Panoramic Image Mosaic System Based On Multi Camera

Panoramic image mosaic is one important research topic in computer vision field, is mainly used to solve the problem of single narrow perspective imaging equipment. Two images with overlapping parts, in the case of not reducing the resolution of the image, are stitched into a wide range of panoramic view of the image by using the panoramic image mosaic technology. With the development of computer and digital image field, panoramic image stitching in the civil, military, medical and other fields has been widely used, for example: virtual reality, medicine, panoramic images, vehicle driving assistant system and so on. When multiple cameras filming scenes for the panoramic image stitching, it will involve lots of issues,such as splicing image background complex, the large of feature points and image registration accuracy is low, video splicing real time difference etc.Aiming at the problem of image mosaic, this paper mainly studies from the following four aspects to research on video mosaic technology:(1)Scene background extraction. Because moving objects often appear in different scenarios,it is difficult that multiple scenes to get the background image at the same time. This paper compares and studies the traditional background modeling method,such as inter frame difference method, Gaussian mixture background model, Codebook and other classical methods, and combined with statistical methods in Statistics,adopted a background extraction method based on statistics. This method can extract the scene pictures, and provide a good mosaic of image stitching material.(2)Image registration. In this paper, the SIFT algorithm is used to extract image feature points, and the rotation, scale, translation and illumination invariant features of the algorithm are analyzed and deduced in detail;Feature point matching based on K-D tree;Using RANSAC algorithm to improve the correct rate of feature points matching;The cumulative error adjustment of image mapping was carried out by using the algorithm of bundle adjustment.(3)Image fusion. For the problem of joint image mosaic, this paper analyzes the common image fusion method, and finally uses the method of gradually in and out.(4)According to the geometric invariant of the camera position,in this paper, the splicing images and transformation matrix are updated every other time to solve the real-time video splicing.The system is designed in two cameras at the same time to monitor the scene, the video stitching. In the scene of moving objects more interference, it can achieve good real-time stitching effect and the next step will be to carry out the relevant research under more cameras.