Hüperboloidpeegliga roboti kaamera pildi kalibreerimine
Date
2012
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Tartu Ülikool
Abstract
Käesolev töö käsitles ühte tehisnägemise arendamisega seotud probleemidest - kuidas luua seos robotile läbi kaamera laekuva pildi ja reaalse maailma vahel. Seose loomiseks on vajalik kaamera kalibreerimine testpildi põhjal. Uuriti võimalust sellise programmi kirjutamiseks, mis on võimeline genereeritud piltide põhjal esitama teisenduse pildi ja mingi kindla reaalse maailma tasandi vahel. Ülesande lahendamiseks on mitmeid üldisi võimalusi, mis põhinevad enamasti kaamera parameetrite hindamisel. Töös käsitletud lähenemine eeldab, et soovitakse kalibreerida mingi kindla tasandi suhtes ruumis.
Lahenduse üldidee seisnes ühe valge ruudu kasutamises kaamera kalibreerimiseks. Kõigepealt tuli kalibratsiooniruut tuvastada. Seejärel rakendati teadmist, et füüsilises ruumis oli kalibratsioonimuster ruut. Observeeritava ruumi ja füüsilise ruumi vahelise teisenduse leidmisel kasutati bilineaarset seost. Võrrandite rohkuse tõttu rakendati kiireima laskumise meetodit, et minimiseerida kõigi küljepikkuste vigasid. Lahenduskäiku rakendava programmi kirjutamiseni ei jõutud.
Töös kirjeldati mitmeid varemloodud või praegu arenduses olevaid kaamera kalibratsiooni tööriistu. Kirjeldustes toodi välja tööriistade vahelisi seoseid ja eripärasid.
Õpiti kasutama POVRay tarkvarapaketti katseandmete simuleerimiseks, lisaks sellele süvendati programmeerimiskeele Python kohta käivaid teadmisi.
Implementeeriti proof of consept tüüpi programm, mis kasutab modifitseeritud versiooni W. Sun [41] poolt kirjeldatud kihilisest nurgatuvastusmeetodist.
Esitati viis pildi kalibreerimiseks vajalikust bilineaarsest teisendusest koos näidislahendusega üksikute kaadrite juhul.
Töö edasiarendamiseks tuleks kirjeldatud kalibratsioonimeetod implementeerida ja seejärel seda vastavalt reaalse maailma katsetulemustele optimeerida.
The aim of this paper is to present a method for camera calibration. The actual implementation of the calibration process itself is not included in this paper, the solution is only theoretical. The camera calibration is considered in this paper in the meaning of spatial calibration and not in the sense of photometric camera calibration. The calibration problem arose in the robotics competition Robotex. The exact location of objects on a certain plane needs to be estimated by the robot in real-time. This means that there is a need for mapping between pixels and real world distances. This paper presents a set of existing methods for solving this camera calibration problem. The test data is generated using POVRay ray-tracing software. This enables predictable test cases for the software. A white square sheet of paper is used as the calibration pattern. A simplified version of W. Sun’s [41] corner detection algorithm is implemented to extract the location of the calibration pattern from an image. After the corners have been extracted, a method based on bilinear interpolation is proposed to calibrate the camera. The information that the calibration pattern is a square in the physical world is used in the calibration method. The proposed method suggests that using more frames increases the accuracy of the calibration. In order to improve the accuracy, the image is divided into subsections that are assumed to have a bilinear transformation from the physical world to the observed image. The next research in this field should implement the suggested method to verify it’s accuracy.
The aim of this paper is to present a method for camera calibration. The actual implementation of the calibration process itself is not included in this paper, the solution is only theoretical. The camera calibration is considered in this paper in the meaning of spatial calibration and not in the sense of photometric camera calibration. The calibration problem arose in the robotics competition Robotex. The exact location of objects on a certain plane needs to be estimated by the robot in real-time. This means that there is a need for mapping between pixels and real world distances. This paper presents a set of existing methods for solving this camera calibration problem. The test data is generated using POVRay ray-tracing software. This enables predictable test cases for the software. A white square sheet of paper is used as the calibration pattern. A simplified version of W. Sun’s [41] corner detection algorithm is implemented to extract the location of the calibration pattern from an image. After the corners have been extracted, a method based on bilinear interpolation is proposed to calibrate the camera. The information that the calibration pattern is a square in the physical world is used in the calibration method. The proposed method suggests that using more frames increases the accuracy of the calibration. In order to improve the accuracy, the image is divided into subsections that are assumed to have a bilinear transformation from the physical world to the observed image. The next research in this field should implement the suggested method to verify it’s accuracy.