High-Accuracy Calibration for a Multiview Microscopic 3-D Measurement System

This article proposes a novel calibration framework for a multiview microscopic 3-D measurement system comprising multiple tilted-cameras with the Scheimpflug condition and one vertical projector. Conventionally, the accurate and independent point correspondences between images and planar targets achieve successful calibration. However, affected by our system's large imaging magnification and anisotropic perspectivity, targets’ poses must be limited in the shallow common focus volume to ensure correspondences’ extracting accuracy of all views. This imaging constraint will lead to insufficiently independent correspondences in calibration computation, where constraints will be insufficient in parameter estimation (under-constraint problems) and bundle adjustment (falling in local optima caused by ambiguous models). To this end, our method first relies on insufficiently independent correspondences to tackle the above problems. In calibrating multiple tilted-cameras, the proposed initialization method incorporates the lens equation as an additional constraint to solve the under-constraint problem. The subsequent multicamera bundle adjustment disambiguates by stereo constraints. Moreover, for the vertical projector, another novel unconventional calibration method utilizes geometric constraints to determine its projection matrix. The experiments prove that the estimation and optimization of parameters for multiple tilted-cameras are significantly enhanced with the smallest reprojection error near 0.1 pixel. In addition, the 3-D measurement result of the standard concentric cylinders has also been applied to demonstrate the accuracy of the vertical projector calibration and the whole calibration framework.