Research On Markless Augmented Reality Algorithm Based On SLAM And Its Application In Cultural Relics Display

Augmented reality technology can increase the sense of reality and interactivity for the display of cultural relics.However,most of the current augmented reality algorithms are implemented based on markers,which cannot get rid of the use restrictions of markers,thus greatly reducing the sense of reality and immersion of augmented reality applications.The design and implementation of unlabeled AR system has always been the focus and difficulty of AR research.In this thesis,Simultaneous Localization And Mapping(SLAM)is used as the basis of AR system.However,due to the size uncertainty of monocular SLAM,the SLAM system must be optimized to be more suitable for augmented reality systems.In addition,SLAM only provides information such as pose estimation and sparse map,on which further spatial perception and analysis are needed to design a complete unlabeled augmented reality system.Since AR application requires high real-time performance,it is necessary to optimize the speed of each module.This research is supported by the national key research and development program sub-project "digital cultural resources of folk tales of folk music and exhibition platform of museum collections".The main research work of this thesis is as follows:1.A method of visual inertia odometer is implemented.This method combines the image information of the camera and the information provided by the inertial measurement unit sensor,provides the dimensional basis for the interaction between the real environment and virtual objects in the augmented reality application,and solves the problems of scale drift and scale uncertainty of the monocular SLAM system.2.A fast visual distance calculation method based on optical flow method and an adaptive selection execution method are designed.The method of fast vision mileage calculation solves the problem of real-time performance of augmented reality system and significantly improves the speed of feature tracking.In this thesis,an adaptive execution algorithm is proposed which adaptively selects the tracking module according to the change of the inertial information sensor.3.A depth information densification algorithm and a fast plane recognition algorithm are designed.The two algorithms solve the problem of spatial perception requirements.The depth information densification algorithm solves the problem that the 3D depth information provided by the SLAM algorithm is too sparse.The sparse depth information can be quickly transmitted to all pixels,and a smooth depth map can be generated as the input information for the depth determination of the virtual model and real scene.Theplanar recognition detection algorithm is based on PROSAC algorithm for fast and real-time planar recognition detection.The obtained plane can be used as the dependent plane of 3D virtual model.Finally,based on the above work,the drawing of the virtual model is completed,the complete augmented reality system is realized,and an application for the display of terracotta warriors and horses relics based on augmented reality is developed.