A novel approach for automation of stereo camera calibration ... - PAR

Table 2 shows that acquisition of sharp images duringmanual positioning of the calibration target is practicallyimpossible for micro scale observation and very difficultand time-consuming process for larger field of views.Proposed system guarantees the optimal quality of inputimages for calibration by implemented sequence of rotations.Tab. 2. Navitar Zoom 6000 Field of View Matrix [12]Tab. 2. Macierz pola widzenia obiektywu Navitar Zoom 6000ParameterDOF[mm]FOV [mm]Mag.WD = 356 mm,0.25 attachment,0.67X TelescopeLens system setupWD = 113 mm0.75X attachment,3.5X TelescopeLow 13.89 1.73High 1.54 0.18Low 93.62 5.98High 14.66 0.93via GigE Vision standard. For the best performance ofcalibration algorithm, the histogram analysis of acquiredimages is done to guarantee the optimal contrast value.14Fig. 7. Camera positioning module: 1 – piezo actuator, 2 – zoomDC servo with encoder, 3 – focus DC Servo with encoder,4 – CCD cameraRys. 7. Moduł pozycjonowania kamer: 1 – piezonapęd, 2 – silnikDC z enkoderem do regulacji powiększenia, 3 – silnik DCz enkoderem do regulacji ostrości, 4 – kamera CCD23Before starting the calibration procedure, it is required toadjust camera orientations, set the optimal focus positionin the motorized lens system and set the optimal lightingconditions. In order to adjust camera orientations thatwere previously set manually by an operator, the visualfeedback system is applied (fig. 6).Fig. 6. Visual feedback system for cameras orientation adjustmentFig. 6. Wizyjna pętla sprzężenia zwrotnego systemu korekcjiorientacji kamerThe laser beam is used to project the cross shape targetpattern. The visual tracking system for both camerasis applied to detect vertical and horizontal lines.An implemented algorithm is used to adjust camera orientationsto point directly in the centre of target cross. Thesecond important step to validate the calibration procedureis to set the optimal focus value. The implementedalgorithm is based on 2D Discrete Fourier Transformation.Navitar’s motorized lens system uses DC Servos withEncoder (fig. 7) that are controlled via RS-232 port in arange up to 3000 steps. Magnetic Hall-Effect sensors areused to reference position location. Starting from the lowposition limit of DC servo, the maximum focus value issearched and saved as a system parameter. The last stagebefore starting the calibration procedure is to adjust theoptimal lighting conditions for detection of the calibrationtarget. LED ring illuminators with the illuminance of ca.18 klux at the distance of ca. 300 mm from the samplewere used. Exposure time of Basler cameras is controlledImages with lower contrast results in longer executiontime of implemented algorithm for circular grid detection.The calibration procedure requires several images of calibrationtarget in different orientations. Images acquiredfrom both cameras are analyzed during the calibrationprocess. The whole process is automated and doesnot require any action from an operator. Calibration targetpoints are detected autonomously by the implementedalgorithm using full resolution (5 Mpx) images. The calibrationprocess is done using a circular grid containing 65control points. Developed software takes the advantagefrom multithreading which allowed for the implementationof fast executing code. Execution time is less than 2 minuteswith the RMS error value below 0.2.4.ConclusionsThe proposed method is dedicated for stereovision systemfor fatigue process monitoring. A condition of scalability isfulfilled by designed calibration module that allows use ofa set of a wide range of size planar calibration targets.Therefore, it is possible to analyze the 3D surface inboth a micro and macro scale. Ultra-high resolution piezoactuatorsare used for automation of the calibration procedure.The automated procedure avoids positioning calibrationstarget out of focus, which is important due tolimited depth of field of the applied vision system. Theproposed solution reduces calibration time and guaranteesrepeatability of the calibration results below, thereprojection error assumed. Implemented algorithm is fullyautonomous. No specialised knowledge or actions is requiredfrom an operator; that makes the calibration processstraightforward compared to other existing solutions.Further work will focus on optimization of the system,which includes research on defining the finest calibrationtarget. Both the size of the pattern and shape of controlpoints will be considered.Pomiary Automatyka Robotyka nr 2/2013237

NAUKAAcknowledgmentsScientific work created within the framework of the“Innovative Systems of Technical Support for SustainableDevelopment of Economy” Strategic Programme withinthe Innovative Economy Operational Programme.References1. Wang Y.-Q., Sutton M.A., Ke X.-D., Schreier H.W.,Reu P.L., Miller T.J., On Error Assessment in StereobasedDeformation Measurements. Part I: TheoreticalDevelopments for Quantitative Estimates, Strain, 2009,Vol. 45, 160–178.2. Dang T., Continuous Stereo Self-Calibration by CameraParameter Tracking, “Transactions on image processing”,Vol. 18, No. 7, 2009, 1536–1550.3. Giesko T., Dual-camera vision system for fatigue monitoring,Materials Science Forum, Vol. 726, 226–232.4. Tsai R.Y., A Versatile Camera Calibration Techniquefor High-Accuracy 3D Machine Vision Metrology UsingOff-Shelf TV Cameras and Lenses, IEEE “Journal onRobotics and Automation”, 1987, Vol. 4, 323–344.5. Liangfu L., Zuren F, Yuanjing F., Accurate Calibrationof Stereo Cameras for Machine Vision, “Journal ofComputer Science & Technology”, 2004, 147–151.6. Urqunhart C.W., Siebert J.P., McDonald J.P., FryerR.J., Active Animate Stereo Vision, The British MachineVision Conference, Surrey, 1993, 75–84.7. Zollner H., Sablatnig R., Comparison ofMethods for Geometric Camera Calibration using PlanarCalibration Targets, 28 th Workshop of the AustrianAssociation for Pattern Recognition, Hagenberg, 2004,237–244.8. [http://docs.opencv.org] – OpenCV v.2.4.3.9. Weng J., Cohen P, Herniou M., Camera Calibrationwith Distortion Models and Accuracy Evaluation,“IEEE Transactions on Pattern Analysis and MachineIntelligence”, 1992, 967–970.10.Bradley D., Heidrich W., Binocular CameraCalibration Using Rectification Error, Canadian Conferenceon Computer and Robot Vision, 2010.11.Schmalz Ch., Forster F., Angelopoulou E.,Camera Calibration: Active versus Passive Targets,“Optical Engineering”, Vol. 50, No. 11, 2011.12.[http://www.navitar.com]13.Ouellet J.N., Hebert P., Developing Assistant Tools forGeometric Camera Calibration: Assessing the Quality ofInput Images, ICPR 2004, [in:] Proceedings of the 17 thInternational Conference, Vol. 4, 2004, 80–83.14.Sharpe W.N. ,Springer Handbook of Experimental SolidMechanics, Springer, 2008, 589-597.15.Yan J.H., Sutton M.A., Deng X., Wei Z., Mixed-modecrack growth in ductile thin-sheet materials under combinedin-plane and out-of-plane loading, “InternationalJournal of Fracture”, Vol. 14, No. 4, 2009, 297–321.Automatyzacja procesu kalibracji systemustereowizyjnegoStreszczenie: Zagadnienie kalibracji kamer w systemachstereowizyjnych było w ostatnich latach podejmowane przezwielu badaczy. Głównym celem tego procesu jestwyznaczenie transformacji pomiędzy współrzędnymi 3D punktóww układzie globalnym i odpowiadającymi im współrzędnymipikselowymi 2D na płaszczyźnie obrazowania kamer. Corazwiększa liczba obszarów, w których zastosowanie znajdująsystemy stereowizyjne, doprowadziła do specjalizacjiwykorzystywanych w nich metod kalibracji kamer. Aktualnie stosowanesą różnorodne wzorce kalibracyjne, a także metodysamokalibracji kamer. W artykule przedstawiony został autorskizautomatyzowany system kalibracji kamer dedykowany dla systemustereowizyjnego, umożliwiającego monitorowanie procesówdestrukcji materiałów na maszynie wytrzymałościowej. Dla zapewnieniamożliwości przeprowadzenia badań zarówno w skalimikro jak i makro, system ten wyposażony jest w zmotoryzowaneobiektywy, umożliwiające regulację zbliżenia i ostrości.Prezentowany system kalibracji poprzez zastosowanie mechatronicznegoukładu pozycjonowania oraz odpowiedniego typoszereguwzorców płaskich, zapewnia kalibrację kamer w pełnymzakresie obserwacji. Rozwiązanie to umożliwia automatyzacjęoraz gwarantuje powtarzalność procesu kalibracji.Słowa kluczowe: stereowizja, kalibracja, badania wytrzymałościowePiotr Garbacz, MScAssistant at the Institute for SustainableTechnologies – National ResearchInstitute in Radom. He received Masterdegree in robotics from WarsawUniversity of Technology in 2010. Heworks in the area of vision systems androbotics.e-mail: piotr.garbacz@itee.radom.plWojciech Mizak, MScAssistant at the Institute for SustainableTechnologies – National ResearchInstitute in Radom. Responsible forprototypes of mechanical andmechatronics design in the field ofautomatic optical inspection,thermography and erosive wear testing.e-mail: wojciech.mizak@itee.radom238