Related Experiment Video
Updated: May 12, 2026
Design and Analysis for Fall Detection System Simplification
Published on: April 6, 2020
HR-YOLO: Segmentation and detection of emergency escape ramp scenes using an integrated HR-net and improved YOLOv12
Guiling Li1, Zuosheng Hu2, Haozhi Zhang2
1College of Electronic and Electrical Engineering, Henan Normal University, Xinxiang, China.
Objective:
With the development of intelligent transportation systems, the demand for automatic recognition and monitoring of critical road safety infrastructure has been increasing. Particularly in high-risk road sections such as mountainous areas and steep downhill stretches, Emergency Escape Ramps (EERs) play a crucial role in preventing severe accidents caused by out-of-control vehicles. Accurate detection of these ramps is essential for enhancing road traffic safety.
Methods:
To address the limitations of existing methods, such as inadequate segmentation accuracy and poor robustness in object detection, this paper proposes a new model (HR-YOLO) that integrates the HR-Net model with an improved YOLOv12 model. To support the segmentation and detection of emergency escape ramp scenarios, we constructed a custom dataset consisting of 411 annotated images across 8 categories. This dataset reflects typical road environments and object types encountered in emergency ramp areas, providing a reliable foundation for model training and evaluation.
Result:
The HR-Net achieves a mIoU and mPA of 85.50% and 91.23%, respectively. The YOLOv12 network attains a mAP and F1 score of 78.7% and 75.9%, respectively, for unsegmented images. The YOLOv12 model enhanced with Coordinate Attention (CAYOLOv12) is combined with the HR-Net model. Compared to the standalone YOLOv12 model, the HR-YOLO model improves mAP and F1 scores by 10% and 9.6%, respectively.
Conclusions:
The proposed approach provides an efficient and reliable technological solution for the intelligent recognition of key infrastructure in traffic safety monitoring systems.
