Polarimetric multi-view 3D reconstruction based on parallax angle and zenith angle optimization
投稿时间:2020-03-28  修订日期:2021-01-15  download
摘要点击次数: 472
全文下载次数: 42
张瑞华 北京大学 地球与空间科学学院 遥感与地理信息系统研究所 北京 100871 ruihuazhang@pku.edu.cn 
施柏鑫 北京大学 数字视频编解码技术国家工程实验室 北京 100871  
杨锦发 北京大学 地球与空间科学学院 遥感与地理信息系统研究所 北京 100871  
赵红颖 北京大学 地球与空间科学学院 遥感与地理信息系统研究所 北京 100871 zhaohy@pku.edu.cn 
左正康 北京大学 地球与空间科学学院 遥感与地理信息系统研究所 北京 100871  
Abstract:Multi-view stereo reconstruction relies on object’ s surface feature, and thus occurred data deficiency in dealing with low-texture regions, the polarization characteristics of the reflection light be fused to observe completely surface under different light surroundings, which can reconstruct depth map of objects by calculating polarization parameters and normal information. However, there appears several problems such as azimuth ambiguity and zenith angle deviation using pure polarization information for 3D reconstruction, which result in distortion of the reconstruction results and even couldn’t get the depth. For objects with low-texture regions, capturing images from around 30 views from four polarizing angles using polarization camera, and using parallax bundle adjustment to optimize camera parameters and points’ coordinate and using passion optimization method to correct zenith angle deviation, the 3D reconstruction algorithm combining the multi-view stereo and polarization information can not only replenish the local data deficiency of multi-view 3D point cloud, but also solve the azimuth ambiguity and zenith angle deviation, and finally get the more accurate 3D reconstruction results.
keywords:polarization 3D reconstruction  bundle adjustment  poisson optimization  image fusion
  HTML  查看/发表评论  下载PDF阅读器

Copyright:《Journal of Infrared And Millimeter Waves》