Application And Analysis of Computer Vision Technology in Traffic Safety

Du Li

doi:10.54097/65rx7v60

Authors

Du Li

DOI:

https://doi.org/10.54097/65rx7v60

Keywords:

Artificial Intelligence, Intelligent Transportation, Computer Vision.

Abstract

In recent years, the integration of intelligent transportation systems and artificial intelligence has been continuously driving profound changes in the field of traffic safety. At present, computer vision technology is increasingly becoming a key tool for enhancing road behavior perception and accident early warning capabilities. Over the past five years, relevant research has achieved significant achievements in areas such as pedestrian and vehicle detection, accident early warning, and damage identification of traffic facilities. A series of algorithms such as YOLO, Faster R-CNN, and CenterNet perform outstandingly in terms of detection accuracy and real-time performance, while multi-object tracking algorithms such as DeepSORT further enhance the stable tracking ability for continuous targets. Despite this, the field still faces many challenges, such as weak system robustness in complex environments, high costs of large-scale labeled data, prominent privacy and security risks, and real-time constraints of edge devices in actual deployment. This paper systematically reviews the advantages and limitations of the existing mainstream methods and offers prospects for future development directions, including multi-sensor fusion, model lightweighting, edge computing advancement, and self-supervised learning applications, with the aim of providing useful references for enhancing the safety and adaptability of intelligent transportation systems.

References

[1]J. Liu, M. Han, & T. Zhang. Using computer vision and street-level videos for tracking pedestrians and vehicles in complex urban environments. Transportation Research Procedia, 97, 221–230 (2025). https://doi.org/10.1016/j.trpro.2025.02.009

[2]L. Zhang, & S. Li. Vision-based pedestrian’s crossing risky behavior analysis from CCTV cameras. Sensors, 22(9), 3451 (2022). https://doi.org/10.3390/s22093451

[3]D. Raj, & P. Mehta. A robust deep learning-based system for pedestrian-aware collision avoidance (YOLOv9). International Journal of Computer and Communication Engineering, 14(2), 67–74 (2025). https://doi.org/10.1016/j.ijcce.2025.01.005

[4]A. Karim, et al. YOLOv8 + DeepSORT for real-time traffic incident detection. Vehicles, 15(9), 382 (2024). https://www.mdpi.com/2032-6653/15/9/382

[5]A. M. Elbasha, & M. M. Abdellatif. AIoT-based smart traffic management system. arXiv preprint arXiv:2502.02821 (2025).

[6]J. Redmon, S. Divvala, R. Girshick, & A. Farhadi. You Only Look Once: Unified, Real-Time Object Detection. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) (pp. 779-788) (2016).

[7]S. Ren, K. He, R. Girshick, & J. Sun. Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks. IEEE Transactions on Pattern Analysis and Machine Intelligence, 39(6), 1137-1149 (2017).

[8]X. Zhou, D. Wang, & P. Krähenbühl. Objects as Points. arXiv preprint arXiv:1904.07850 (2019).

[9]N. Wojke, A. Bewley, & D. Paulus. Simple Online and Realtime Tracking with a Deep Association Metric. In 2017 IEEE International Conference on Image Processing (ICIP) (pp. 3645-3649) (2017).

[10]A. M. Roy, & J. Bhaduri. A computer vision enabled damage detection model with improved YOLOv5 based on transformer prediction head. arXiv preprint arXiv:2303.04275 (2023).

[11]A. S. Abdelrahman, M. Abdel-Aty, & D. Wang. Video-to-text pedestrian monitoring (VTPM) using computer vision. arXiv preprint arXiv:2408.11649 (2024).

[12]A. Aboah, B. Wang, U. Bagci, & Y. Adu-Gyamfi. Real-time multi-class helmet violation detection using computer vision. arXiv preprint arXiv:2304.08256 (2023).

[13]J. Hu. YOLOv7 with attention mechanisms for pedestrian detection. Sensors, 24(21), 6922. (2024).

[14]H. Ghahremannezhad, H. Shi, & C. Liu. Real-time accident detection in traffic surveillance using deep learning. arXiv preprint arXiv:2208.06461 (2022).