Acta Geodaetica et Cartographica Sinica ›› 2024, Vol. 53 ›› Issue (7): 1355-1370.doi: 10.11947/j.AGCS.2024.20230482

• Photogrammetry and Remote Sensing • Previous Articles     Next Articles

3D scene graph representation and application for intelligent indoor spaces

Shengjun TANG1,2(), Siqi DU2, Weixi WANG1,2(), Renzhong GUO1,2   

  1. 1.State Key Laboratory of Subtropical Building and Urban Science, Shenzhen University, Shenzhen 518061, China
    2.School of Architecture and Urban Planning, Shenzhen University, Shenzhen 518061, China
  • Received:2023-10-20 Published:2024-08-12
  • Contact: Weixi WANG E-mail:shengjuntang@szu.edu.cn;wangwx@szu.edu.cn
  • About author:TANG Shengjun (1991—), male, PhD, associate researcher, majors in urban 3D element structured reconstruction and multi-sensor fusion mapping. E-mail: shengjuntang@szu.edu.cn
  • Supported by:
    The Natural Science Foundation of Guangdong Province(2024A1515030061);Research Project of Shenzhen Science and Technology Innovation Committee(KJZD20230923115508017);The Research Project of State Key Laboratory of Subtropical Building and Urban Science(2023ZB18)

Abstract:

Existing methods for indoor 3D scene representation focus on object-oriented descriptions, with element representations limited to object-level semantic understanding. These methods lack the ability to express complex relational information within indoor scenes. Addressing the demands of intelligent indoor space tasks, there is a critical need for a structured model that can comprehensively and accurately describe the geometry, semantics, and relationships of indoor elements, while also supporting semantic retrieval and analytical reasoning. Based on the fundamental theory of 3D scene graphs, this paper innovatively proposes a 3D scene graph representation model tailored for intelligent indoor spaces. It systematically introduces the hierarchical organization, geometric representation, semantic description, and relational description methods of indoor 3D scene graphs. A conceptual model is established that uniformly describes the geometry, semantics, and relationships of indoor elements. Additionally, this graph model is compatible with existing 3D scene representation methods, ensuring good data compatibility. Finally, a comprehensive multi-level relational 3D scene graph model is constructed based on the publicly available IFC model. This model's application capabilities, potential, and limitations are systematically explored and analyzed through applications such as complex scene retrieval and topological analysis, in conjunction with large language models. The results demonstrate that the indoor 3D scene graph model possesses complex computation and analysis capabilities, can be directly integrated with large language models, and enables complex scene analysis applications through simple natural language prompts.

Key words: indoor modeling, graph model, IFC, CityGML, large language model

CLC Number: