Design Of Augmented Reality System Based On Image Recognition Technology

In recent years,urbanization and informatization operations have become the main content of military operations in various countries.Compared with the target information expression form of traditional equipment,AR is a technology that can dynamically superimpose information other than human physiological perception into the real environment.The use of AR equipment can more intuitively and diversify the expression of some target information that cannot be directly felt,enhances the combatants' understanding of battlefield target recognition and target information,so that the combatants can react quickly,grasp the opportunity to win,and improve the military's combat informatization level.Aiming at common vehicle targets on the battlefield,based on image recognition technology,this thesis designs a vehicle that can accurately identify vehicle targets in the camera's field of view,and accurately superimpose corresponding virtual information on the target position on the display side for users to intuitively view and interact AR system.The main work of this thesis is as follows:(1)The design of the target image recognition module of the system.Aiming at the problems of the traditional feature detection algorithm,which consumes a lot of time in descriptor calculation and the low accuracy rate caused by image transformation,this thesis studies the optimization method of feature extraction and description combining the scale space feature detection SURF and binary descriptor FREAK.In the feature matching part,this thesis adopts the nearest neighbor ratio and two-way matching strategy to improve the matching accuracy.Finally,the RANSAC algorithm is used to further eliminate mismatches.The outdoor scene test verifies that the method in this thesis can produce accurate recognition results.Experiments show that the average accuracy of this method is better than SURF algorithm and FREAK algorithm respectively.At the same time,the method in this thesis is less time-consuming than the SURF algorithm,which can reduce the time-consuming calculation while ensuring the accuracy.(2)The design of the tracking and 3D registration module of the system.Research the offline part to establish the 3D structure of the target and the online part to use the PnP algorithm to obtain registration information to register the 3D target.Aiming at the time-consuming problem of frame-by-frame feature detection and description,the KLT tracking algorithm based on Gaussian pyramid is studied to quickly track and register the target.The experiment introduces the reprojection error,which shows that the registration error of the design in this thesis is low and stable.By changing the scene conditions to verify that the design in this thesis can achieve the effect of accurately superimposing virtual objects to achieve AR.At the same time,experiments verify that the time-consuming tracking and registration is reduced after adding the tracking algorithm,which shows that the method in this thesis can quickly track and register the subsequent target frames.(3)The design of the virtual and real fusion module of the system and system realization.Use 3dsMax and other software to design and realize the 2D and 3D virtual model of the target.Use the Unity3D development engine to design the UI interface and interactive operations,and complete the design of the AR system for vehicle target recognition.At the same time,the overall effect of the system in outdoor scenes is realized,which shows that the system in this thesis can realize the function of identifying and tracking vehicle targets and superimposing the display of virtual model information,presenting multiple information,smooth interface display and obvious interactive effects,indicating that the system is practical.In summary,this thesis analyzes and designs the system.Design the target image recognition module,study the optimization method of feature extraction and description,so that the accuracy rate is improved and the calculation time is reduced.Design the 3D tracking registration module,study the error-stable 3D tracking registration method,and add the KLT tracking algorithm based on the Gaussian pyramid to achieve fast tracking and registration of the target.The virtual and real fusion module designs the virtual model and interface interaction.The final design realizes an AR system that can accurately identify and track military vehicle targets and correctly superimpose and display virtual models,making the system more practical.