Volume 11, Issue 1 (Spring & Summer 2025)
KJES 2025, 11(1): 152-174 |
Back to browse issues page
Download citation:
BibTeX | RIS | EndNote | Medlars | ProCite | Reference Manager | RefWorks
Send citation to:
BibTeX | RIS | EndNote | Medlars | ProCite | Reference Manager | RefWorks
Send citation to:
Farhangi M, Milan A, Fallahi G, Khankeshi-Zadeh E. Depth estimation and 3D reconstruction from a single image based on the MiDaS deep learning model. KJES 2025; 11 (1) :152-174
URL: http://gnf.khu.ac.ir/article-1-2934-en.html
URL: http://gnf.khu.ac.ir/article-1-2934-en.html
1- Shahid Beheshti University
2- Shahid Beheshti University ,a_milan@sbu.ac.ir
3- K. N. Toosi University of Technology
2- Shahid Beheshti University ,
3- K. N. Toosi University of Technology
Abstract: (271 Views)
3D reconstruction plays an important role in surveying and close-range photogrammetry, facilitating the accurate extraction of geometric information from objects and their surrounding environment. However, conventional methods in this field typically require multi-view images along with positional and angular data of the camera, which can pose limitations in certain practical applications. This study introduces a novel approach based on the MiDaS deep learning model, one of the most accurate architectures for monocular depth estimation, which is capable of generating a relative depth map from a single 2D image. The final 3D model is then extracted using the Poisson Surface Reconstruction algorithm, without the need for spatial information or camera orientation data. To evaluate the performance of the proposed method, the resulting 3D model was compared against a reference model produced by the conventional photogrammetry method. The results showed a Root Mean Square Error (RMSE) of 0.775 centimeters, confirming the appropriate accuracy and reliability of the proposed approach under multi-view data limitations. This study demonstrates the high potential of deep learning models like MiDaS in 3D reconstruction and surveying applications, and highlights that using more advanced versions such as DPT could further improve accuracy in future research and applications.
Keywords: Depth Estimation, 3D Reconstruction, Single Image, Deep Learning, Photogrammetry, Machine Vision.
Type of Study: Original Research |
Subject:
Remote sensing and GIS
Received: 2025/03/25 | Accepted: 2025/08/16 | Published: 2025/08/25
Received: 2025/03/25 | Accepted: 2025/08/16 | Published: 2025/08/25
Send email to the article author
| Rights and permissions | |
 |
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License. |
