Calibration of a Line-Structured Light Vision System

Sammanfattning

Vision-based measurement systems based on line-structured light triangulation rely on accurate calibration to convert pixel coordinates into millimeter coordinates. Optimation supplies such measurement systems to process industries worldwide. To avoid losing valuable production hours, calibration needs to be quick and reliable. The procedure in use today establishes a two-dimensional mapping from the image plane to the laser plane, producing accurate calibrations. Although the calibration is accurate, it can suffer near the edges of the sensor, and recalibration is a process that can involve multiple exposures of a calibration target at multiple positions within the measurement volume. This thesis develops a two-stage calibration procedure for a Scheimpflug-tilted laser triangulation system. The intrinsic camera parameters are estimated once during system setup from images of a ChArUco target. The laser plane is then recovered from a single exposure of a parallel-cylinder target, using sub-pixel stripe centers extracted with a multi-scale Hessian-based pipeline. The calibration is validated against three reference objects and compared against the existing homography-based calibration. The intrinsic calibration showed that the 3.6° Scheimpflug tilt is absorbed by tangential distortion and explicit Scheimpflug modeling is not required at this tilt. The recovered light plane reproduces the cylinder separation and diameter to within 0.4 mm and 0.003 mm of their nominal values. Reconstruction is close to sub-millimeter accuracy (≈ ± 2 mm error), with a systematic height bias at larger heights. The same bias appears with opposite sign in the existing homography-based calibration.