Real Scene Modeling Based On Multi-level Fusion For VR

With the advent of VR boxes in recent years,virtual reality technology has attracted worldwide attention and achieved rapid development and tremendous progress.Virtual reality provides a new platform that can be widely used in military,technology,games,education and other fields.At present,the virtual scene captured by the VR panoramic camera only has 3DoF viewing freedom.To obtain 6DoF degrees of freedom,the scene must be modeled in three dimensions.The virtual scene in the 3D game is generally modeled by professional 3D modeling software,and the real scene first needs to perform three-dimensional measurement on the actual scene through the active method or the passive method,and then convert the measurement data into a three-dimensional model.However,both active and passive methods have their own limitations.The three-dimensional models built often have various problems such as voids and noise.This three-dimensional model with sputum applied to VR will seriously affect the imaging effect.This paper explores the VR 3D modeling of real scenes.The results of the full text are summarized as follows.This paper proposes a VR real scene modeling method based on multi-level 3D point cloud fusion.This method firstly acquires the sparse seed 3D point cloud of the whole large scene through multi-view 3D reconstruction technology,and uses this 3D point cloud as the skeleton to integrate.The three-dimensional point cloud of each sub-area collected by the ZED camera is finally filled and repaired by the local three-dimensional point cloud collected by the SENSE depth scanner.Through this multi-level fusion,a perfect 3D model of a large real scene suitable for VR is obtained.In the multi-level 3D point cloud fusion of large scenes,the two types of 3D point cloud are inconsistent and the problems caused by density inconsistency are explored.A PSO-ICP point cloud registration method is explored.The method firstly preprocesses the 3D point cloud,and then uses the PSO algorithm to coarsely register the point cloud data so that its relative position is closer to the real position.Finally,the registration result is used as the initial position of the ICP algorithm registration,and further Reduce the error of point cloud registration and improve the registration accuracy.In this paper,HTC VIVE is used as the hardware platform,and Unity3D is the software platform.The 3D model of VR real scene obtained based on multi-level fusion is imported.The VR virtual scene which is very close to the real scene is created,and the 6DoF degree of freedom is realized.After wearing the HTC VIVE helmet,You can walk in a virtual scene as if you are in a real scene.The multi-level fusion based 3D modeling method studied in this paper makes full use of the advantages of various 3D reconstruction techniques,and can realize VR modeling of large real scenes,achieving 6DoF effects,which can be applied not only to the VR field,but also to further expand.Applied to areas such as AR.