our proposed system demonstrates superior detection capabilities and robustness, making it a viable solution for real-time perception in autonomous vehicles.

Autonomous Vehicles,YOLOv11, Swin-Transformer, Edge-cloud