基于RGB-D图像的室内场景CAD模型重建

Indoor scene CAD model reconstruction based on RGB-D image

针对传统三维重建方法中存在的依赖图组和计算复杂等问题,文中在重建过程中引入大规模CAD模型库,提出了一种基于单视角RGB-D图像的室内场景CAD模型重建方法。首先,利用FCN和点云数据估计室内场景布局;同时,利用Mask R-CNN和点云数据检索与待重建对象最匹配的CAD模型;最后,组合上述结果完成场景重建。实验证明,文中方法仅需单视角RGB-D图像即可完成三维重建,且重建图与原场景相似度高,有较好的鲁棒性。

In the view of relying on chart series and high computational complexity of traditional 3 D reconstruction methods,a new indoor scene CAD model reconstruction method based ona single-view RGB-D image,by introducing a large-scale CAD model library,is proposed.Firstly,point cloud and FCN are utilized to reconstruct roomlayout.Meanwhile,retrieval CAD model of furniture with point cloud and instance segmentation obtained by Mask R-CNN.Finally,compose results above to complete the scene reconstruction.Experiments show the proposed reconstruction method,only requiring a single-view image,can reconstruct 3 D scene,which has high similarity with the original scene,as well as has strong robustness.