摘要
由于引入光流的路径插帧算法中存在边缘或遮挡问题,从而导致路径集构建失败,存在使算法健壮性降低的不足。为了克服此不足,文中通过对现有的引入光流的路径插帧算法的深入研究,使用邻域光流的方向指定路径的大致方向,提出跳转点及其邻域内的光流组合共同指导路径的构建,并利用CUDA模型进行GPGPU加速。使得改进后的算法加快了生成中间帧的速度,并防止了由于边缘或遮挡而导致的路径构建失败的问题,在提高算法健壮性的同时满足了实时性的需求。
Since path-based interpolation method involved with optical flow exists path construction failure caused by edges or occluded regions,the robustness of the algorithm decreases highly. In order to conquer the draw back,conduct a deeper analysis about existing path interpolation method with optical flow,use the direction of optical flow in neighborhood to decide the direction of path,effectively prevent path construction failure by using both the transition point and its optical flow in the neighborhood to construct path set,and greatly accelerate the speed of generating intermediate frames by using CUDA model for GPGPU acceleration,improving the robustness of the algorithm while meeting real-time requirements.
出处
《计算机技术与发展》
2015年第3期11-14,共4页
Computer Technology and Development
基金
国家重大科技专项课题(2009ZX04001-111)
关键词
光流法
图像插值
基于路径
能量最小化
CUDA模型
optical flow method
image interpolation
path-based
energy minimization
CUDA model
