嫦娥一号电荷耦合元件(charge coupled device,CCD)相机同轨前后视影像组成的立体像对的基高比较大,立体像对在撞击坑边缘、山脉等地形起伏区域存在较大的遮挡和变形,影响了DEM的精度。为弥补这一不足,根据嫦娥一号CCD相机邻轨下视短基线立体像对成像模型,本文提出了一种自适应窗口归一化互相关(normalized cross correlation,NCC)-相位相关(phase-only correlation,POC)视差估计的DEM生成方法。首先,根据影像纹理特征计算各像素点对应的自适应窗口大小,通过NCC方法估计立体像对左右片对应窗口的整数像素位移;然后,用POC方法估计亚像素位移;最后,将窗口的总位移作为窗口中心像素点的视差,依次计算所有像素点的视差,得到立体像对稠密亚像素视差图。模拟实验结果表明,本文方法视差估计精度可达±0.075像素,利用空间分辨率为120m的嫦娥一号CCD邻轨下视影像生成的月面DEM与LOLA 30m平面分辨率的DEM基本一致。
In order to construct a DEM of the moon's surface from narrow baseline stereo matching between nadir images of Chang'E-1 CCD camera of adjacent tracks, a NCC-POC based adaptive disparity estimation method is proposed. The estimated disparity map is used to construct the DEM. In the algorithm, NCC (normalized cross correlation) is applied with a texture based adaptive window to scan the pair of stereo images pixel by pixel. This process determines the integer disparity between corresponding pixels, and the POC (rhase-only correlation) is applied to estimate high accuracy sub-pixel shifts. The total disparity between the corresponding pixels of the image pair is calculated as the sum- mation of NCC measured integer disparity and the POC measure sub-pixel disparity. Via this two step disparity estimation procedure, a dense high accuracy disparity map was generated after essential post- processing to eliminate outlier values. Simulated experimental results show that the proposed method can achieve an accuracy of 0. 075 pixel of RMSE. The DEM generated by the proposed method is consistent with the DEM at a resolution of 30 m, obtained by LOLA.