Bulletin of Surveying and Mapping ›› 2022, Vol. 0 ›› Issue (9): 45-51.doi: 10.13474/j.cnki.11-2246.2022.0262

Previous Articles     Next Articles

Indoor adherent point cloud segmentation method based on joint optimization of minimal cut and deep learning

QIAN Jianguo1, ZHANG Yuqi1, TANG Shengjun2,3,4,5, WANG Weixi2,3,4,5, LI Xiaoming2,3,4,5   

  1. 1. School of Geomatics, Liaoning Technology University, Fuxin 123000, China;
    2. Research Institute of Smart City, School of Architecture and Urban Planning, Shenzhen University, Shenzhen 518061, China;
    3. Key Laboratory of Urban Natural Resources Monitoring and Simulation, Ministry of Natural Resources, Shenzhen 518061, China;
    4. Key Laboratory of Spatial Information Intelligent Perception and Service, Shenzhen, Shenzhen 518061, China;
    5. Key Laboratory of Urban Spatial Information Engineering, Guangdong Province, Shenzhen 518061, China
  • Received:2021-10-09 Published:2022-09-30

Abstract: With the development of digital city, the demand for 3D point cloud structuring as well as the accuracy requirement of urban 3D model reconstruction is getting higher and higher. How to effectively and accurately segment indoor semantic models and 3D reconstruction is a current hot research issue. Point cloud segmentation classification is an important basis for indoor point cloud structuring, and how to segment the adherent point cloud components accurately and use them for indoor point cloud structuring is a difficult problem in current urban modeling. This paper proposes a segmentation and classification method for indoor adhesion point cloud data, which firstly uses deep learning network to process indoor point cloud data, then classifies the point cloud data with labeled point cloud to get the target labeled point cloud, and uses Euclidean algorithm to cluster and segment the target point cloud, calculates the coordinates of each target centroid and horizontal radius by the enclosing box information of indoor semantic components. Finally, we use point cloud minimization to achieve accurate segmentation of the indoor adherent point cloud. In this paper, three sets of data obtained from indoor scenes are used to evaluate the accuracy and effectiveness of this segmentation method. The experimental results show that the segmentation optimization method proposed in this paper has high accuracy and data integrity.

Key words: indoor adherent point cloud, deep learning, labeled point cloud classification, Euclidean algorithm, minimum cut

CLC Number: