Photogrammetric Area-Based Least Square Image Matching for Surface Reconstruction

Abstract

In computer vision and robotics community, anormalized cross correlation image matching is widely known as a robust method to determine conjugate points between two overlapping images. In photogrammetric community, however, this method is less favor due to stringent requirements of precision. To achieve such a high standard, a least square adjustment is utilized to minimize a cost function of the image matching process, and then the sum of the residual errors of the cost function is employed to judge the precision and reliability of the match. This paper elaborates the least square image matching adjustment to match conjugate points for surface reconstructions in a highly convergent imaging network or in a wide baseline of stereo images.