Camera calibration by vision threads with pixel-resolved focus measurement

点击翻译按钮获取中文摘要

Imaging systems like cameras and projectors are widely used for two-dimensional- and three-dimensional coordinate measurement or machine vision. For example, this is done in fringe-projection- and fringe-reflection techniques, where projected or displayed pattern are observed with cameras. The optical imaging device, for instance a CCD or CMOS camera, maps light rays from object space into 2D points in image space. The mapping function (including lens distortion) must be determined in order to build the correct object-image relation. This is done by camera calibration, which are therefore fundamental for visual measurement processes.The classic and common approach of camera calibration assumes the model of a pinhole camera. The photogrammetric calibration based on it has been researched widely and is reaching its limits more often. New research tasks in optical metrology like efficient multi camera- or comprehensive projector or monitor calibration, however, require more accurately and flexible procedures. Instead of further adapting the pinhole model, the more sophisticated approach of the generic camera calibration or so-called vision-ray-calibration describes independent rays for every pixel. It compensates the effects of lens distortions more precisely and is more flexible then other techniques, but still requires thorough investigation to achieve its full potential.With the benefits of the vision-ray-calibration comes the need for intense data acquisition and evaluation. The typical calibration pattern like checkerboards are not sufficient anymore. Instead the phase shifting technique is used on a display to get the required spatial data density. For the calibration it takes, however, a lot of time to reconstruct the exact position and alignment of the display positions, until the vision rays are determined from the data.In this proposal we present an alternative optimization method, which uses vision-threads instead of vision-rays. These connect the viewpoints of a pixel between the different display places. If they become "tensed", positions and alignment are transferred directly into the desired constellation. With this method an optimization should be faster and more accurate than before. Furthermore, we present a way to calculate the focal length for each pixel from the same data, which is at one hand used to support the vision-threads and at the other hand provide users an easy access to the focal information of their systems.