测绘通报 ›› 2024, Vol. 0 ›› Issue (12): 101-105.doi: 10.13474/j.cnki.11-2246.2024.1216

• 工程测量分会年会优选论文 • 上一篇    下一篇

基于分割一切模型SAM的实景三维场景语义分割

李锋1,2, 薛梅1,2, 詹勇1,3, 杨元1,3   

  1. 1. 重庆市测绘科学技术研究院, 重庆 401120;
    2. 自然资源部智能城市时空信息与装备工程技术创新中心, 重庆 401120;
    3. 重庆市勘测院智能城市空间技术创新中心, 重庆 401120
  • 收稿日期:2024-07-24 发布日期:2024-12-27
  • 作者简介:李锋(1983-),男,硕士,正高级工程师,主要研究方向为实景三维领域。E-mail:lifeng@cqkcy.com
  • 基金资助:
    重庆市科研机构绩效激励引导专项(CSTB2023JXJL-YFX0060);重庆市规划和自然资源科研项目(KJ-2023039)

Semantic segmentation of 3D real scene based on segment anything model

LI Feng1,2, XUE Mei1,2, ZHAN Yong1,3, YANG Yuan1,3   

  1. 1. Chongqing Institute of Surveying and Mapping Science and Technology, Chongqing 401120, China;
    2. Technology Innovation Center for Spatio-temporal Information and Equipment of Intelligent City, Ministry of Natural Resources, Chongqing 401120, China;
    3. Technology Innovation Center for Intelligent City Space of Chongqing Survey Institute, Chongqing 401120, China
  • Received:2024-07-24 Published:2024-12-27

摘要: 基于深度学习和计算机视觉技术的场景语义分割是当前研究的热点。本文提出了包含“场景输入—预处理—模型推断—语义分割”的实景三维场景语义分割框架,通过将实景三维场景作为输入,按照正交投影的方式转为多视图二维图像,开展分割推理,获得分割掩码并进行处理,实现了实景三维对象拣选、单体化、语义化处理。试验结果表明,本文方法具有较好的语义分割效果和运行效率。

关键词: 分割一切模型, 实景三维, 语义分割

Abstract: Scene semantic segmentation based on deep learning and computer vision technology is currently a hot research topic. This paper proposes a 3D real scene semantic segmentation framework that includes “scene input-preprocessing-model inference-semantic segmentation”. By transforming the 3D real scene as input into multi-view 2D images through orthogonal projection, segmentation inference is carried out, and segmentation masks are generated and further processed,achieving the object selection, singulation, and semantic processing of 3D real scene.The experiments show that the method has good semantic segmentation performance and efficiency.

Key words: segment anything model, 3D real scene, semantic segmentation

中图分类号: