Calibration Technique For Phase-Aided Optical3D Measurement Systems

Phase-aided optical three-dimensional (3D) measurement (PAOM-3D) has theadvantages of high accuracy, high density, rapid measurement, simple structure, gooduniversality and flexibility. The PAOM-3D is the most representative method amongoptical3D measurement techniques using structural illumination, and it hasincreasingly extensive applications in various fields including industrialmanufacturing, surveying&navigation, cultural heritage, medicine clinics,entertainment, and so on.The calibration methodology and related techniques of the PAOM-3D system arestudied in this dissertation, which aims to improve the accuracy of system calibrationand find more feasible in situ calibration approach. After briefly introducing theworking principle, research status, development trend and key techniques of thePAOM-3D, this dissertation pays more attention to the image coordinate and the3Dcoordinate of the benchmark, which are two essentials determining the accuracy ofsystem calibration.Circular landmark is the commonly used benchmark in the calibration. In orderto obtain the image coordinate of the circular landmark center, the gray leveldistribution around the circular landmark image is modeled with the surface createdby elliptic rotation of the complementary error function, and then the center can belocated with sub-pixel accuracy using surface fitting algorithm. Experiment results forboth simulated and practical images demonstrate that the proposed method showsgood robustness against the image noise. The eccentricity error of center locationalgorithm caused by the asymmetric perspective projection is discussed, and thiseccentricity error is corrected with the combination of Chen’s camera calibrationbased on circles and Heikkil’s formulation for the eccentricity error.Monocular “Camera-Projector” structure is the common setup for3D sensorbased on the PAOM-3D technique. Since planar target is commonly used in thecalibration of the monocular3D sensor, a planar calibration target pattern, which canachieve auto-coding for benchmarks by using positional relationship invariance, isdesigned for conveniently calibrating the3D sensor. Taking into account theequivalence on the modeling for projector and camera, the principle of bundleadjustment is introduced into the process of calibrating the monocular3D sensor, which can effectively reduce the system calibration error arising from the inaccurate3D coordinates of the benchmarks. Experiment results show that the proposed methodcan reach relatively high calibration accuracy while utilizing calibration target withlow manufacture precision.When measuring objects of large scale and complex topology, a more feasiblescheme is to build up a multi-node optical3D measurement network consisting ofmultiple binocular3D sensors. An analysis of previous literature shows that the3Dcalibration target is the most suitable one for measurement network calibration.Therefore, the coded landmarks which are taken as benchmarks are used to build the3D calibration target that is self-adaptive to the measurement volume, then theEuclidean reconstruction and bundle adjustment are employed to reconstruct the3Dcoordinates of benchmarks accurately. Finally the multi-node3D measurementnetwork can be calibrated conveniently with the reconstructed target. The accuracy oftarget reconstruction and system calibration are well evaluated in the experiments on areduced scale. The proposed approach of measurement network calibration is appliedto an engineering case study, in which a three-node measurement network to inspectthe silica ceramic crucible automatically was successfully calibrated.