摘要
立体视觉计算过程中的匹配值计算极其耗时,整幅图的匹配值由各个参考像点的匹配值计算构成,因此,如果能够将前后两次参考像点的匹配值计算中重复的部分提取出来,那么就可以节约部分计算时间。将前后两次匹配值计算所涉及的领域像素分布情况分析清楚,结合所选取的计算步骤,将各个子步骤的中间输出结果缓存起来,以便下一次计算时直接使用而不需重新计算。如果匹配所用的邻域子图与整幅图相比非常小,则这种类似于流水作业的时间重叠计算方式可以获得很好的加速性能,同时该加速性能也会随着邻域尺寸的加大而提高,这得益于所省略的中间计算步骤较多的原因。
It is very time consuming as matching value computing in binocular stereo vision, which all the referring pixel will be figured out a matching value according to the near area of peer image. If the redundant computation between two conjunction pixels can be found, a faster method will be derived. By analyze the process of matching computing and the relative position of the near-areas, the computing result of sub-steps which is identical in these two pixel will be buffered and can be use in next step immediately. If the near-area is much smaller that that whole image, a very ideal speed up can be obtain similar to streaming technology and the speed up increase as the near-area getting larger.
出处
《微计算机信息》
北大核心
2007年第05X期293-294,312,共3页
Control & Automation
基金
广东省自科基金(5301033)
关键词
匹配值
视差
双目立体视觉
快速计算
Matching Value,disparity,binocular stereo vision,fast computing
