Research on 3D Reconstruction Methods Based on Deep Learning
DOI:
https://doi.org/10.62051/rbpr6z70Keywords:
3D Reconstruction; Methods; Deep Learning.Abstract
Deep learning applications have been applied extensively and have made tremendous strides in the 3D reconstruction field in recent years. This paper offers a methodical review of deep learning-based techniques for single-view images, multi-view images, and video-based sequence approaches. For single-view methods, we focus on depth estimation using Convolutional Neural Networks (CNNs) and image-to-depth mapping using Generative Adversarial Networks (GANs). For multi-view methods, we explore 3D reconstruction based on multi-view stereo matching method, 3D points cloud reconstruction method and stereo flow estimation method. For video-based methods, we introduce depth estimation method based on optical flow and video sequence modeling using Recurrent Neural Networks (RNNs). Generally, the multi-view image method is more accurate and sophisticated than the single-view image method, while the method based on video sequences is more challenging and complex. Different 3D reconstruction methods depend on specific application scenarios and requirements. This review provides a considerable insight in research for 3D reconstruction and also make the conclusion as well as future prospect for this field.
Downloads
References
Bhoi, A. Monocular depth estimation: A survey. arXiv preprint arXiv:1901.09402. 2019
Ming, Y., Meng, X., Fan, C., & Yu, H. (2021). Deep learning for monocular depth estimation: A review. Neurocomputing, 438, 14-33.
Aleotti, F., Tosi, F., Poggi, M., & Mattoccia, S. (2018). Generative adversarial networks for unsupervised monocular depth prediction. ECCV workshops. 2018.
Roca-Pardinas, J., Lorenzo, H., Arias, P., & Armesto, J. From laser point clouds to surfaces: Statistical nonparametric methods for three-dimensional reconstruction. Computer-Aided Design, 40(5), 646-652, 2008
Iglhaut, J., Cabo, C., Puliti, S., Piermattei, L., O’Connor, J., & Rosette, J. Structure from motion photogrammetry in forestry: A review. Current Forestry Reports, 5, 155-168, 2019.
Zhou, Y., Zhang, L., Xing, C., Xie, P., & Cao, Y. Target three-dimensional reconstruction from the multi-view radar image sequence. IEEE Access, 7, 36722-36735, 2019
Luo, X., Huang, J. B., Szeliski, R., Matzen, K., & Kopf, J. Consistent video depth estimation. ACM ToG, 39(4), 71-1, 2020
Wang, Y., Wang, P., Yang, Z., Luo, C., Yang, Y., & Xu, W. Unos: Unified unsupervised optical-flow and stereo-depth estimation by watching videos. In CVPR (pp. 8071-8081, 2019
Lu, W., Cui, J., Chang, Y., & Zhang, L. A video prediction method based on optical flow estimation and pixel generation. IEEE Access, 9, 100395-100406, 2021
Teed, Z., & Deng, J. Deepv2d: Video to depth with differentiable structure from motion. arXiv preprint arXiv:1812.04605, 2018.
Downloads
Published
Conference Proceedings Volume
Section
License
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.