Research And Implementation Of Mobile Augmented Reality System Based On Monocular Visual-Inertial Odometry

With the continuous development of science and technology and the rapid progress of the mobile application market,people have put forward higher requirements for augmented reality technology running on mobile smart phones.As an emerging display and interaction technology,mobile augmented reality technology has attracted more and more researchers and developers' attention.However,the traditional augmented reality technology is still limited to be applied in a small area or a predetermined area in most cases.Because the traditional 3D registration method requires manual marking in the environment,it cannot meet the application requirements under large-scale areas and without prior knowledge.Aiming at the shortcomings of traditional mobile augmented reality technology,a mobile augmented reality system based on monocular visual inertial history meter is designed and implemented in this dissertation.The main work is as follows:Based on the research on unmarked 3D registration technology,a monocular visual-inertial odometry position and pose estimation algorithm based on adaptive prediction-prediction mechanism is proposed,which can realize effective system position and attitude tracking in an environment with unfamiliar and verbose texture features.This algorithm uses FAST features and Lucas-Kanade sparse optical flow as the visual tracking front end,adds feature selection algorithm based on vision attention mechanism and motion prediction to the image feature extraction part,and the adaptive control module for dynamically adjusting the algorithm parameters.The algorithm achieved better accuracy and effectiveness on public data sets,and verified the effectiveness of the algorithm.Based on the research on loop-closure detection technology,a relocation algorithm based on GMS feature matcher is proposed.The algorithm uses the improved DBoW3 as the visual database for retrieving loopback information,and comprehensively considers the feature similarity and effectiveness to optimize the retrieval evaluation function.Finally,the GMS feature matcher is used for loop-closure feature matching and screening.Based on the above improvements,the speed of the algorithm has been greatly improved,and the effectiveness of the algorithm has been verified on public data sets.The practical test results show that the system can achieve pose estimation more stably and achieve a robust augmented reality effect.