摘要
设计了一种高质量和快速的立体匹配方法，首先在粗分辨率的影像上获取立体影像的重叠区域和初始视差值，而后采用互信息作为匹配测度并采用一种通过一维路径聚合的全局优化方法得到精确的视差值。匹配方法采用图形处理器进行了加速，得到了很好的加速比。实验结果表明了该匹配方法的正确性和高效率性。
A stereo matching algorithm which is good matching quality and quick is designed. Firstly, an initial disparity and overlap based on coarse-resolution images is got. Then a mutual information is used as matching-measure and through one dimension path aggregation the accurate disparity will be got. The matching algorithm uses GPU to accelerate the process, and a good acceleration ratio will be achieved. Finally, the experiment prove it right and efficient.
出处
《计量学报》
CSCD
北大核心
2014年第2期143-146,共4页
Acta Metrologica Sinica
基金
江西省数字国土重点实验室开放基金（DLLJ201303）
关键词
计量学
无人机
立体匹配
互信息
半全局
图形处理器并行
Metrology
Helicopter
Stereo matching
Mutual information
Semi-global
Graphic processing unit parallel
