Network Orientation Using the Scaled Orthographic Projection for Parameter Initialization

Bundle adjustment is a well known and reliable method used in photogrammetry for image network orientation which requires relatively accurate initial approximations of image orientations and point coordinates. The initialization problem has proved difficult and a variety of initialization techniques have been proposed. We present a new method that takes advantage of the scaled orthographic projection to compute direct linear solutions that approximate the image network geometry. The algorithm includes two relative orientation methods adapted from previous work, and incorporates a recently developed orthographic bundle adjustment method. Following initial network orientation, image coordinates are corrected for perspective to obtain an intermediate solution which is converted into perspective projection parameters and a final bundle adjustment is performed. The method has been tested using several image sets and has proven to be effective at various fields of view, with a variety of imaging network geometries, and with different object point geometries.