Article Details

Abstract

To address the issues of low accuracy and poor real-time performance in trailing vehicle behavior recognition under complex traffic environments, this paper proposes an intelligent analysis algorithm based on multimodal fusion and spatiotemporal clustering. First, the algorithm collects multi-dimensional vehicle data through a high-definition camera system, and then initiates a three-level screening process. It completes the initial grouping based on license plates. When the number of captures of a vehicle reaches a preset threshold (such as 3 times), it verifies the consistency of information using a feature-matching degree model. Finally, it uses an improved DBSCAN clustering algorithm to conduct trajectory analysis. Compared with traditional manual inspection (with an average time consumption of more than 48 hours and a missed detection rate of 35%), this algorithm significantly improves the efficiency and accuracy of trailing behavior recognition. Moreover, relying on the collaborative architecture of hardware and algorithms, it ensures robustness in complex scenarios and ultimately provides a reliable technical solution for intelligent traffic security.

Keywords

trailing vehicle detection; multimodal feature fusion; spatiotemporal trajectory analysis

References

1. [1] Shi Shixi High risk area access control and tailgating warning based on video detection [J]. Mechanical Design and Manufacturing Engineering, 2021, 50 (11): 91-97
2. [2] Zhang Yashu Time Space Pattern Recognition Method Based on Big Data Analysis [J]. Wireless Internet Technology, 2025, 22 (6): 16-19
3. [3] Bao Jinyu Research on Vehicle Intelligent Monitoring Technology Based on Video Analysis [D]. Beijing: North China University of Technology, 2019
4. [4] Wang Guilin, Chen Xiaoling, Yue Jiahao, etc DBSCAN-SMOTEENN-RF Joint Algorithm and Its Application in 3D Geological Modeling [J]. Geography and Geographic Information Science, 2025, 41 (3): 19-26
5. [5] Li Zheyu Optimization method for customized bus routes based on refined spatiotemporal clustering of demand [D]. Suzhou: Soochow University, 2025
6. [6] Hunan Normal University A measurement method for industrial quality indicators based on multimodal temporal clustering: 202510831830.7 [P]. 20225-12-02
7. [7] Zhu Shiliang Research and Application of Remote Sensing Object Detection Algorithm Based on Deep Learning [D]. Beijing: Beijing University of Information Science and Technology, 2024
8. [8] Wang Mingdao Research on Human Action Recognition Technology for Skeletal Sequence [D]. Beijing: Beijing University of Posts and Telecommunications, 2024
9. [9] Wang Tenghui Research on the joint positioning method of Beidou and 5G [D]. Xi'an: University of Chinese Academy of Sciences (National Time Service Center of Chinese Academy of Sciences), 2021
10. [10] Xie Zhihao, Li Guogang Heterogeneous CNN Accelerator for Collaborative Design of Software and Hardware [J]. Journal of Huaqiao University (Natural Science Edition), 2025, 46 (2): 209-216

About the authors

Lin Jianmin (1983-), male, Han ethnicity, from Shaoxing, Zhejiang, is an artificial intelligence engineer with a bachelor's degree. His research interests include artificial intelligence and new materials.