Automated Target-Free Network Orienation and Camera Calibration

Abstract

ABSTRACT:Automated close-range photogrammetric network orientation and camera calibration has traditionally been associated with the use of coded targets in the object space to allow for an initial relative orientation (RO) and subsequent spatial resection of the images. However, over the last decade, advances coming mainly from the computer vision (CV) community have allowed for fully automated orientation via feature-based matching techniques. There are a number of advantages in such methodologies for various types of applications, as well as for cases where the use of artificial targets might be not possible or preferable, for example when attempting calibration from low-level aerial imagery, as with UAVs, or when calibrating long-focal length lenses where small image scales call for inconveniently large coded targets. While there are now a number of CV-based algorithms for multi-image orientation within narrow-baseline networks, with accompanying open-source software, from a photogrammetric standpoint the results are typically disappointing as the metric integrity of the resulting models is generally poor, or even unknown. The objective addressed in this paper is target-free automatic multi-image orientation, maintaining metric integrity, within networks that incorporate wide-baseline imagery. The focus is on both the development of a methodology that overcomes the shortcomings that can be present in current CV algorithms, and on the photogrammetric priorities and requirements that exist in current processing pipelines. This paper also reports on the application of the proposed methodology to automated target-free camera self-calibration and discusses the process via practical examples.

Network Design and Quality Checks in Automatic Orientation of Close-Range Photogrammetric Blocks

“…In other words, considering the precision of both, a common SfM software package and the GCP, hardly any difference is noticeable when using many tie points (as usually an SfM routine does) or just a few. As reported by Stamatopoulos and Fraser [ 15 ], the problem could be much more serious if the deformation is assessed using a certain number of artificial targets. In that case, the image coordinate accuracy of the points extracted by the SfM on natural features would probably be less precise compared to the ones extracted by image matching on the artificial target, and the stochastic model would probably have produced some unwanted deformation.…”

“…The high number of image observations used by SfM procedures does not in itself guarantee the correctness of the result and the required quality level. Metric performances of SfM should therefore be more thoroughly investigated in order to extend the use of these techniques to image metrology [ 15 ].…”

Digital Workflow for the Conservation of Bahrain Built Heritage: The Sheik Isa Bin Ali House

“…Both lenses were calibrated beforehand, so that calibration parameters were fixed during photogrammetric processing. The approach used for camera calibration is the target-less method proposed in (Barazzetti et al, 2011) and (Stamatopoulos and Fraser, 2014). Rooms were processed in different ways depending on their geometry.…”

Network Design in Close-Range Photogrammetry With Short Baseline Images

“…As mentioned, the camera was previously calibrated by acquiring a set of images of an object with a good texture, following the rules presented in Remondino and Fraser (2006). The method followed the principle of markerless calibration presented in Barazzetti et al (2011) and Stamatopoulos and Fraser (2014). For each software, calibration was carried out independently with the same image dataset.…”