Three-Dimensional Structure Measurement And Optimization Method Of Indoor Scene Based On Single Image
Ronghe Wang,Xinhai Zhang,Bo Zhang,Jianning Bi,Xiaolei Guo
DOI: https://doi.org/10.1007/978-981-13-6508-9_98
2020-01-01
Abstract:In the paper, we detect 2D surfaces and recognize 3D structures in images, and generate occluded part of three-dimensional structure. We detect objects vanish points and geometric segmentation lines, and generate the vector and normal plane. We measure the objects boundaries which have been detected. In the process, we have found the best linear segmentation positions, generated buildings, and other 3D models in room. We propose some effective assumptions to recognize the object model from images by geometric reasoning. At the same time, we put forward structure prediction technology combined with the volume reasoning by parameter representation of spatial objects. We detect scene model relationships of each other by combining image-rich appearance with geometric features. We proposed image geometric modeling grammar framework according to previous discriminative classifier. It is used to represent the physical structure of the visual component. The framework has broken the traditional probability texture context grammar tree model. We proposed spatial context theory and model generation rules. Finally, on the public existing data sets, we proved that we only use structure prediction and linear segmentation of scene to recover 3D structure by a lot of experiments, comparing to restoration algorithm using whole image structure, ours method can produce more credible scene entity models and space constraint relationships.