Journal of the Korean Geotechnical Society (한국지반공학회논문집)

Korean Geotechnical Society (한국지반공학회)
권호 표지
DOI QR코드

DOI QR Code

Effect of Learning Data on the Semantic Segmentation of Railroad Tunnel Using Deep Learning

딥러닝을 활용한 철도 터널 객체 분할에 학습 데이터가 미치는 영향

  • Received : 2021.10.29
  • Accepted : 2021.11.16
  • Published : 2021.11.30
Download PDF
⟨ Previous Next ⟩

Abstract

Scan-to-BIM can be precisely mod eled by measuring structures with Light Detection And Ranging (LiDAR) and build ing a 3D BIM (Building Information Modeling) model based on it, but has a limitation in that it consumes a lot of manpower, time, and cost. To overcome these limitations, studies are being conducted to perform semantic segmentation of 3D point cloud data applying deep learning algorithms, but studies on how segmentation result changes depending on learning data are insufficient. In this study, a parametric study was conducted to determine how the size and track type of railroad tunnels constituting learning data affect the semantic segmentation of railroad tunnels through deep learning. As a result of the parametric study, the similar size of the tunnels used for learning and testing, the higher segmentation accuracy, and the better results when learning through a double-track tunnel than a single-line tunnel. In addition, when the training data is composed of two or more tunnels, overall accuracy (OA) and mean intersection over union (MIoU) increased by 10% to 50%, it has been confirmed that various configurations of learning data can contribute to efficient learning.

Scan-to-BIM은 라이다(Light Detection And Ranging, LiDAR)로 구조물을 계측하고 이를 바탕으로 3D BIM(Building Information Modeling) 모델을 구축하는 방법으로 정밀한 모델링이 가능하지만 많은 인력과 시간, 비용이 소모된다는 한계를 가진다. 이러한 한계를 극복하기 위해 포인트 클라우드 데이터를 대상으로 딥러닝(Deep learning) 알고리즘을 적용하여 구조물의 객체 분할(Semantic segmentation)을 수행하는 연구들이 진행되고 있으나 학습 데이터에 따라 객체 분할 정확도가 어떻게 변화하는지에 대한 연구는 미흡한 실정이다. 본 연구에서는 딥러닝을 통한 철도 터널의 객체 분할에 학습 데이터를 구성하는 철도 터널의 크기, 선로 유형 등이 어떤 영향을 미치는지 확인하기 위해 매개변수 연구를 수행하였다. 매개변수 연구 결과, 학습과 테스트에 사용한 터널의 크기가 비슷할수록, 단선 터널보다는 복선 터널로 학습하는 경우에 더 높은 객체 분할 성능을 보였다. 또한, 학습 데이터를 두 가지 이상의 터널로 구성하면 전체 정확도(Overall Accuracy, OA)와 MIoU(Mean Intersection over Union)가 적게는 10%에서 많게는 50%가량 증가하였는데 이로부터 학습 데이터를 다양하게 구성하는 것이 효율적인 학습에 기여할 수 있음을 확인하였다.

Keywords

Acknowledgement

본 연구는 한국철도기술연구원 주요사업 '기존 철도구조물의 BIM기반 유지관리를 위한 역설계 모델링 기술 개발(PK2103C2)'과 정부(과학기술정보통신부)의 재원으로 한국연구재단의 지원(No. 2019R1C1C1008326)을 받아 수행되었으며, 이에 깊은 감사를 드립니다.

References

  1. Hergunsel, M. F. (2011), "Benefits of Building Information Modeling for Construction Managers and BIM based Scheduling", Master's thesis, Worcester Polytechnic Institute.
  2. Hong, S., Park, I.-S., Heo, J., and Choi, H. (2012), "Indoor 3D Modeling Approach based on Terrestrial LiDAR", KSCE Journal of Civil and Environmental Engineering Research, Vol.32, No.5D, pp.527-532 (in Korean).
  3. Jaderberg, M., Simonyan, K., Zisserman, A., and Kavukcuoglu, K. (2015), "Spatial transformer networks", Advances in Neural Information Processing Systems.
  4. Kim, H., Yoon, J., and Sim, S.-H. (2020), "Automated Bridge Component Recognition from Point Clouds Using Deep Learning", Structural Control Health Monitoring, Vol.27, No.9, e2591. https://doi.org/10.1002/stc.2591
  5. Kreider, R., Messner, J., and Dubler, C. (2010), "Determining the Frequency and Impact of applying BIM for Different Purposes on Projects", Proceedings 6th International Conference on Innovation in Architecture, Engineering and Construction (AEC), pp.1-10.
  6. Landrieu, L. and Simonovsky, M. (2018), "Large-scale Point Cloud Semantic Segmentation with Superpoint Graphs", Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp.4558-4567.
  7. Lee, J.S., Park, J., and Ryu, Y.-M. (2021), "Semantic Segmentation of Bridge Components based on Hierarchical Point Cloud Model", Automation in Construction, Vol.130, 103847. https://doi.org/10.1016/j.autcon.2021.103847
  8. Lee, O.-G. and Sim, J.-Y. (2021), "Point Cloud Acquisition and Preprocessing based on LiDAR", Broadcasting and Media Magazine, Vol.26, No.2, pp.9-17 (in Korean).
  9. Meng, J. (2020), "Research on 3D spatial semantic segmentation based on PointNet", Master's thesis, Chonnam National University (in Korean).
  10. Ministry of Economy and Finance (2020), "The Comprehensive Plan for the Korean New Deal", (in Korean).
  11. Ministry of Land, Infrastructure and Transport (2018), "Smart Construction Technology Roadmap", (in Korean).
  12. Park, J. (2021), "Method of Automating Scan-to-BIM Process based on Deep Learning", Master's thesis, Sejong University (in Korean).
  13. Park, T.-S. and Lee, S.-H. (2015), "Analyses of Existing Tunnel Liner Behaviors Caused by Excavation of Upper Layer with Using Laser Scanning Technology", Journal of the Korean Geotechnical Society, Vol.31, No.10, pp.29-36 (in Korean). https://doi.org/10.7843/KGS.2015.31.10.29
  14. Qi, C.R., Su, H., Mo, K., and Guibas, L.J. (2017), "PointNet, Deep Learning on Point Sets from 3D Classification and Segmentation", Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp.77-85.
  15. Rao, Y., Zhang, M., Cheng, Z., Xue, J., Pu, J., and Wang, Z. (2021), "Semantic Point Cloud Segmentation Using Fast Deep Neural Network and DCRF", Sensors, Vol.21, 2731. https://doi.org/10.3390/s21082731
  16. Robert, L. (1986), "Stochastic Sampling in Computer Graphics", ACM Transaction on Graphics, Vol.5, No.1, pp.51-72. https://doi.org/10.1145/7529.8927
  17. Soilan, M., Novoa, A., Sanchez-Rodriguez, A., Riveiro, B., and Arias, P. (2020), "Semantic Segmentation of Point Clouds with Pointnet and Kpconv Architectures Applied to Railway Tunnels", ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol.V-2-2020. pp.281-288. https://doi.org/10.5194/isprs-annals-V-2-2020-281-2020
  18. Wang, Y., Sun, Y., Liu, Z., Sarma, S.E., Bronstein, M.M., and Solomon, J.M. (2019), "Dynamic Graph CNN for Learning on Point Clouds", ACM Transaction on Graphics, Vol.38, No.5, 146.