Stereo Matching Algorithm Based on Multi-frame Information Fusion and Fast Global Smoothing
WANG Zihao
SHI Chaoxia
WANG Yanqing
Abstract：Traditional stereo matching algorithms are mainly based on the single-frame binocular images for parallax estima-tion,and often cannot predict a good parallax value under the conditions of occlusion,weak texture,and mirror image switch,etc.This paper proposes a stereo matching algorithm of MVSGM(multi-view semi global matching)based on multi-frame parallax esti-mation.When calculating the initial cost of stereo matching,the current frame is projected onto the reference frame to obtain the cost information for fusion.Meanwhile,aiming at the estimation of parallax boundary fuzzy,an improved fast global smooth(IFGS)algorithm based on the second order gradient constraint is proposed.The algorithm has both fast solving and good smoothing effect.Middlebury,ETH3D and KITTI data sets are used for testing,and multi-frame cost fusion and fast global smoothing were added to the classical method SGM respectively in ablation experiments.The final obtained parallax map showed significant improvement in the error indexes RMSE,AVGE and bad2.0,which proved the effectiveness of the proposed method.
Keywords：stereo matchingdisparity estimationmulti-view geometryglobal smoothing
Publication Date：2025-10-20
Online Publishing Date：2026-01-16(First online date of this platform, not the publication date of the document)
Pages：7( 2766-2772 )