Research on vehicle monitoring and early warning in expressway work zones based on video recognition

Expressway work zones contain numerous potential safety hazards, and traffic conditions are an important factor affecting work-zone safety. To support real-time traffic monitoring and early warning, this paper proposes a video recognition-based method for vehicle detection, tracking, traffic-flow statistics, and speed estimation in expressway work zones. First, based on a constructed traffic video dataset, a scenario-oriented improved YOLOv8 vehicle detection algorithm is developed. FasterNet is introduced to reconstruct the backbone network and reduce parameter redundancy, a P2 small-object detection head is added to improve the detection capability for distant vehicles, and MPDIoU Loss is adopted to enhance bounding-box regression accuracy under dense traffic and partial occlusion conditions. Ablation experiments show that the proposed method improves vehicle detection accuracy and small-object detection performance in expressway work-zone scenarios. Compared with the original YOLOv8n model, the improved model increases mAP@0.5 from 96.2% to 98.1% and reduces the number of parameters from 3.01 M to 1.66 M. Although the introduction of the P2 detection head slightly increases FLOPs from 8.1 G to 9.1 G, the model achieves better detection performance while maintaining a relatively low parameter count, which improves its applicability to work-zone vehicle monitoring tasks. Subsequently, the improved YOLOv8 model is integrated with the DeepSORT tracking algorithm to realize vehicle tracking, traffic-flow statistics, and speed estimation. The speed estimation results are compared with manually calibrated vehicle speeds, and the average relative error is 6.4%, indicating that the proposed method can provide preliminary speed input for warning decision-making. Finally, a vehicle monitoring and early warning framework for expressway work zones is established by combining video-based perception results with dynamic speed-limit calculation and safe braking distance analysis. The results provide preliminary technical support for intelligent traffic safety management in expressway work zones.