Research On Key Technologies Of Automatical Panoramagram Generation Based On Binocular Fisheye

Nowadays, panoramic cameras are becoming increasingly popular in computer vision an virtual reality. Traditional virtual reality build the virtual environments generated by 3D graphics, in which the information to represent the environment is kept internally as geometry and texture maps. Using panoramic images to generate virtual environments enables realistic exploration of indoor and outdoor environments, which is much more immersed. Since fisheye cameras have wide field of view, it’s a tendency to equip fsheye cameras in panoramic cameras. Because of the distortions of fisheye images, it is hard to generate 360-degree images by fisheye images. In this paper, we will discuss how to generate panoramic images with two fishey images.By establishing the correspondence between equirectangular and fisheye, we can link the equirectangular with fisheye including extrinsic parameters and radial distortion. We’ll design a special calibration environment to create more feature points. Then using SIFT to find the control points between two equirectangular images. By using these control points to optimize, we’ll obtain the extrinsic parameters and the coefficients of radial distortion function. Finally, a spherical panorama is obtained by stitching two equiratangular. Comparing with other methods, we find out the approach in this paper is more quickly, easy to be implemented and effective.Considering inaccuracy of the parameter optimization method, we’ll altere the way of getting the intrinsic parameters, which is called chess-board calibration. By establishing the correspondence between planar pattern and fisheye, we’ll obtain the image projection function of fisheye cameras. To test the proposed method, we’ll compare with parameter optimization method, and notice that the new method was much easier for equiratangules to stitch. The improved methods is more accurate for the estimates of intrinsic parameters, which can effectively reduce the error of stitching tapes.Through studying the two models, we notice that there was same “ghosting” when differece depth of field was stitch by a same depth. Considering the depth of field caused the “ghosting”, we’ll calibrate the images by using four different kinds of depth to reduce the parallax error. In order to test the method, we’ll use the method to handle a couple of fisheye images, and compared with the method before. This method can balance the error of different depth, and has a certain effect to eliminate ghosting.