基于高斯局部连续致密化的稀疏多视图三维重建算法OA

Sparse multi-view 3D reconstruction finds wide application in virtual reality,digital human modeling,and robot vi-sion.However,existing methods still exhibit clear limitations in detail representation and geometric accuracy due to insuffi-cient initial reconstruction information.This paper proposed a sparse multi-view 3D reconstruction algorithm based on Gaussian local continuous densification to improve reconstruction quality and robustness under sparse input conditions.The algorithm built on an initial 3D Gaussian model.It densified sparse Gaussian regions by interpolating new Gaussian points with neighbor-hood geometric continuity to enhance local structure representation.The loss function incorporated depth regularization and normal regularization terms to optimize geometric consistency and suppress artifact generation.Experiments on self-collected datasets and public datasets demonstrate that the proposed algorithm significantly improves novel-view synthesis quality.It out-performs multiple existing sparse multi-view 3D reconstruction algorithms on several evaluation metrics.The proposed algo-rithm exhibits strong generalization ability and reconstruction accuracy.