Research On Key Technologies Of Panoramic Video Augmented Reality System

Augmented reality is the superposition of virtual information generated by computer and the real world,and panoramic video is to stitch the captured video to be the dynamic video which can be dragged at any angles.In some protected and inconvenient for users to enter scenes,such as cultural relics,museums,scenic spots and so on,we can capture panoramic video clips and add the virtual information generated by augmented reality technology to the stitched panoramic video background,so as to provide more abundant information for the users.The research introduces the augmented reality technology into the panoramic video to propose the construction scheme of the panoramic video augmented reality system and the key technologies of the system are studied.The main contents and results of this study include the following two aspects:(1)In panoramic video mosaics,against the high complexity of panoramic video stitching algorithm,the stitching cracks of video mosaic and "GHOST" phenomenon caused by moving objects,we propose a panoramic video mosaic method based on improved ORB(Oriented FAST and Rotated BRIEF)feature detection,which not only has better real-time performance,but also can effectively restrain the stitching cracks and "GHOST" phenomenon;(2)In order to reduce the impact of illumination,occlusion and rotation transformation on 3D registration method based on marker,we propose ARTool Kit 3D registration based on dynamic feature template matching.In a complicated environment,the dynamic template matching algorithm has high real-time performance and marker recognition rate,and ARTool Kit registration based on dynamic feature template matching has good stability and robustness;(3)In order to solve the large error and slow running speed of 3D registration based on combination of marker and natural texture feature,we propose affine reprojection hybrid virtual-reality registration method,which can complete the registration when the feature graph are blocked or invisible.The proposed registration method is superior in the accuracy,robustness and speed;(4)In order to solve that the 3D registration based on natural texture feature was easily affected by complicated environment and that the target tracking and detection algorithms were time-consuming and low precision,we propose an improved KCF(Kernerlized Correlation Filter)tracking and registration method.The improved KCF algorithm generally performs better than KCF algorithm,TLD(Tracking-Learning-Detection)algorithm,Struck(Structured output tracking with kernel)algorithm and CT(Compressive Tracking)algorithm because the target location and virtual information have high fusion degree.The improved KCF algorithm also improves the real-time performance,stability and robustness of the augmented reality system.Based on the above research on the panoramic video augmented reality system,we stitch and display the shot panoramic video clips,and then superimpose the virtual information to build a panoramic video augmented reality prototype system,which has good real-time performance,robustness and high practical value.