Research On Self-calibration Method Based On Image Sequences

Camera self-calibration plays an important role in the field of Computer Vision.It’s anecessary step for any application involved in the relationship between2D images and3Dobjects.Because of traditional camera calibration methods generally need auxiliaryequipment(e.g,calibration board),that makes it has significant limitations in some practicalapplications,especially involved in autonomous robot navigation, automated drivingassistant system,object identification and so on.It needs to provide real time3Dmeasurement information and abundant structural imformation,and therefore if has a widelyforeground and huge potential.In this paper,some commonly used methods of camera self-calibration will beintroduced and analyzed based on the research of camera self-calibration methods at homeand abroad in recent years. Finally this paper will use the image sequencesuncalibrated,through trifocal tensor as the breakthrough point, putting forward aself-calbration method based on stratified reconstruction,which combinate with the trifocaltensor.The main work0f research is as follows:(1) Described the research situation at home and aborad,theory value and applicationprospect about self-calibration.(2) Described some commonly used theoretical knowledge,including the basic model ofcamera,the epipolar geometric theory and related knowledge for the theory research of laterpaper.(3) On the research of feature point correspondence among imges,we focus on thecorrespondence method of two views based on SURF(speed up robust features),andfollowly we describe correspondence method of three views based on point transfer usingtrifocal tensor.(4) Described the method of calculation about trifocal tensor based onpoint-point-point,and discussed the advantages and disadvantages of it.Finally proposed theimproved algorithm according to the existing problems of those methods.(5) Described some commonly used methods of camera self-calibration,and proposed aself-calbration method based on stratified reconstruction,which combinate with the trifocal tensor.(6) Using OpenCV2.0powerful processing capabilities(It is an open source library,offers a wide range of related aspects such as image processing, mathematical operationfunctions and classes),established a camera self-calibration system based on OpenCV2.0and VS2008,and provided technical support for further research.