Information Annotation And 3D Tracking Of Single Object In Augmented Reality Scene

Augmented reality(AR)is a technology of virtual and real fusion.It projects virtual information into a real three-dimensional scene,which can bring surreal visual experience to users in the real world and provide an intelligent way for human beings to understand the surroundings.Among of existing AR technologies,one extensive application is attaching annotations to real-world object for explanation,which helps human beings recognize the object more quickly and intuitively.In order to add these labeling information to real objects in real time and accurately,a stable and efficient 3D object tracking method is necessary.However,existing work cannot preserve annotation constraints(readability,layout and visual continuity)in a low-computational AR setting.In the augmented reality scene,there will be occlusion,crossover and other problems,which will have a negative impact on the user's receiving of the relevant information of the object.The existing mainstream 3D object tracking platform also has shortcomings in terms of real-time performance and stability.Some problems,such as jitter and offset,will occur when information is marked on real objects,seriously affecting user experience.In this paper,we propose a two-phase continuous annotation optimization method for achieving effective and efficient real-time annotation on 3D objects.The approach includes an offline analysis phase and a real-time rendering phase:the offline analysis phase aims to optimize the annotation layout of a set of virtual cameras,under the guidance of readability,layout constraints and visual coherence;the real-time phase leverages those information to adapt the real-time layout based on realistic camera positions.Based on network architecture,we designed a set of 3D object tracking system,which draws on the advantages of various tracking methods.We put the pretreatment module and target detection module needing large amount of calculation on the server,at the same time put the target tracking module,which real-time demand is high,on the client.The server and the client communicate instantaneously through the network module,which realizes the real-time detection,matching and tracking,and provides a stable platform for the annotation placement.The main contributions of this paper are as follows:1.This paper proposes a method to annotate the information of a single object in the augmented reality scenes.This method optimizes the overall layout from the perspective of visual coherence and significantly reduces the overall moving distance of labels.It can satisfy the constraints of layout optimization in real time.The users can get the annotation information without ambiguity whether they are in the state of motion or in the state of rest.2.A 3D object tracking system based on network is constructed.The system combines the existing methods,preprocessing and target detection in the server and real-time tracking in the client,which effectively improves the efficiency of 3D object tracking.Through experiment test,it can meet the real-time requirement of augmented reality scene.3.A two-dimensional label layout optimization algorithm is proposed.The algorithm calculates the location of labels based on the center point and 2D convex hull of the graph,which can obtain a layout close to the shape of the graph.