Research On The Method Of Camera Self-Calibration Based On The Vanishing Points

Camera calibration is an important technology in the field of machine vision. Itis mainly used in mobile robot navigation, scene reconstruction, visual surveillanceand virtual reality and other related fields. Camera calibration can be divided intothree categories: The first method is the traditional camera calibration, which isbased on the accurate three dimensional calibration target. The second method is thecamera self-calibration. This method is not related with specific targets and specificscenes. The camera self-calibration uses the self-constrains of the camera forcalibration. There is also another calibration method, which is based on the activevision. It can precisely control the parameters of the motions of the camera (such astranslation or rotation) for calibration. Although this method eliminates thedependent of traditional method about three-dimensional target, however, it alsorequires a known accurate camera platform. There are still some other calibrationmethods, such as the method of Zhengyou Zhang, which is based on the flattemplate target and the methods of based on the algorithms of neural networks orgenetic algorithms and so on.This paper proposed the methods of camera self-calibration based on thevanishing points. In the projective geometry, the vanishing point is the image pointof infinity point. In the space, the infinity point of a line represents the direction ofthe line, which is unrelated with the location of the line and coordinates of the points.Putting the infinity point into the projection transformation formula, the fourthcolumn vector of homography (displacement column vector) is zero, this means theprojective matrix between the infinity points in space and the vanishing points ofimage is unrelated with the displacement of the camera. It can get a correspondentmatrix between two images which is not related with displacement of camera. Forthis, it makes the camera freely move in space and it has no effect to the calibration.This paper mainly proposed two self-calibration methods, which are all based on thethe transformation mastrix, using the vanishing points to compute between twoimages. The one self-calibration method is using a transformation matrix and anadditional constraint condition for calibration. It maks only use two images forcalibration. And the other method is using at least two different transformationmatrixes for calibration. It can accurately and stablely calibrate by using moreimages. For the introduction of the vanishing point, the camera can get the images ofthe target at any location with any angle in space. Thus, in some certainenvironments (in the environment, it can easy access the vanishing point), the twomethod proposed in this paper can achieve on-line self-calibration. The online camera self-calibration means that the camera can be auto re-calibrated withoutstopping the ongoing task when its parameters change when being used.