Research On Patch Selection Algorithm For Mobile AR Cloud System

Mobile devices have become an important application platform of Augmented Reality technology.In order to experience the convenient life brought by AR technology anytime and anywhere,the application of mobile AR technology is gradually rising.With the development of mobile AR technology,the drawbacks of mobile devices are gradually emerging.Because mobile devices are inherent embedded systems,there are shortcomings in database scalability and computing power,and the performance of mobile AR applications is also limited.The performance of mobile AR system includes the scalability of tag content and the real-time performance of the system.How to take into account the scalability and real-time performance of mobile AR system has become an urgent problem to be solved in the development of mobile AR technology.Based on the above problems,this paper introduces cloud technology to solve the current dilemma of action AR system.Related work and research show that the cloud itself is extensible and can share computing and reduce system maintenance costs for mobile devices.This paper explores the solution based on cloud technology,and discusses the cloud server scheme attempt and action AR cloud scheme optimization.In this paper,two schemes,S-MARC and D-MARC,are designed.The former makes a bold attempt on the idea of cloud technology support action AR and studies more possibilities of combining with positioning information,while the latter discusses the design optimization scheme of cloud-based action AR in detail,carries on the image patch interception and the key patch selection optimization to the mobile client,and carries on the classified storage to the cloud database.The image patch interception module proposes Improved-YOLO neural network to detect the target of the image and find out the part that the user is really interested in.The key patch selection module selects the most representative key patch for calculation and transmission based on the image similarity calculation,thus reducing the traffic bandwidth consumption of the system and simplifying the amount of computation.The cloud server uses ImprovedYOLO neural network to classify and sort out the storage content in the database in order to reduce the matching time.In this paper,30 volunteers were invited to carry out the experimental research.Their user evaluation shows that the mobile augmented reality cloud scheme can be implemented in the two systems,and good feedback can be obtained in the evaluation.The experimental results show that the two schemes proposed in this paper have good tracking performance in common camera motion and are superior to other schemes in real-time performance index,and help to improve the current dilemma of action AR and build users' confidence in mobile AR applications.Finally,this paper summarizes the subject and discusses the parts of the subject that need to be improved.