The Research On Image Registration Based On Omni-directional Vision

The Omni-directional vision sensor can capture grand-view images which cover horizontally 360 degree, vertically 240 degree at once.And it can easily transform the input images to human friendly images such as cylinder expand Images. The Omni-directional vision has greatly developed since it has been researched by experts. These advantages are suitable for many applications requiring that a large field of view, including surveillance, tele-conference and tele-presence in virtual reality and robotics.This paper mainly studies the registration technology of the Omni-directional image based on Super-resolution (SR) image fusion technology. The paper focuses on the image registration methods which are suitable for the Omni-directional image. Firstly, this paper discusses the significance and introduces the primary aspects of the image registration, analyzes the definition and the principle of the image registration and some ordinary image registration methods.This paper researches the imaging principle of Omni-directional vision. In order to revise the distortion of Omni-directional image, the cylinder unwrapping algorithm is researched based on the conversing imaging principle. Experiments show that the unwrapping algorithm has high fidelity.According to the characteristics of the Omni-directional image, the paper presents a semi-automatic registration method based on feature and a registration method based on Harris corner detection. The semi-automatic registration method can provide precise points of the same name to ensure the accuracy of image registration and reduce the time consumption. Harris corner detection has rotation invariance and adapt to the registration of the Omni-directional image. As the Harris corner detection calculates the value of each point and then chooses the most advantage of the neighboring field, the Omni-directional image registration has a more satisfactory result.In order to improve the image registration accuracy, this paper proposes a registration method based on wavelet transform. The method has greatly enhanced the registration accuracy through the iterative process layer by layer.some experiments show that the Omni-directional image registration based on wavelet transform has high accuracy and high reliablity.