|
In many applications, need wide viewing angle and even full 360 ° viewing angle of the image in order to obtain the required information in their respective fields. However, due to current technology, and physical structural limitations, the current image can not be captured in a single time to obtain a full panoramic image panoramic image stitching technology for this problem presented its solution. The fish-eye image is used on the camera lens has a wider viewing angle fisheye lens, thereby reducing the number of images and image stitching times, improve the quality of the final image mosaic effect. In this paper, fisheye panoramic image stitching algorithm for each module involves an analysis of the shortcomings of the algorithm proposed a new method and a certain degree of improvement. Include: ① fisheye image correction: the traditional method of latitude and longitude on the fisheye image correction, image shooting high quality requirements, you need to ensure that the image is not possible during shooting occurred pitch conversion, otherwise it will lead to the corrected image loss of vertical invariance thus affecting subsequent mosaic effect. To address this problem, this paper proposes a feature points between two images using the information to calculate the estimate fisheye images when shooting pitch error method, using the pitch estimate fisheye images for precise correction, ensure that the image exists pitch error When, after correction still has a vertical invariance. Denoising feature point ②: For the fisheye image feature points analyzed, a histogram of the image feature point that the positional relationship of statistical methods, as SIFT feature point matching algorithm higher accuracy, Therefore, the input matching feature points of a certain class of polyethylene, by the histogram, statistical feature point represented by the positional relationship between the images, the position information with the same peaks stretching, while the obvious noise suppressing process information , so as to achieve the feature points denoising. ③ transformation matrix calculation: Based on the characteristics of the fish-eye image, after the correction due precisely fisheye image, with only deflection transformation, these points as the a priori knowledge of the parameters using the transformation matrix as the two fish-eye image position transformation matrix, the location can be a more accurate representation of the matrix after correction precise positional relationship fisheye image, greatly reducing the computational difficulty, while improving the quality of stitching. ④ image fusion: the traditional linear alignment fusion splicing a lesser extent in the image, it will produce some degree of ghosting, resulting in the final panoramic image information redundancy, solve this problem, this paper presents a gray value for pixel difference is too large weight of special processing methods to eliminate ghost images, experiments show that a good effect. Upon completion of the analysis of the specific algorithm for each process improvements, using software engineering approach to the system modular, process-oriented, with a certain build a maintainability, complete implementation of image panorama stitching process.
|