Research On The Self-calibration Method Based On Binocular Image Sequences

The camera calibration is one of the key technologies from2-D image to obtain3-Dinformation, and is the most important role in computer vision, and be mostly used inthe side of3-D reconstruction, navigation, visual surveillance. So the research of thecamera calibration has become the hot research fields of computer vision. Thedifference between the camera self-calibration and the traditional camera calibration isthat obtaining the camera intrinsic parameters by the image itself, and this method canbe applied in on-line3-D reconstruction, therefore it not only has high academic valuebut also has a widely foreground.The research of camera self-calibration has been carried out at home and abroad,but the self-calibration for varying intrinsic parameters based on binocular imagesequence is little, most of them are based on monocular vision camera self-calibrationor camera parameters unchanged. But the accuracy of self-calibration based onbinocular vision in practical application is better than the accuracy of self-calibrationbased on monocular vision，and the parameters of the camera stays the same is notpossible. So vary inner parameters of self-calibration based on binocular vision is moreacademic value. The paper researches some methods of self-calibration for varyingintrinsic parameters based on binocular vision. The main content of the work follows:(1). Access to the domestic and international literature about self-calibration, andintroduce its research status. Elaborate the research background, the research value, theresearch significance and the application prospect of the self-calibration.(2). Described the camera model and multi-view geometry which is the theoreticalfoundation of the camera self-calibration.(3). The paper puts forward a kind of binocular vision self calibration methodbased on affine projection model. This method solves the affine projection matrixfirstly, then we can get the part parameters. And then, we suppose the main pointcoordinate is located in the center of the image to work out the rest of the parameters.Finally the camera nonlinear distortion for regional adjustment is introduced toimprove the accuracy of the calibration. The experimental result data show that thismethod has a high accuracy and practical binocular self calibration.(4). This paper presents a new self calibration based on binocular vision. Firstlythe method normalizes the matching point coordinates to calculate basis matrix, then obtains the camera parameters by solving Kruppa equation,then gets the focal of twocamera which are intrinsic parameters of binocular vision system; then use the natureof the nature matrix to get two-camera relative position which are external parametersof binocular vision system; Finally used the residuals of image point coordinate toobtain nonlinear optimization by internal and external camera parameters, till the resultof internal and external camera parameters of binocular vision system is satisfactory.The experimental result data show that this method is availability.