基于卷积神经网络的室内场景三维重建技术研究

三维场景重建技术是计算机视觉领域的十分重要的研究课题。传统三维场景重建大多是专业工程师通过手工制图实现,效率不高且成本较高。对此提出一种基于卷积神经网络的三维场景重建方法。该方法在对2D图像进行语义分割的基础上,提取分割后的室内场景元素图像块,训练一个基于卷积神经网络的三维模型匹配模型;再将匹配得到的三维模型结合深度图构造的残缺三维模型,进一步进行组合,从而完成室内场景的三维重建工作。实验验证了该方法的可行性和优异性。

3D RECONSTRUCTION OF INDOOR SCENE BASED ON CNN

3D scene reconstruction technology is one of the most important research topics in the field of computer vision. Traditional 3D scene reconstruction is mostly achieved by professional engineers through manual drawing, which is inefficient and costly. This paper proposed 3D scene reconstruction method based on CNN. Based on the semantic segmentation of 2D images, we extracted the segmented image blocks of indoor scene elements, and trained a 3D model matching model based on CNN. Then, the matched 3D model was combined with the incomplete 3D model constructed by depth map, and further combined to complete the 3D reconstruction of indoor scenes. The feasibility and superiority of this method are verified by experiments.