Seamless Panorama Stitching Technology Of Fish-eye Image

Panoramic image refers to the image with viewing angle of 360 degrees, to view the whole scene circle around. Panoramic image has a wide range of applications, such as street view maps, three-dimensional GIS. The panoramic image can be obtained by the panoramic camera composed of multiple lenses, or by stitching multiple images captured by a single camera indoor after shooting. Since the panoramic camera is expensive, and the use of multiple lenses requires higher precision of registration, it isn’t applied widely yet. So the use of multiple images stitching technology to obtain panoramic image is now a more widespread method. In the image stitching, we also can use the fisheye image with viewing angle of up to 180 degree to improve stitching efficiency, and speed up the progress of work.This paper mainly analyzed advantages and disadvantages of each module related to the field of image stitching algorithm existed, and on the basis of the improved to some extent by the existing algorithms., and on this basis completed a certain degree of improvement of the existing algorithms, then developed a fisheye image stitching system of practical value by related module algorithm realized. Finally, realized quick browsing the reconstruction of real 3D sense by designing of interface to show the image overlying with the pixel 3D obtained though camera parameters calibration using the known coordinates of the control points.The relevant existed algorithm improvements proposed in the paper mainly includes:Fisheye Image Correction: Traditional fisheye image correction algorithms are based on the 2D model, which is directly to establish the corresponding relationship between calibration image and the original image. Correcting precision of this algorithm is low, and information loss is serious for fish eye image close to the 180 degree view angle on the image edge. The 3D correction model of fish eye image in this paper has a correct angle range of [0°,180°] theoretically, so that is not limited to the view angle of image. Using the Least Square method in solving model parameters makes the error for compute results of model parameters minimized.Using the Successive over Relaxation method makes the procedure run faster, but also more robust, then means that it still can reach high precision in a large amount of computation time for solving linear equations in the solution of image homography matrix compared to the traditional Gauss elimination method, the Square Root method, and-Cramer rule method.