中文摘要：

航空图像光流场是低空运动目标检测与变化信息获取的基础,通常将图像金字塔结构引入数值过程以增强全局收敛性。然而,金字塔结构往往是由底层至顶层的递进方式构建,其忽略图像的几何成像过程,造成微小光流或不能得到光流的问题,导致难以支撑后续建模与分析。本文提出了一种以顶层图像为基准的逆向金字塔结构,首先依据中心投影定量地计算出顶层图像的降采样因子,使得顶层图像光流能够反映所设定的地面目标位移阈值;其次,结合顶层与原始图像,以等比方式确定中间层降采样因子;最后,利用高斯平滑与图像插值得到中间层图像,并形成金字塔。对比试验与分析表明,逆向金字塔可准确地计算航空图像光流场,在抑制地面微小位移方面具有优势。

英文摘要：

Aerial image optical flow field is the foundation for detecting moving objects at low altitude and obtaining change information. In general,the image pyramid structure is embedded in numerical procedure in order to enhance the convergence globally. However,more often than not,the pyramid structure is constructed using a bottom-up approach progressively,ignoring the geometry imaging process. In particular,when the ground objects moving it will lead to miss optical flow or the optical flow too small that could hardly sustain the subsequent modeling and analyzing issues. So a backward pyramid structure is proposed on the foundation of top-level standard image. Firstly,down sampled factors of top-level image are calculated quantitatively through central projection,which making the optical flow in top-level image represent the shifting threshold of the set ground target. Secondly,combining top-level image with its original,the down sampled factors in middle layer are confirmed in a constant proportion way. Finally,the image of middle layer is achieved by Gaussian smoothing and image interpolation,and meanwhile the pyramid is formed. The comparative experiments and analysis illustrate that the backward pyramid can calculate the optic flow field in aerial image accurately,and it has advantages in restraining small ground displacement.