仿射重投影混合虚实注册方法

Affine reprojection hybrid virtual-reality registration method

针对特征图形结合自然纹理特征的虚实注册误差较大、运行速度慢的问题,提出仿射重投影混合虚实注册方法。通过建立摄像机、真实场景以及特征坐标系之间对应关系的仿射坐标系,且采用ORB算法进行特征检测及后续特征模板的构造,并利用仿射重投影矩阵来恢复特征模板在当前图像上的投影位置信息以计算虚实注册矩阵。采用Open GL生成的立方体模型与真实场景之间进行虚实注册仿真实验。仿真结果表明:在特征图形被遮挡或不可见时,本文方法在精度、健壮性以及速率方面表现优异,提高了增强现实(AR)系统的实时性和鲁棒性。

In order to solve the problem of large error and slow running speed of virtual-reality registration method combining feature marker with natural texture feature, affine reprojection hybrid virtual-reality registration method is proposed. Through establishing affine coordinate system with corresponding relationship between camera, real scene and feature coordinates, and using oriented FAST and rotated BRIEF （ORB） algorithm for feature detection and construction of subsequent feature template, and use affine reprojection matrix to restore the projection position information of feature template on current image to calculate virtual-reality registration matrix. Simulation is carried out using 0penGL generated cube model and real scene. Simulation results certify when the feature marker is occluded or invisible, the proposed registration method is superior in precision, robustness and rate;and real-time and robustness performance of augmented reality（AR） system are improved.