Research Of Machine Vision Applied In Augmented Reality

Augmented reality is a research branch developed on virtual reality technology,which could superimpose virtual three-dimensional objects into real scenes and guarantee their relative position fixed.However,in addition to visual effects,in order to enhance the realism of the user experience and expand the scope of application,the combination of augmented reality and other hardware modules that provide auditory and tactile effects and interact with the real world is also urgently needed.This paper built a machine vision system based on augmented reality technology and image classification technology,implemented the linkage mechanism of smart glasses,server and hardware modules,and simulates the machine vision system,which demonstrates the engineering practical significance of this machine vision system.The content and contribution of this paper mainly have the following four points:Firstly,the paper introduced the new augmented reality-based machine vision system structure,described the functions and implementation principles of the main module smart glasses,servers and multiple MCU-controlled execution units.The process of augmented reality and other hardware modules cooperating based on the TCP/IP protocol is demonstrated by a specific process example.At the same time,it is pointed out that the control core part of the system is the machine vision module,and the result of image classification also determines the output of the augmented reality display and the execution unit.Secondly,this paper compared the traditional feature point detection algorithms of the visual part in augmented reality technology.The superimposed image Gaussian pyramid solved the shortcoming of the ORB feature descriptor without scale invariance.Experimentally,compared with SIFT,SURF and native ORB algorithm,the pyramid ORB algorithm implemented in this paper comprehensively evaluates the efficiency and feature matching rate highest.Then the paper completed the registration and rendering in augmented reality based on the pyramid ORB algorithm,and realized the effect of virtual and real fusion.Thirdly,this paper compares the performance of traditional image classification algorithm and convolutional neural network based image classification algorithm.This paper collects 8000 images from ten categories and divides the training dataset and verification dataset according to the ratio of 3:1.In the traditional algorithm experiment,this paper extracted the image features by using SIFT,SURF,raw ORB,pyramid ORB algorithm and then input the features to decision tree classifier after dimension reduction,and conclusion is that pyramid ORB got the best performance whose prediction accuracy in verification dataset exceeded 75%.Then this paper trained an image classifier by using 16-layer convolutional neural network,and the experimental results show that the classifier performs well and the prediction accuracy reached 98.5% and 97.65% on training dataset and verification dataset,far exceeding the traditional algorithm.Finally,this paper simulated the application scene of this machine vision system.In this application scenario,the user can see the effect of augmented reality fusion by wearing smart glasses,and can also get a message prompt UI according to the different scenes seen by the user.