实时多波门图像归一化相关算法的设计与实现

Design and realization of real-time multi-range-gate image normalized relation arithmetic

摘要:

大模板、多波门的相关算法在图像信噪比低、重复模式较多的情况下具有稳定性好、解算精度高的优点，因此其在视场运动分析应用中倍受关注。本文针对大模板、多波门相关运算实时性的问题，提出了一种并行流水运算结构。该结构能够完成对ITU656数字视频码流进行32×32的模板实时归一化相关运算，依据当前码流在图像中的位置信息，实时切换模板，实现多波门搜索。实验结果表明，该结构相关运算仅延迟1.04 μs，有效地解决了运算实时性问题。

Abstract:

Large templates and multi-range-gate relation arithmetic is highlighted in the application of purview movement analysis due to its excellent stability and accurate calculation under the situation of low Signal to Noise Ratio(SNR) and with more models in image. Aiming at real-timing requirement of large templates and multi-range-gate relation calculation, this article presents a kind of parallel pipelining calculating structure. The structure can calculate normalized relation using the template as large as 32×32 in real time when inputted in digital video coding of ITU656. Meanwhile the structure can change the templates in real time according to the position information of the image in the video coding to realize multi-range-gate searching. The results of experiments show that the structure of relation calculation has resolved the real time problem effectively just with a delay of 1.04 μs.