Indoor scene model completion method based on plane constraint

Abstract

The invention discloses an indoor scene model completion method based on plane constraint. The method comprises the following steps: on the basis of dense indoor three-dimensional reconstruction, carrying out three-dimensional plane detection by using a recycling RANSAC method, segmenting an indoor environment into different plane areas and object areas according to a plane detection result, and then segmenting objects which are separated from each other in space by using an adjacency relation of mesh vertexes in the object areas; then, carrying out hole detection on different plane areas and different object models respectively, completing hole completion tasks respectively, then placing completion results in the same three-dimensional space one by one, and finally completing the three-dimensional model hole completion task of the indoor full scene. According to the method, the interference between holes in different objects is eliminated, the complementation precision of the object model is greatly improved, the complementation effect is improved, and a more robust plane complementation effect can be realized under the constraint of a plane equation when the plane model in a scene is complemented.

Description

technical field [0001] This patent belongs to the field of 3D model reconstruction and 3D model completion in computer graphics. It is an indoor 3D scene model completion method that makes full use of plane constraints in artificial indoor space, and can further improve the reconstruction effect of real-time dense reconstruction. It can be applied to AR and VR fields in the future. Background technique [0002] In the fields of virtual reality (VR), augmented reality (AR) or mixed reality (MR), it is often necessary to interact virtual digital scenes, objects or other digital information with the real environment. Or the interface where the object is transformed into a virtual scene or object, which plays a vital role in the process of interaction between the real environment and the virtual object. [0003] Although the existing 3D reconstruction technology can reconstruct the completed dense 3D model, this reconstruction process often requires a sensor to slowly scan all ...

AI Technical Summary

Problems solved by technology

One is model completion based on 3D point cloud data. For example, the PCN algorithm can complete the missing point cloud, but the point cloud model is poor in describing the details and texture of the 3D model, and often cannot be used in VR and other fields. application

The second is the hole completion based on the dense 3D mesh model, but the existing completion algorithm based on the dense mesh model is only effective for a single object model. There will be mutual interference, which makes the final completion effect very unsatisfactory, so this method cannot solve the hole completion task of the entire indoor full-scene environment model

Images