Real-Time Detection of Heavy-Duty Vehicle Lane Violations: A Comparative Analysis of Yolov11 and Yolov12

Abstract

The unnecessary occupation of the left lane on highways and main roads by heavy-duty vehicles (such as trucks and buses) poses a significant traffic safety problem. Due to their low speeds and large sizes, these vehicles disrupt traffic flow, forcing other drivers into sudden braking or lane-changing maneuvers, thereby increasing the risk of accidents. Studies conducted indicate that left-lane violations by commercial vehicles contribute to nearly 15% of traffic accident globally. In this study, an AI-based prototype system was developed to detect and prevent left-lane violations by heavy-duty vehicles. The system employs a vehicle-mounted camera that analyzes lane markings, road barriersand surrounding vehicles in real time. The captured data were processed using two state-of-the-art object detection models, YOLOv11 and YOLOv12, and their performances were comparatively evaluated. Experimental results demonstrate that YOLOv12 outperforms YOLOv11 in terms of overall performance, achieving higher values in both mAP@50 (85.6%) and precision (89%). YOLOv12 yielded superior detection results for the car (93.6%), bus (72.6%), and lane violation (96.6%) classes. However, for the truck class, YOLOv11 achieved slightly better accuracy (86.0%) compared to YOLOv12 (80.6%). Training curves further revealed that YOLOv12 stabilized its losses more rapidly and exhibited a more consistent learning process. In conclusion, the proposed system provides real-time detection of left-lane violations and delivers visual and auditory warnings to drivers, thereby encouraging safer lane usage. The comparative analysis of YOLOv11 and YOLOv12 highlights that YOLOv12 generally offers superior performance, while class-specific variations underline the importance of model selection in traffic safety applications.

Keywords

• Heavy-Duty Vehicles
• Lane Occupation
• Safe Driving
• Traffic Congestion
• YOLOv11

References

1. Global status report on road safety (2021). https://www.itf-oecd.org/sites/default/files/docs/irtad-road-safety-annual-report-2021.pdf
2. Weng, J., & Meng, Q. (2011). Analysis of driver behavior in freeway lane-changing events using vehicle trajectory data. Journal of Transportation Safety & Security, 3(2), 117–129
3. Leduc, G. (2009), Road traffic data: Collection methods and applications. Working Papers on Energy, Transport and Climate Change, No. 1, JRC, European Commission.
4. Zhang, L., & Du, B. (2017). High-level semantics extraction from traffic images using deep convolutional neural networks. Neurocomputing, 239, 64–73.
5. Redmon, J., & Farhadi, A. (2018). YOLOv3: An incremental improvement. arXiv preprint arXiv:1804.02767. Bongiorno, C., Wilson, R., & Wright, S. (2022). Advanced vehicle monitoring for heavy vehicles: Addressing lane violations using AI systems. IEEE Transactions on Intelligent Transportation Systems, 23(9), 14532–14541.
6. Weng, J., & Meng, Q. (2011). Analysis of driver behavior in freeway lane-changing events using vehicle trajectory data. Journal of Transportation Safety & Security, 3(3), 153-167.
7. Zhang, K., & Du, X. (2017). Lane detection based on deep convolutional neural network. IEEE Intelligent Transportation Systems Conference (ITSC), 1-6.
8. Chen, L., Englund, C., & Wang, H. (2019). Cooperative intelligent transport systems: A V2X-based approach for reducing traffic accidents. IEEE Transactions on Intelligent Transportation Systems, 20(1), 178-186.

9. Jetson Nano [Online]. Available: https://www.nvidia.com/tr-tr/autonomous-machines/embedded-systems/jetson-nano/product-development/
10. Jetson Nano Brings AI Computing to Everyone, https://developer.nvidia.com/blog/jetson-nano-ai-computing/
11. Kwang-Hyun Baek,Kyusam Lim, Gye Su Kim, Suki Kim (2010). A multi-lane MIPI CSI receiver for mobile camera applications, IEEE Transactions on Consumer Electronics
12. Sapkota, Ranjan, et al. "YOLO advances to its genesis: a decadal and comprehensive review of the You Only Look Once (YOLO) series." Artificial Intelligence Review 58.9 (2025): 274.