面向裸眼3D的半全局立体匹配视差优化方法

针对基于双目相机的裸眼3D显示的实时性和真实性问题,研究高精度、高效的双目立体匹配算法来获取高质量的深度信息。采用半全局立体匹配算法获取初步的视差估计量,在此基础上,建立基于一致性检验、中值滤波、多层次均值滤波和视差绘制方法的组合优化方法。实验结果表明,视差优化方法能够使视差的均方根误差下降40%左右、像素错误匹配比率下降20%以上,从而获取高精度的深度信息,为裸眼3D的虚拟视点实时精确绘制奠定技术基础。

Semi-global Stereo Matching Parallax Optimization Method forNaked-eye 3D

Aiming at the problem of real-time and authenticity of naked-eye 3D display based on binocular camera, a high-precision and efficient binocular stereo matching algorithm was studied to obtain high-quality depth information. A semi-global stereo matching algorithm was used to obtain the preliminary disparity estimator. On this basis, a combined optimization method based on consistency testing, median filtering, multi-level mean filtering and disparity rendering methods was established. The experimental results show that the proposed parallax optimization method can reduce the root-mean-square error of parallax by about 40% and the pixel mismatch ratio by more than 20%, so as to obtain high-precision depth information, and it lays the technical foundation for real-time accurate rendering of naked eye 3D virtual viewpoints.