Multi-Camera Calibration With Imperfect 3-D Targets for Imaging Biological Systems

High-resolution multi-view stereo camera systems provide an opportunity for precise measurement of shapes, deformations, and displacements. These systems are especially useful for instrumentation and measurement of small-scale biological systems. Often, these are so precise that their measurements have greater precision than the achievable manufacturing precision of calibration targets, limiting the accuracy of template-based calibrations. This work extends the calibration of camera systems using imperfect planar calibration targets to 3D targets, allowing robust calibration of these high-resolution stereo systems. Results are reported for three separate camera systems, designed for human-scale, skin surface, and single-organ imaging. Recovery of target imperfections, to within 20 μm for a 40 mm target, and improved reprojection errors are demonstrated. The proposed calibration of imperfect targets is shown to be more robust to imperfections in the points chosen to constrain the calibration than previous methods. For systems with relatively precise targets, small improvements are shown in calibration results (from 0.91 pixels to 0.59 pixels of Euclidean error) with limited change in measurements from the system. For imprecise targets, the results show notable improvements in reprojection errors (improving from 4.30 pixels to 1.34 pixels and 5.85 pixels to 0.61 pixels of Euclidean error) and improvements in reconstruction based measurement, reconstructing an additional 10% of the subject.