Research On Super-Resolution Reconstruction Based On Omnidirectional Vision

Catadioptric Omnidirectional vision can capture an omnidirectional image at video rate and can easily transform the input image to human friendly images such as a common perspective image and a panoramic image. These advantages are suitable for many applications requiring that a large field of view is images, including surveillance, tele-conference and tele-presence in virtual reality and robotics. However, the resolution of the transformed image is lower than that of conventional camera image, because 360 degree field of view is captured by only one CCD. Moreover the output image would be de degraded due to the blur and noise caused by the imaging system. These disadvantages make it restrictive for some applications where remote and high definition surveillance is needed.Super-resolution (SR) image fusion is the technology of reconstructing a frame of image with high resolution from a group of warped, blurred and noised low-resolution (LR) images or video sequence about the same scene. The goals of this technology are to break through the resolution limitation of the imaging hardware facilities, to make up the loss of spatial resolution during the acquisition and transmission of images by the help of the data fusion of complementary information between multi-frame images. It can not only improve the visual effect of image, but also be convenient and beneficial for computer to analyze, process, recognize and understand the image. So far, the technology of Super-resolution image fusion has been extensively applied to remote sensing, military, public security, computer vision, medical imaging, image compression and so on. Therefore, in this paper, the measurement and means for enhancing the resolution and definition of the omnidirectional images is researched cased on the SR technique.Firstly, an improved unwarping processing method based on forward mapping and Shepard scattered data interpolation method is brought forward. As such, it avoids distortion induced when interpolation is applied in the omnidirectional image directly. The experimental results indicate that the improved method is an effective cure for the nonlinear distortion of omnidirectional images as well as effective promoter for the smoothness of gray surface and precision of interpolation.Secondly, the causation of the de decline of the resolution and the special movement of the circumferential resolution and radial resolution is analyzed. The measurement and scheme to enhance the circumferential resolution and radial resolution based on super-resolution fusion is put forward. Test result shows the feasibility of the scheme.At last, Super-resolution omnidirectional restoration based on POCS is researched. The reference frame is transformed to panoramic image which is considered as the original estimate. After registration, the pixels in the LR image need be mapped onto the estimate panoramic image plane. The range restriction and similarity restriction are used as the convex restrictions. Estimated value is projected on the convex iteratively, bringing the result approach to the perfect HR image. Test result shows the validity of the scheme.