Multi-view stereo based on adaptive expansion propagation

Through computer vision and image processing techniques, a set of images from a scene can be reconstructed in 3D to recover a 3D model of the scene, in which dense reconstruction is a crucial part, and most existing algorithms cannot effectively recover weak texture regions in dense reconstruction....

Full description

Saved in:
Bibliographic Details
Published in2024 7th International Conference on Computer Information Science and Application Technology (CISAT) pp. 507 - 511
Main Authors Nie, Zixuan, Fu, Haohai
Format Conference Proceeding
LanguageEnglish
Published IEEE 12.07.2024
Subjects
3D reconstruction
Adaptation models
Computational modeling
Image reconstruction
Information science
multi-view stereo
Optimization
PatchMatch
Point cloud compression
propagation
Solid modeling
Surface reconstruction
Surface texture
Three-dimensional displays
weakly textured regions
Online AccessGet full text

Cover

More Information
Summary:Through computer vision and image processing techniques, a set of images from a scene can be reconstructed in 3D to recover a 3D model of the scene, in which dense reconstruction is a crucial part, and most existing algorithms cannot effectively recover weak texture regions in dense reconstruction. In response to this issue, this article proposes a hypothesis propagation strategy based on adaptive extension. In the hypothesis propagation stage of PatchMatch MVS, an adaptive mechanism for region propagation is introduced, and an update counter is set to regulate the expansion of the sampling area. This ensures that there are enough excellent hypotheses for sampling in this algorithm iteration, and the quality of sampling hypotheses is not limited by fixed sampling pixel positions, thereby improving the probability of excellent hypotheses being sampled in the sampling space. It is demonstrated in the ETH3D dataset that our method reconstructs models with higher integrity, and fewer voids and smoother surfaces in the reconstruction of weakly textured regions.
DOI:10.1109/CISAT62382.2024.10695320