Research on the technical route of large-scale 3D reconstruction based on gaussian splatting
Meng Zhang
School of Chengdu Academy of Fine Arts, Sichuan Conservatory of Music
Qingwei Zhou
School of Artificial Intelligence,Sichuan Tourism University
Shun Chen
School of Artificial Intelligence,Sichuan Tourism University
DOI: https://doi.org/10.59429/esta.v12i4.12678
Keywords: gaussian splatting; large-scale 3D reconstruction; point cloud optimization; dynamic scene adaptation; accuracy evaluation
Abstract
Large-scale 3D reconstruction holds crucial application value in fields such as digital twins, smart cities, and cultural heritage protection. However, it faces technical bottlenecks including massive data volume, low detail restoration accuracy, and poor adaptability to dynamic scenes. Gaussian Splatting technology, leveraging the advantages of flexible point cloud representation, high rendering efficiency, and strong detail preservation capability, provides a new approach to address the challenges of large-scale reconstruction. This paper systematically sorts out the core principles and technical characteristics of Gaussian Splatting technology. It constructs a complete technical route for large-scale 3D reconstruction from four dimensions: data acquisition and preprocessing, Gaussian point cloud generation and optimization, dynamic scene adaptation, and accuracy evaluation. Combined with practical application scenarios, the technical advantages and existing challenges are analyzed, providing theoretical reference and practical guidance for the engineering application of this technology in complex large-scale scenes.
References
[1] H. Xie, Y. Liu, and Z. Wang, "A review of large-scale 3D reconstruction technologies," IEEE Trans. Circuits Syst. Video Technol., vol. 32, no. 7, pp. 4567–4582, Jul. 2022, doi: 10.1109/TCSVT.2021.3133456.
[2] M. Kerbl, B. Riegler, G. Kopanas, et al., "3D Gaussian splatting for real-time radiance field rendering," in Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR), Jun. 2023, pp. 10699–10709, doi: 10.1109/CVPR52729.2023.01035.
[3] Y. Li, J. Zhang, and H. Chen, "Efficient large-scale 3D reconstruction using adaptive Gaussian splatting," IEEE Trans. Vis. Comput. Graph., vol. 29, no. 12, pp. 4890–4901, Dec. 2023, doi: 10.1109/TVCG.2023.3290087.
[4] Z. Wang, L. Zhao, and C. Li, "Dynamic scene reconstruction with Gaussian splatting: A survey," Comput. Graph. Forum, vol. 42, no. 6, pp. 345–362, Sep. 2023, doi: 10.1111/cgf.14887.
[5] J. Chen, X. Wang, and S. Liu, "Multi-source data fusion for large-scale 3D reconstruction," IEEE Trans. Instrum. Meas., vol. 71, pp. 1–12, 2022, Art. no. 5018212, doi: 10.1109/TIM.2022.3193456.
[6] J. Redmon, S. Divvala, R. Girshick, et al., "You only look once: Unified, real-time object detection," in Proc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), Jun. 2016, pp. 779–788, doi: 10.1109/CVPR.2016.91.
[7] Z. Wang, A. C. Bovik, H. R. Sheikh, et al., "Image quality assessment: From error visibility to structural similarity," IEEE Trans. Image Process., vol. 13, no. 4, pp. 600–612, Apr. 2004, doi: 10.1109/TIP.2003.819861.