基于视频云联网技术的隧道车辆检测技术研究

Research on Tunnel Vehicle Detection Technology Based on Video Cloud Networking Technology

摘要: 本文针对高速公路隧道光照复杂、车辆密集及遮挡频繁等环境，研究了基于视频云联网的隧道车辆检测算法。为提升复杂环境下的检测精度，在YOLOv5主干网络中嵌入DBB模块，增强模型对多尺度车辆特征的提取与融合能力。其次，搭建隧道车辆视频云联网数据集，涵盖不同时段、不同隧道场景下的车辆样本。实验结果表明，检测精确度提升0.5个百分点，达到96.5%。mAP@0.5:0.95指标至80.0%。同时，模型GPU显存占用量降低36.1%，由7.56G降至4.83G。本研究为隧道车辆智能监测系统的高效部署提供了可行的技术方案。并验证了该方法在视频云联网环境下的工程应用可行性。

Abstract: Aiming at the complex environment of highway tunnels with intricate illumination, dense vehicles and frequent occlusion, this paper studies a tunnel vehicle detection algorithm based on video cloud networking technology. To improve the detection accuracy in complex scenarios, the DBB module is embedded into the backbone network of YOLOv5, which enhances the model’s capability of multi-scale vehicle feature extraction and fusion. Secondly, a tunnel vehicle video cloud networking dataset is established, covering vehicle samples collected from different time periods and various tunnel scenarios. Experimental results show that the detection precision is increased by 0.5 percentage points, reaching 96.5%, and the mAP@0.5:0.95 index is improved to 80.0%. Meanwhile, the GPU memory consumption of the model is reduced by 36.1%, decreasing from 7.56G to 4.83G. This study provides a feasible technical solution for the efficient deployment of intelligent tunnel vehicle monitoring systems. The engineering practicability of the proposed method in the video cloud networking scenario is further validated.