基于深度信息虚拟视点生成

Virtual Viewpoint Generation Based on Depth Information

虚拟视点生成是指通过两个或者多个摄像机,拍摄同一个三维场景不同角度的图像,根据这些已知图像可以通过某种算法计算出两个摄像机中间虚拟视点处的图像。不管是在计算机图形学领域还是在计算机视觉领域虚拟视点生成都是当前的研究热点之一,并且虚拟视点在很多方面都有较好的应用前景,例如远程医疗、远程会议、立体电影、自由视点电视等。提出一种能够生成虚拟视点的有效算法,能很好生成中间视点的图像。算法包括三步:深度预处理,基于深度信息的反向投影,重投影。对于实验结果,使用PSNR与SSIM进行检验,证明合成的图像在客观和主观上都比较自然。

Virtual viewpoint generation refers to taking images of different angles of the same 3D scene by two or more cameras. Based on these known images, the image of the virtual viewpoint in the middle of two cameras can be calculated by some algorithm. Whether in the field of com- puter graphics or in the field of computer vision, virtual viewpoint generation is one of the current research hot spots, and the virtual view- point has good application prospect in many aspects such as telemedicine, teleconferencing, stereoscopic movie, free viewpoint TV and so on. An effective algorithm that can generate virtual viewpoints is proposed, which can generate the image of intermediate viewpoints well. The algorithm consists of three steps： depth preprocessing, reverse projection based on depth information, reprojection. For the experimen- tal results, the use of PSNR and SSIM tests proved that the synthetic images obiectivelv and subiectivelv more natural.