一种应用于虚拟现实的基于网格变形的立体图像生成方法

摘要

文章针对虚拟现实系统，介绍了立体图像的生成方法。提出了利用单个相机通过不同的投影变换来拍摄双眼图像。基于单相机拍摄的方法，分析了立体图像的视差形成机制，给出了有头跟踪设备和无头跟踪设备两种情况下双眼图像的视差变换矩阵。进一步分析了视差大小、感知深度、感知变形和模型变形之间的联系，将视差的产生归结为模型的几何变形。提出可以通过模型的几何变形来控制视差和感知物体的变形，从而给出了一种基于模型网格变形的非线性深度映射方法。该方法能有效的减少因感知物体的变形而带来的视觉瑕疵，并能消除在头跟踪系统中因深度映射导致的图像漂移。%The paper introduced a novel method for stereoscopic image creation in virtual reality system.In this meth-od,the left-and-right images were produced through different projection transformations in a single camera space,and the two types of projection matrices were derived respectively for stereoscopic head-tracked and non-head-tracked systems. Based on the single-camera method,we analyzed the relations among disparity,perceived depth,deformation of perceived objects and model mesh warping.The disparity and perceived deformation were modeled as 3D warping,so the disparity and the perceived deformation could be controlled by deforming the model mesh.Finally,a nonlinear perceived depth mapping method using mesh warping was proposed,which could reduce the artifacts caused by perceived distortion,and could avoid the image shift caused by perceived depth mapping in stereoscopic head-tracked system when the viewer moved his head left and right.