Research of Panoramic Images Formation Methods
Students Name: Tsap Serhii Ihorovych
Qualification Level: magister
Speciality: Radioelectronic devices, systems and complexes
Institute: Institute of Telecommunications, Radioelectronics and Electronic Engineering
Mode of Study: full
Academic Year: 2022-2023 н.р.
Language of Defence: ukrainian
Abstract: A panoramic image is a separate wide-angle image of the environment around the camera. Panoramas are widely used in fields such as robotics, computer vision, video surveillance, and virtual reality. They are also used for commercial purposes such as entertainment, interactive television, real estate and virtual tourism. [1]. Over the past 15 years, panoramic shooting systems have progressed significantly. Not only specialists can create and display panoramas. Thanks to the abundance of software available [2], anyone with a computer and a camera can create panoramas. In order to create a panorama from several separate images, it is necessary to perform some processes on them [3]. First, with the help of shaping methods, find the key points on these images and highlight those that coincide with each other on both images. Then compare them with each other using homography and affine transformations and get a single image. But often this will not be enough, due to possible differences in color, brightness or in a slight change of perspective between the images that were stitched into one. In such cases, the "mosaic" effect will be noticeable [4]. To remove it already on the finished panorama, various methods of color and brightness correction are used, as well as mixing methods, such as alpha mixing, or multi-band mixing using Gaussian and Laplace pyramids. The main methods of forming panoramic images include: SIFT, SURF, ORB, BRISK and AKAZE [3,5]. These methods have their own algorithms for finding key points for further comparison between them. Basically, they are based on the search for special points, passing the image pixels through certain filters, thus looking for maximum changes in the brightness gradient, or contrast. RANSAC and PROSAC methods are used for geometric image correction [6]. The RANSAC algorithm is an iterative method of estimating model parameters based on random samples. PROSAC ¬ progressive sampling consensus algorithm.It uses a linear ordering defined on the set of matches by the similarity function used to establish prior matches. The purpose of the work – study the process of forming a panoramic image from fragments. The object of research – process of forming panoramic images from fragments. The subject of research – algorithm for creating panoramic images using software. This master’s thesis is devoted to the research of methods of forming panoramic images. The following methods of image formation: SIFT, SURF, ORB, BRISK, AKAZE are reviewed and their work algorithms are described. Software packages for working with panoramic images were considered and the OpenCV library was selected for further work. Based on it, a program was written in the Paython language for conducting experiments. Based on the results of these experiments, conclusions were made regarding the effectiveness of the methods and their comparison with each other. In terms of performance, BRISK turned out to be the best, it has the largest number of found points and the smallest relative time to find them. And in terms of speed - ORB. It surpasses its competitors in terms of stay time by more than two times. But at the same time, he finds the fewest points of all. Deterioration in the quality of stitched images was not noticed. Recommendations for improving the images obtained as a result of the experiment are also given. Key words: panorama, OpenCV, correction, matching methods, key points. References 7. Gledhill, D., Tian, G. Y., Taylor, D., & Clarke, D. (2003). Panoramic imaging—a review. Computers and Graphics, 27(3), 435- 445. https://doi.org/10.1016/S0097-8493(03)00038-48. Huang, Ho Chao and Yi-Ping Hung. “Panoramic Stereo Imaging System with Automatic Disparity Warping and Seaming.” Graph. Model. Image Process. 60 (1998): 196-208. 9. Szeliski R. Computer Vision: Algorithms and Applications. New York: Springer-Verlag, 2010. 812 p. 10.Steedly D., Pal C., Szeliski R. Efficiently Registering Video into Panoramic Mosaics // Tenth IEEE International Conference on Computer Vision, 2005. Vol. 2. P. 1300–1307. 11.Rublee E., Rabaud V., Konolige K., Bradski G. ORB: An efficient alternative to SIFT or SURF // Proceedings of the 2011 International Conference on Computer Vision, 2011. P. 2564-2571. 12.Куцаченко Н. Г. Сравнение алгоритмов трекинга движения при реализации средствами OpenCV, Системний аналіз та інформаційні технології: матеріали 19-ї Міжнародної науково-технічної конференції SAIT 2015, Київ 22-25 червня 2017 р.