| Being an important branch of digital image processing, image mosaic becomes a hot topic in recent years. This technology can use more than one images, which has overlapping areas with each other, to form a large, high resolution picture. Image mosaic technology has broken the limit of image acquisition device which is inaccurate and difficult to use. It has been widely used in panorama mosaic, medical imaging, virtual reality and military fields.In the present study, we focused on the accuracy and the speed of image matching, which are the keys to image mosaic technology, have become the focus. Under the relevant work, this dissertation first introduces the significance and the application of image mosaic, especially the development of image matching’s development and the existing algorithms. Then, it summarizes the basic knowledge and theories of image mosaic, which includes coordinate systems of computer vision, the derivation and the estimation of the relational model. In the later part of this dissertation, the process of image mosaic and the key technologies used in the mosaic process is summarized.Base on the analysis of relevant work, we proposed a novel method, which combines the hardware with the software, to complete the image matching in twice. Base on the inertia sensor with magnetometer and deep sensor Xtion, the first matching can be applied to the rotation and the translation of camera in three axes. At the same time, it is insensitive to the influence of environment, especially the luminance. With the result of first matching, the second process of matching is carried out by the improved NCC and RANSAC algorithms, which solves the matching of Harris corner and relational model estimation respectively, in real-time under the high accuracy. Combined with the two matching process, the accuracy and the speed of the system are guaranteed. Base on the analysis of system error in the aspect of source and feature, we proposed a correctional method to improve the performance of the system by correctional matrix, which can change the system into a closed-loop form. The experiments were performed based on actual image data, and the results of the image mosaic are compared with the classical algorithm in accuracy, real-time performance and robustness, From the competition results, our method were proved to be competitive and efficient. |
