Combined Bundle Adjustment of Panoramic and Central Perspective Images

Based on a strict mathematical model for rotating line cameras developed at the Institute of Photogrammetry and Remote Sensing of the Dresden University of Technology, a self-calibrating bundle adjustment of panoramic image data was implemented. This model has been extended by the option to involve central perspective images of a conventional camera. A panoramic bundle adjustment is a suitable tool for precise 3D modelling of indoor scenes or city squares in combination with the analysis of very high resolution texture information, as only a few panoramas are necessary for stereo capture of 360° scenes. When recording an object with a complex structure, some parts of the object may not be captured with these panoramic images. In this case it may be desirable to take additional central perspective images using a conventional digital camera in order to avoid such “hidden” object areas. These additional images shall be integrated in the panoramic bundle adjustment in order to warrant full stereoscopic coverage and at the same time stabilize the bundle geometry. For this purpose it was necessary to combine two different sets of observation equations, one for panoramic and one for central perspective images, in a single adjustment calculation. In this paper the combined adjustment will be presented as well as results based on a 360° test field consisting of ca. 120 retroreflective targets. Fig. 1: 360°-Panorama of test field (courtyard within the campus of Dresden University of Technology)