3d Reconstruction With Multi-View Texture Mapping
NEURAL INFORMATION PROCESSING (ICONIP 2017), PT III(2017)
摘要
In this paper, a novel 3D reconstruction with multi-view texture mapping method based on Kinect 2 is proposed. Camera poses of all chosen key frames are optimized according to photometric consistency. Optimized camera poses can make the projected point from vertices to different views get closer. A small range of translations with limited calculation is added in this method. A new form of data term and smoothness term in Markov Random Field (MRF) objective function is presented. The outlier images are rejected before view selection and Poisson blending are applied in the end. Experimental results show that our method achieves a high-quality 3D model with high fidelity texture.
更多查看译文
关键词
Camera poses optimization,Markov Random Field (MRF),Texture mapping
AI 理解论文
溯源树
样例
生成溯源树,研究论文发展脉络