Evaluation of Flexible Structured Light Calibration Using Circles

Abstract

We introduce a novel structured light calibration framework that prioritizes calibration flexibility while maintaining satisfactory accuracy. Contrary to existing approaches, our proposed method facilitates the extrinsic and intrinsic projector calibration of a dual-camera (or single RGB-D camera) structured light system by projecting circles onto (non-)planar and potentially dynamically changing scenes across multiple distances. Our method reconstructs light rays by detecting circle-centers, triangulates their 3D positions and filters outliers using RANSAC.Finally, a two-step bundle adjustment process optimizes the structured light parameters. We evaluate our method using synthetically rendered images under relevant scene- and lighting conditions, including the simulation of circlecenter detection errors, different distances, deviations of initial intrinsics and different environments. We also perform an ablation study that justifies the design choices for the components of the algorithm. Our implementation achieves a rotational accuracy of better than 1◦, translational accuracy of better than 1cm and a mean reprojection error of approximately 1 pixel. Based on our experimental results, we expect our approach to be applicable for use-cases in which more flexible structured light calibration techniques are required, such as automotive headlight calibration.