基于CUDA的快速全景环带图像展开

摘要

全景环形镜头将以透镜为中心的360.视场内的景物投影到环形区域内,具有大视场和无穷远的景深.由于全景环形镜头所拍摄到的图像具有较大的失真,为了方便人眼观看,需要利用坐标变换,将全景环形镜头获得的环带图像展开为条幅状图像.图像变换过程需要大量处理的数据,尤其对于高分辨率图像而言,采用传统的基于中央处理器( CPU)的坐标转换算法耗时较长,图像停顿和滞后严重,无法实现实时显示.为解决此问题,文中提出采用最新出现的基于通用图形处理器( GPGPU)的并行处理技术实现图形的展开,大规模并行计算的使用缩短了图形展开的时间,从而实现了大视场、高分辨率、实时监控显示.%Panoramic Annular Lens (PAL) can form annular image circling the optical axis with field range of 360 degree and infinite depth of field. The annular image captured by PAL must be stretched into a rectangular image in order to be observed and measured conveniently. For a high resolution image, the traditional method of stretching process conducted by CPU will cost too much time. Propose a novel method using the latest GPU technology - CUDA. With this method, it can achieve a processing speed 10 times faster than traditional processing method by CPU, which makes the real-time high resolution panoramic detection possible.