Research On Efficient Organization And Dispatch Methods For Scene Data In Head-Mounted Augmented Reality

With the continuous exploration of the digital world by humans,augmented reality technology,widely known as the new generation of human-computer interaction technology,has attracted widespread attention.In recent years,augmented reality technology has been widely used in military,education,industrial and other fields.Augmented reality can superimpose virtual 3D scene data onto the real environment,providing users with a deep immersive virtual-real fusion experience.However,with the increasing production demand of people,large-scale and various types of 3D scene data are increasing day by day.Due to the obvious geographical and spatial attributes of 3D scene data,how to use geographical spatial information to assist in quickly responding to massive heterogeneous 3D scene data is a key factor in achieving efficient visualization of augmented reality terminal data.Therefore,this paper focuses on the efficient organization and dispatch of massive heterogeneous 3D scene data in head-mounted augmented reality scenes,and the main research work and achievements are as follows:(1)Analyzed and defined the semantic expression features of 3D scene data in augmented reality scenes,proposed a combination method of semantic dimensions based on space,perception and appearance,and based on this,referred to the Geo JSON data standard to define the JSON format of 3D scene data expression specifications.For efficient access to large-scale 3D scene data,based on the characteristics of No SQL databases,proposed the 3DHash 3D spatial data encoding and constructed a spatial data query algorithm based on 3DHash encoding,which greatly improves the efficiency of 3D scene data access.(2)Designed seven typical behavior interaction modes for head-mounted augmented reality devices and built a behavior pattern recognition neural network BINet using long short-term memory networks to recognize the seven behavior interaction modes based on the temporal pose changes of the user,achieving a recognition accuracy of 91%.According to previous research results on visual cognitive load,reasonable data update strategies for augmented reality visualization scenes under different behavior interaction modes are proposed.(3)To quickly respond to 3D scene data loading requests for head-mounted augmented reality terminals and ensure the smoothness and stability of terminal visualization scenes,a multi-level cache dispatch process for 3D scene data visualization scenes for head-mounted augmented reality is designed based on the user’s current pose attributes,using data cache dispatch strategies based on spatial proximity and visual visibility,combined with the research on the organization and behavior pattern interaction of 3D scene data in the previous sections,effectively improving the visualization speed and concurrency efficiency of head-mounted augmented reality terminals.(4)Using Holo Lens 2.0 head-mounted augmented reality devices,a prototype system for visualizing massive heterogeneous 3D scene data with augmented reality is designed and built,and application research experiments are conducted in the Chinese Academy of Sciences’ Tianhe Science Park.The experimental results show that it can still achieve second-level response when facing a data scale of millions,and the frame rate of the terminal visualization scene is basically maintained between 40-50 FPS,effectively avoiding overload of virtual-real fusion scene cognitive load,achieving ideal research experimental effects.