With the continuous growth of highway traffic flow, accurate and efficient traffic event detection has become critical for ensuring road safety and smooth operations. A highway traffic event detection method based on the interaction of large and small models was proposed. First, the YOLOv8 small model leverages its efficient object localization mechanism to quickly perform preliminary detection and screening of traffic targets from incoming video frames, accurately identifying potential event regions. Then, a Transformer-based large model conducts in-depth analysis of the candidate regions, achieving precise recognition of traffic event types through global feature extraction and semantic understanding. By combining the small model’s rapid screening with the large model’s detailed analysis, the proposed method balances detection speed and accuracy. Experimental results on a self-built highway traffic event dataset demonstrate that this approach significantly outperforms single-model methods in terms of precision, recall, and other metrics, providing an efficient and reliable technical solution for highway traffic management.
QI Yu-liang, SHI Zhao-nian, SUN Ye-teng, LIU Hai-yu, KANG Su-mei.
Method for Traffic Incident Detection on Expressways Based on Interaction between Large and Small Models[J].
Science Technology and Engineering, 2026, 26(6): 2627-2635 DOI:10.12404/j.issn.1671-1815.2502927
YOLO(you only look once)目标检测算法是当前深度学习领域中经典且高效的一阶段目标检测方法,因其简洁的结构、快速的检测速度和实时性,以及出色的多尺度检测能力、全局上下文信息利用和多任务学习支持,广泛应用于交通监控、无人机图像分析和自动驾驶等多个实际场景中。YOLOv8相比前代模型在精度、速度、结构设计和部署灵活性等方面实现了显著的提升。
ShiJunqing, ChenLinwu, LiSulan, et al. Urban road traffic incident detection algorithm based on CNN[J]. Journal of Highway and Transportation Research and Development, 2022, 39(3): 176-182.
XuePanfen, JinKaibin. Research on traffic anomaly event detection algorithm based on improved YOLOv6 model[J]. Technology of Highway and Transport, 2025, 41(1): 176-181, 189.
LiangYu. Traffic incident detection system based on 3D convolutional neural network[J]. Western China Communications Science & Technology, 2023(5): 176-177, 199.
ZhangBing, ZouShaoquan, LuChunlin, et al. A framework for highway traffic incident detection based on VAE-CatBoost[J]. Journal of Chongqing University of Technology(Natural Science Edition), 2024, 38(8): 197-207.
[12]
SunG, YanK, FanC. Deep learning-based vehicle tracking and traffic event detection[C]//4th International Conference on Computer Vision, Image and Deep Learning (CVIDL). Zhuhai: CVIDL, 2023: 522-526.
LuoYuxin, ZhangBing, XueYunqiang. Weighted random forest based on variable analysis and particle swarm optimization traffic incident detection method[J]. Science Technology and Engineering, 2021, 21(14): 6044-6049.
WangZhen, ZhaiKe, XueSai, et al. Unsupervised anomaly detection based on universal visual transformer and attentionen hancement[J]. Journal of Beijing Jiaotong University, 2025, 49(3): 14-22.
ZhangFeng, HuangShixin, HuaQiang, et al. Novel image classification model based on Depth-wise convolution neural network and visual Transformer[J]. Computer Science, 2024, 51(2): 196-204.
YangTaocun, ShiWeifeng, LiGuohua, et al. Application of large-scale AI vision models to detect foreign object intrusion into railway lines[J]. China Railway, 2025(1): 23-29, 48.
GuoJialin, ZhiMin, YinYanjun, et al. Review of research on CNN and visual transformer hybrid models in image processing[J]. Journal of Frontiers of Computer Science and Technology, 2025, 19(1): 30-44.