Open AccessGPU based multi-scale depth map calculation for 3D reconstruction
Author(s) -
Tong Wu,
Hui Wang,
Yanling Wang,
Min Liang,
Jie Li
Publication year - 2021
Publication title -
journal of physics. conference series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.21
H-Index - 85
eISSN - 1742-6596
pISSN - 1742-6588
DOI - 10.1088/1742-6596/1920/1/012075
Subject(s) - depth map , scale (ratio) , computer science , computation , texture mapping , point cloud , texture (cosmology) , artificial intelligence , computer vision , 3d reconstruction , algorithm , computer graphics (images) , image (mathematics) , cartography , geography
When using multi-view stereo (MVS) method to calculate depth map for reconstructing dense 3D model, the model precision is often influenced by low-texture areas. To solve this problem, in this paper, we present an efficient multi-scale depth map based MVS method. First, the coarsest scale depth map is calculated which can easily capture the rich texture information in low-texture areas. Then we upsample this depth map to higher scale, and four forward-backward reprojections are used to reduce the error of depth difference between the last scale and the current scale. In addition, a GPU and CPU based cooperative optimization architecture is built to optimize and accelerate depth map computation in Patch Match (PM) stage. Experimental results show that, the proposed method can reconstruct quite accurate and dense point clouds compared to state-of-the-art methods.