On Some Optimal Methods For Improving Omnidirectional Vision Image Quality

Omnidirectional vision system is composed of photosensitive components, imaging lens and quadratic rotation curve surface reflector, which can capture an omnidirectional image at video rate and an image in certain scope of vertical view. Omnidirectional vision system has the unified imaging, 360 degree view and rotation invariant etc. advantages. It has been applied widely to many engineering fields requiring a large field of view images due to its outstanding advantages, such as 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 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.This dissertation is supported by National Natural Sceinces Fundations"On omnidirectional vision resolution Improving Methods". The goals of this dissertation 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.Firstly, an improved unwrapping 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 single viewpoint constraint is a principal optical characteristic for most catadioptric omnidirectional vision. A new calibration method of single viewpoint constraint for the catadioptric omni-directional vision is proposed. Firstly, an image correction algorithm is obtained by training a neural network. Then, according to characteristics of the space circular perspective projection, the corrected image of the mirror boundary is used to estimate its position and attitude relative to the camera to guide the calibration. Since the estimates conducted based on actual imaging model rather than the simplified model, the estimate error is largely reduced, and the calibration accuracy is significantly improved. Experiments are conducted on simulated images and real images to show the accuracy and the effectiveness of the proposed methods.Thirdly, a radial space variant point spread function (RV-PSF) modeling and image restoration method for the COV are studied. Firstly, a radial edge detection based SV-PSF modeling method is proposed, rest on the analysis of the blur regions'distribution pattern in the COV image. Since the fuzzy image zone is clearly identified, the proposed method models global SV-PSF for the COV at limited regions where radial edges locate, avoiding the huge amount of calculation of edge detection and image segmentation. Experiment results indicate that our method is flexible and effective.Fourthly, the causation of the 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.Omnidirectional Visions systems have been a hot research object in the area of computer visiton due to its widespread application on engineering fields; it is a key point for more extensible cover of its engineering applications to improve the imaging quality of omnidirectional image. Allowing for the particularity of omnidirectional Visions systems, this dissertation aims at putting forward some new approaches to improve the imaging quality. It is of in-depth significance that the acquired results will promote the further development of the engineering applications of omnidirectional visions systems...