Image Fusion Algorithm For Panoramic Video Stitching And Its GPU Implementation

Panoramic video stitching system use video streaming technology to stitch overlap region and make it a higher resolution video, finally output a wider horizon panoramic video. Panoramic video stitching can be used for Virtual reality, security monitoring, auxiliary car, remote desktop and so on. This paper focuses on camera calibration and color correction algorithm of image fusion with GPU implementation, and our research are constructed as follows.First we build a panoramic image stitching architecture based on geometric correction and optical flow transformation. Generally image registration is based on feature points matching, which performs good at simple stitching effect. But it will cause ghost or artifact in complex scenes such as with occlusions and great parallax and so on. This architecture obtain homography matrix by geometric correction and project- ing images to the same plane, then we use energy function to estimate the optical flow of overlap region, which we could get the high quality stitching results, and largely remove ghost on overlap region.Next we propose a SURF auto calibration method based on blending region to deal with the inaccuracy of camera parameters. First, we calculate the overlap region by rough intrinsic and extrinsic parameters of camera. Then we do SURF feature matching on the blending region which calculated by adding width of overlap region to reduce the effort for precise blending region detection. And finally we get accurate homography.Finally we provid an online color correction algorithm based on temporal information to solve the problem that the color changes suddenly and not fluently in the videos. Therefore we plus color information in the process of color balancing and color mapping among few near frames to make color more factually intra-frames and color change more smoothly between the frames.Our article stitches HD videos(1920?1080), with large amount of temporal and computational complexity. To make the video in real time, GPU optimization is implemented on image fusion algorithm, so as to improve the stitching process speed.