Study On The Realistic Scene Reproduction And Self-localization Based On Panoramic Images

With the development of computer and pattern recognition technology,the demand of the experience in the view of the realistic scene is higher and higher,hoping the computer vision system can provide more abundant information.In this thesis,the realistic scene reproduction is implemented based on the image correction and splicing technology.It also can search fast and match the query image by the research and development of image matching technology.Then the user can get a view of the most similar panoramic image and could understand the surrounding geographical environment information.To implement the system,the main research content involves the following several aspects:(1)A series of natural scene images are collected by using the fisheye lens within the scope of the campus.Then do the correction of the fisheye images.Establish the two different panoramic image libraries:the panoramic image library for displaying and the panoramic image library for the further study of the feature extraction and image matching based on the two correction algorithms:the correction of based on spherical matrix and the correction of based on camera calibration.(2)The Gist feature extraction algorithm is proposed based on scale space for the query image through introducing the scale space theory and establishing the Gaussian pyramid image.In view of the panoramic image,adopt the Gist of feature extraction based on Gabor filter groups with dividing into the sub image which is the same with the query image.(3)Implement the preliminary search based on vocabulary tree search.First of all,the panoramic image library are clustered by the K-means cluster to build vocabulary tree,and then the query image and the panoramic image respectively are quantified by a set of weight vectors.According to the image similarity measure,a few similar panoramic images are detected which is implementing the preliminary search.(4)Combined with scene Gist eigenvector the feature fusion algorithm is proposed based on SIFT feature points.Implement the accurate positioning on the basis of the traditional SIFT feature matching and combined with the ideas of the scenes Gist feature extraction.The correctness and effectiveness of the matching algorithm is verified by analyzing the results of the experiment.At the same time,this thesis improves speed of the system.(5)Development and design the system of the realistic scene reproduction and self-position based on panoramic images.It has a certain application value which runs stable and smoothly.When entering a query image,the user could search accurately a panoramic image by the positioning system,he can understand the environment geographic information in time.At the same time the multi-azimuth observation is implemented.In this thesis,the panoramic image library is established by using the fisheye lens and the basic operations of the correction and the splicing technology.The image positioning technology is studied on the basis of implement the realistic scene reproduction.Detect the image rapidly through the preliminary search and accurate positioning,which is convenient for the user to understand the surrounding environment timely.