Automated Target-free Camera Calibration

Automatic camera calibration is now a well-established procedure. The process generally involves photographing a target array to form a network of images in a geometric configuration suited to self-calibration. Full automation can be implemented if coded targets are employed, as these provide initial image point correspondences necessary for network exterior orientation. While the use of coded and indeed any artificial targeting facilitates high accuracy recovery of camera calibration parameters, it is also fair to say that the employment of targets can be inconvenient in some practical circumstances, 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. Fortunately, accompanying the adoption of so-called structure-from-motion (SfM) approaches in photogrammetric network orientation, there is the prospect of fully automated camera calibration without the need for artificial targets. Instead of the image-point correspondence problem being overcome through the use of coded targets, feature-based matching is employed to provide the necessary point matches to support exterior orientation. Whereas it is not uncommon to achieve an accuracy of camera calibration of 0.1 pixel or better through the use of coded targets, the resolution from the feature-based matching process is generally closer to 0.3 pixel. This difference is offset, however, by the fact that whereas 100 or so targets might be employed, the SfM approach can easily involve 10,000 or more feature points on a feature-rich object, thus leading to a very comprehensive calibration result. This paper reports on the application of the SfM approach to automated target-free camera self-calibration and discusses the process via practical examples.