Object Reconstruction Using Part Labeling and Composition from Monocular View

• Main Authors: Chen, Wei-Cheng, 陳威丞
• Other Authors: Lin, I-Chen
• Format: Others
• Language: zh-TW
• Published: 2013
• Online Access: http://ndltd.ncl.edu.tw/handle/35835077710599840283

碩士 === 國立交通大學 === 多媒體工程研究所 === 102 === It is challenging to reconstruct the target object from monocular view because of the limitation of the input data. Otherwise, since range scan is an incomplete and noisy data, we incorporate the color image captured from the same viewpoint to provide extra cues. Because of the growing popularity of real world objects, it is not enough to reconstruct the target object only by deforming the existing models. In this thesis, we propose a part-assembly approach to reconstruct the structure. First, we retrieve the most similar model, which is called representative model from database to guide the model alignment. With the representative model, we use the primary facet axes between the point cloud and the model to align all the database models. Finally, we deform the most similar part retrieved from database to fit the target object and compose them to reconstruct the final structure we want.