2D Layout Estimation And 3D Layout Reconstruction Of Indoor Scene

3D topologies of indoor scenes are crucial to semantic understanding and reconstruction of scenes.It is a challenge field of computer vision and has got a lot of attention.The research is very useful in applications like real estate and furniture design.In recent years,with the rapid development of artificial intelligence and deep learning,it is a hot issue of many researchers to combine this method of deep learning to solve this problem.In this paper,we present a fully automated solution for reconstructing a 3D topology of an indoor scene from a single 2D image.Based on our knowledge,this is the first fully automated 3D topology reconstruction based on a single 2D image in an indoor scene.Our proposed method is consisted of three key algorithm modules:(1)A novel fully convolutional neural network(FCN)is proposed in this paper.The network can directly regresses the existence strength of room structure lines with a single stage while largely promoting the feature quality.(2)We propose a novel Topology Anchor Points Optimization to automatically identify the layout topology of the input image.It get feature maps extracted by convolutional neural network to optimize the 2D topology structure in further.(3)In this paper,we propose a 3D topology reconstruction algorithm.The algorithm uses only two-dimensional topological anchor points to perform nonlinear optimization with equality constraints and inequality constraints.Finally we can get the final 3D layout of the scene.In order to verify the robustness and effectiveness of our proposed algorithm,we have carried out a relatively complete experimental proof on the public datasets LSUN,Hedau and 3DGP.By comparing the results,we can qualitatively evaluate the effect of the method and pass the calculation results.The average pixel error can quantitatively evaluate the performance of the algorithm.The experimental results show that the proposed method can deal with different layout topologies in different images,and can realize accurate 3D topology reconstruction on various images,which is obviously improved compared with the work in the field in recent years.