Research Objectives Stitching Algorithm For Panoramic Imaging

As the rapid development of information technology, virtual reality, machine vision, computer simulation, and intelligent monitoring system are widely employed in our daily life. Since panoramic image mosaic is the kernel technology of the above mentioned technologies, the needs of panoramic image mosaics becomes more and more impendency. Moreover, panoramic image mosaic has broad prospects both in national defense industry and national economics, such as geodesy, medical image, military reconnaissance, video session and so on, which indicates that it is important for theory research and engineering practice to research on panoramic image mosaic.The flow path of panoramic image mosaic has three steps, namely image acquisition and preprocessing, image registration and image fusion. Because the process of image acquisition and preprocessing is relatively easy, this paper focuses on image registration and image fusion. Thereinto, the process of image registration contains feature extraction, search strategy and transform estimation. And this paper research on the last two aspects. Therefore, the main contents researched in this paper are as follows.1) Make comparative research on several feature point extraction. This paper briefly summaries five common feature point extraction methods, and analyzes their advantages and disadvantages. In order to find out relative complete evaluation criteria, existing evaluation criteria are summarized. Then the proposed evaluation criteria are utilized to compare the performances of five feature point extraction methods. The experiment results indicate that the performance of SIFT operator surpasses others. Therefore, this paper selects SIFT operator as feature point extraction.2) Propose an image matching algorithm based on artificial bee colony algorithm. After extracting feature points by SIFT operator, it’s necessary to employ proper search strategy to realize image matching. Since it’s hard to match image by point-to-point method, this paper employs Hausdorff distance as similarity measure. At the same time, an image matching strategy based on artificial bee colony algorithm is proposed to solve the problems of the existing algorithm with slow speed and low precision. In addition, experiments are made to check the performance of the proposed algorithm.3) In order to select proper image fusion algorithm, this paper compares several image fusion algorithm.Through studying on the key technology of panoramic image mosaic, on the one hand, fast and efficient image registration algorithm is proposed; on the other hand, advantages and disadvantages of the common image fusion algorithms are analyzed, which could provide reference when selecting image fusion algorithms. And all of the above mentioned establish a firm foundation.