Scholarlink

<p>With the increasing demand for smart mobile devices in park environments, there are higher requirements for their ability to adapt to complex scenarios. In unique park scenes such as lakeside paths and irregular stairs, the blurred road edges make it difficult for traditional algorithms to accurately distinguish between drivable and restricted areas, posing a significant challenge to the path planning and stable operation of smart mobile devices. To enhance the safe operation of these devices on complex paths, this study proposes an improved semantic segmentation method for park drivable areas, based on the Short-Term Dense Connection (STDC) network. This method effectively addresses issues such as imprecise road edge segmentation, insufficient road position awareness, and the limited generalization capability of existing datasets. Utilizing the STDC network as a backbone for key image feature extraction, a Multi-scale Coordinate Attention Module (MCAM) is introduced to bolster road position awareness and refine edge segmentation. Moreover, an Atrous Spatial Pyramid Pooling (ASPP) module is incorporated, allowing the network to manage local details and global contextual information within images. Besides, a dataset tailored to park environments is constructed using real-world and simulated park scene data for training and validation. The experimental results show that the algorithm achieved a mIoU of 98.91% on the self-built dataset, which is 3.07% higher than the STDC network. Additionally, its mIoU on the public dataset is 2.91% higher than the STDC network.</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

Improvement of park drivable area segmentation method based on STDCSeg network

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年4月3日

查看全文
  • Ting Zhang,
  • Yuansheng Liu,
  • Yangyang Fan,
  • Ming Lu

摘要

With the increasing demand for smart mobile devices in park environments, there are higher requirements for their ability to adapt to complex scenarios. In unique park scenes such as lakeside paths and irregular stairs, the blurred road edges make it difficult for traditional algorithms to accurately distinguish between drivable and restricted areas, posing a significant challenge to the path planning and stable operation of smart mobile devices. To enhance the safe operation of these devices on complex paths, this study proposes an improved semantic segmentation method for park drivable areas, based on the Short-Term Dense Connection (STDC) network. This method effectively addresses issues such as imprecise road edge segmentation, insufficient road position awareness, and the limited generalization capability of existing datasets. Utilizing the STDC network as a backbone for key image feature extraction, a Multi-scale Coordinate Attention Module (MCAM) is introduced to bolster road position awareness and refine edge segmentation. Moreover, an Atrous Spatial Pyramid Pooling (ASPP) module is incorporated, allowing the network to manage local details and global contextual information within images. Besides, a dataset tailored to park environments is constructed using real-world and simulated park scene data for training and validation. The experimental results show that the algorithm achieved a mIoU of 98.91% on the self-built dataset, which is 3.07% higher than the STDC network. Additionally, its mIoU on the public dataset is 2.91% higher than the STDC network.