| Popis: |
Industrial 3D cameras are limited by their measuring range. We expand it by attaching the camera to a robot wrist. However, this setup introduces unknown flange-camera and world-base transformations we determine using Robot-World-Hand-Eye (RWHE) calibration. Each robot joint also adds an error we compensate with robot kinematics (RK) calibration. In this thesis, we present a simultaneous RK+RWHE calibration procedure for a 3D camera on an industrial robot. To parametrize the system, we used the modified complete and parametrically continuous (MCPC) model. We use a bundle adjustment approach to find calibrated parameters, which minimize the distance between known point locations and detected calibration points. The calibration points lie on a ChArUco board and we present a robust method to extract them. We evaluate results of calibration for simultaneous RK+RWHE, optimization-based RWHE and two traditional RWHE methods for different numbers of calibration poses. Using the 3D camera as a reference and a leave-one-out approach, we found that RK+RWHE calibration halves the average error of RWHE methods. We show that RK+RWHE methods require >18 (RWHE >12) samples for good results, plateauing at >42 samples (RWHE >18). Qualitative results confirm good point cloud alignement, highlighting the need of RK+RWHE calibration for high-precision 3D reconstruction of scenes much larger than the camera measuring range. Industrijske 3D kamere imajo omejeno merilno območje. Merilni prostor povečamo tako, da pritrdimo kamero na zapestje robota. S tem v sistem dodamo dve neznani transformaciji vrh-kamera in osnova-baza, ki ju dobimo s kalibracijo robot-osnova-roka-oko (RWHE). Dodamo tudi napako v členih robota, ki jo zmanjšamo s kalibracijo robotske kinematike (RK). V tej nalogi predstavimo simultano RK+RWHE kalibracijo sistema 3D kamere pritrjene na vrh industrijskega robota. Za parametrizacijo sistema smo uporabili modificiran popoln in parametrično zvezen (MCPC) model. Kalibrirane parametre dobimo z nelinearno minimizacijo razdalj med zaznanimi in znanimi kalibracijskimi točkami. Te točke zaznamo na ChArUco plošči z robustnim algoritmom. Rezultate kalibracije ovrednotimo za simultano RK+RWHE, optimizacijsko RWHE in tradicionalno RWHE kalibracijo za različna števila vzorcev. S 3D kamero kot referenco in pristopom izvzemanja enega vzorca ugotovimo, da RK+RWHE metode razpolovijo povprečno napako v primerjavi z RWHE metodami. Primerjali smo natančnost kalibracije v odvisnosti od števila vzorcev in odkrili, da RK+RWHE metode potrebujejo >18 (RWHE >12) vzorcev za dobre rezultate in se ustalijo pri >42 (RWHE >18) vzorcih. Kvalitativni rezultati potrjujejo dobro ujemanje oblakov točk in poudarjajo pomembnost RK+RWHE kalibracije za 3D rekonstrukcijo scen veliko večjih od merilnega območja kamere. |
