<p>Traffic scene detection plays a crucial role in autonomous driving, with object detection being a fundamental task within this domain. However, deploying large models to in-vehicle platforms and achieving real-time detection in complex traffic scenarios is challenging. To address this issue, we propose an improved version of the YOLOv8 algorithm, named YOLO-EDGE. First, GhostConv is introduced in the backbone to replace several standard convolutions, thereby reducing model parameters. Second, a novel C2FEMA module is proposed, which maintains the lightweight advantages and captures richer gradient flow information while improving algorithm accuracy. Lastly, the neck of the original YOLOv8 network is improved by transforming the PAN-FPN structure into a more lightweight SlimNECK structure, which enhances feature information and further improves feature extraction capabilities. Experimental results on the PASCAL VOC dataset show that YOLO-EDGE reduces the model size by 7%, achieving a final size of only 20.9M compared to the baseline YOLOv8s. The number of parameters decreases by 7.6%, while the mean average precision (mAP@0.5) increases by 2.2% and the mAP@0.5:0.95 improves by 3.2%. These results demonstrate the effectiveness of YOLO-EDGE. Furthermore, YOLO-EDGE achieves a mean average precision (mAP@0.5) of 64.9% on the multi-object detection dataset UA-DETRAC, further demonstrating the algorithm’s versatility.</p>