Real-Scale 3D Reconstruction with Monocular Zoom Technology

Jinao Song,Jie Li,Hao Fan,Lin Qi,Shu Zhang,Yong Chen,Junyu Dong
DOI: https://doi.org/10.1109/tim.2024.3497052
IF: 5.6
2024-01-01
IEEE Transactions on Instrumentation and Measurement
Abstract:We propose a method that is able to use the monocular zoom technology for real-scale 3D reconstruction of the scene. To reconstruct the scene, we take a sequence of zoomed-in and zoomed-out figures. First, we can estimate zoomed-in camera parameters by using the known zoomed-out camera parameters, which avoids calibrating the camera parameters twice. Then, we use the SfM method (COLMAP) to reconstruct free-scale translations among these figures. After that, as we have pairs of zoom frames in the same scene, we can calculate the true scale of the scene by comparing the ratio between the free-scale translation of a pair of zoom frames and the difference in zoomed-out and the zoomed-in focal length. Finally, we use RAFT-stereo to compute the depth of the scene. In detail, we select two adjacent figures taken at the same focal length, make a stereo correction for them, and remove the non-covision area of the corrected images. This way, we obtain a more accurate matching of these images and then get a dense real-scale 3D reconstruction. Experimental results have demonstrated that our method achieves good performance on monocular 3D reconstruction with the real scale.
What problem does this paper attempt to address?