大幅宽面阵相机多片探测器影像拼接方法

A Stitching Method for Multi-Detector Images of Large Width Area-Array Cameras

大幅宽面阵相机通常采用多片探测器拼幅成像的方式来增大幅宽。为了提高数据的使用效率,用户要求提供几何无损和视觉无缝的拼接产品。由于分片探测器间重叠区域小甚至无重叠区,拼接后影像畸变增大等原因,传统像方拼接方法难以适用。针对此问题,文章提出一种基于传感器校正的大幅宽面阵相机多片探测器影像高精度几何拼接方法。该方法首先根据卫星相机分片探测器的物理结构及其成像几何特性,虚拟一个完全理想的无畸变面阵影像,影像覆盖范围与原始分片探测器影像覆盖范围相同;在严格标定分片探测器像元指向的基础上,建立虚拟影像与原始分片影像的像点坐标换算关系,并将原始分片影像的像点灰度值赋给虚拟影像对应像点,进而实现对分片影像的无缝拼接和畸变校正。经对仿真数据和在轨数据的试验验证,结果表明:对于不同探测器拼接方式和搭接区重叠情况,文章中所提方法可获得视觉无缝和几何无损的拼接影像,影像的片间拼接精度和影像内部几何畸变均优于1个像素。

Most of the large width area-array cameras use multiple detectors stitching imaging to increase the width.In order to improve the efficiency of data use,users make a request to acquire geometrically lossless stitching products.The traditional stitching method is difficult to apply due to the small or even no overlapping area between detectors and the increased distortion after stitching.Aiming at this problem,this paper proposes a sensor corrected method based on virtual area-array imaging of the object space to realize high-precision geometric stitching for multi-detector images of large width area-array cameras.This method firstly simulates a completely ideal undistorted area-array image based on the physical structure and imaging geometric characteristics of the multiple area-array detectors of the satellite camera,and the virtual image coverage is the same as that of the original slice detector images;then,the two-dimensional orientation of the pixels of each area-array detector is strictly calibrated;finally,based on the object consistency,the pixel coordinate conversion relationship is established between the virtual image and the original image,and the pixel gray value of the original image is assigned to the virtual image to achieve seamless stitching and distortion correction.The experimental results of simulation data and on-orbit real data show that,for different detector stitching conditions and overlapping areas between detectors,the method proposed in this paper can obtain visually seamless and geometrically lossless stitched images,and the image stitching accuracy and internal geometric distortion are better than 1 pixel.