摘要
基于深度图像的绘制(DIBR)广泛应用于虚拟视点的合成,但是目前实现DIBR的算法复杂度都比较高,很难较实时地应用到3DTV系统中。采用单路纹理图像和其对应的深度图像进行虚拟视点的合成,在图形处理单元(GPU)上应用CUDA(Compute Unified Device Architecture)技术实现了基于深度图像的绘制。通过在NVIDIA Telsa C2050图形卡上运行,绘制分辨力1 024×768和640×480的图像速率分别达到了15 f/s(帧/秒)和24 f/s,分别能够准实时或实时地应用到3DTV系统中;同时本文的绘制方法有效地节约了传输带宽,绘制图像的主观质量良好。
Depth-Image-Based rendering is widely applied in virtual viewpoint synthesis, but at present the implementation of DIBR algorithm is too complex to be applied to the real-time3DTV system. A virtual viewpoint synthesis method which utilizes a single texture image with its corresponding depth map is presented in this paper. The proposed method is implemented on the graphics processing unit （GPU） through CUDA （Compute Unified De- vice Architecture） technology. Running on NVIDIA Telsa C2050 graphics card to render images with resolutions of 1 024 × 768 and 640 × 480, the pro- posed method reaches rates of 15 fps and 24 fps respectively. This method can be applied to less demanding real-time 3DTV system. Meanwhile it is not only effective in saving transmission bandwidth, but also achieves better rendering quality subjectively.
出处
《电视技术》
北大核心
2012年第11期11-14,26,共5页
Video Engineering
基金
国家自然科学基金重点项目(60832003
61172096)
上海市科委重点项目(10510500500)
上海市教委科研创新重点项目(09ZZ90)
