全方位图像展开的并行优化  

Parallel optimization for developing omnidirectional image

作　　者：陈曈[1] 杨东勇[1] 石洗凡[2] 

机构地区：[1]浙江工业大学计算机科学与技术学院,浙江杭州310023 [2]浙江大学计算机科学与技术学院,浙江杭州310027

出　　处：《计算机工程与设计》2010年第22期4862-4865,共4页Computer Engineering and Design

基　　金：浙江省科技计划基金项目(2009C31111)

摘　　要：结合查表法的基于双线性内插值法的全方位图像展开算法较近邻取样法提高了图像的展开质量。当透视图像和全景图像同时作为全方位图像展开时,全方位视觉软件单线程执行时运算量大,影响了全方位图像展开的实时性。为了解决该问题,分析了结合查表法的基于双线性内插值法的全方位图像展开算法的原理,提出了基于多核CPU通过OpenMP程序实现全方位视觉图像展开算法并行化处理。实验结果表明,并行处理大大降低了展开成全景图所耗的时间,视频实时处理能力达到640*480分辨率。Combine with trigonometrical function table checking method, quadratic interpolation method improve the unwrapping quality of the omnidirectional image on a certain degree. While the perspective and panorama images are unwrapped simultaneously from omnidirectional image, the whole unwrapping performance needs to be further improved because of the huge calculated amount of omnidirectional vision software on the single thread. And the unwrapping real-time is to be affected. To resolve this problem, by analyzing the unwrapping algorithms a kind of method is proposed that on the multiple-core CPU, the unwrapping algorithms can be realized parallelization when it is implemented by the OpenMP. The result indicated that it can largely reduce the time of the panorama to be unwrapped, the ability of video real time processing can reach 640＊480 resolution.

关 键 词：全方位图像 展开算法 OPENMP 并行化 全景图 双线性内插值法 

分 类 号：TP39[自动化与计算机技术—计算机应用技术]