Scholarlink

<p>Emergency escape ramps can prevent traffic accidents at critical moments, saving lives and minimizing property damage, making their study of significant practical importance. However, existing datasets primarily focus on urban roads, with a lack of datasets for emergency escape ramp scenes, limiting further research. To address this issue, we first collected a dataset of emergency escape ramp scenes from highways and mountain roads. Subsequently, we proposed a dual-branch semantic segmentation method based on the DeepLabV3 + architecture. In this approach, a new branch is introduced to incorporate depth maps for auxiliary training. Additionally, Strip Pooling (SP) is integrated into the Atrous Spatial Pyramid Pooling (ASPP) module, and Coordinate Attention (CA) is added after the ASPP framework. This method achieves a Mean Intersection over Union (MIoU) of 80.80% and a Mean Pixel Accuracy (MPA) of 88.95%. Compared with state-of-the-art (SOTA) models, our method achieves a better balance between accuracy and model complexity, making it well-suited for deployment on edge devices. Additionally, testing on publicly available datasets showed that our method outperforms the results before the improvements. This is the first attempt at segmenting emergency escape ramp scenes, offering new insights for the study of emergency escape ramps on highways and mountain roads. Our code is released at <a href="https://github.com/Huzuosheng590/ED-DeepLab-An-Efficient-Dual-Branch-DeepLab-Network-with-Depth-Fusion.git">https://github.com/Huzuosheng590/ED-DeepLab-An-Efficient-Dual-Branch-DeepLab-Network-with-Depth-Fusion.git</a>.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Semantic Segmentation of Emergency Escape Ramp Scenes for Traffic Injury Prevention

  • 期刊论文

  • 发表日期：

    2026年2月2日

查看全文
  • Zuosheng Hu,
  • Guiling Li,
  • Luping Wang

摘要

Emergency escape ramps can prevent traffic accidents at critical moments, saving lives and minimizing property damage, making their study of significant practical importance. However, existing datasets primarily focus on urban roads, with a lack of datasets for emergency escape ramp scenes, limiting further research. To address this issue, we first collected a dataset of emergency escape ramp scenes from highways and mountain roads. Subsequently, we proposed a dual-branch semantic segmentation method based on the DeepLabV3 + architecture. In this approach, a new branch is introduced to incorporate depth maps for auxiliary training. Additionally, Strip Pooling (SP) is integrated into the Atrous Spatial Pyramid Pooling (ASPP) module, and Coordinate Attention (CA) is added after the ASPP framework. This method achieves a Mean Intersection over Union (MIoU) of 80.80% and a Mean Pixel Accuracy (MPA) of 88.95%. Compared with state-of-the-art (SOTA) models, our method achieves a better balance between accuracy and model complexity, making it well-suited for deployment on edge devices. Additionally, testing on publicly available datasets showed that our method outperforms the results before the improvements. This is the first attempt at segmenting emergency escape ramp scenes, offering new insights for the study of emergency escape ramps on highways and mountain roads. Our code is released at https://github.com/Huzuosheng590/ED-DeepLab-An-Efficient-Dual-Branch-DeepLab-Network-with-Depth-Fusion.git.