Research On Visual Positioning Based On Urban Street View

As the development of smart cities progresses,people's demand for accurate location of space becomes more and more urgent.The current outdoor positioning method is mainly based on the Global Navigation Satellite System(GNSS),which mainly includes the United States GPS,the Russian GLONASS,and China's Beidou system.However,its positioning accuracy and application are limited by the attenuation of signals in transmission,especially in highdensity buildings and other built-up areas in urban areas.The current publicly available streets view can display the fine picture quality and full-element information of urban scenes,integrate computer vision and GIS technology,and provide a new method for spatial position location in cities,visual positioning.Aiming at the problem of traditional visual positioning relying on three-dimensional modeling and low real-time performance,this paper studies a visual positioning method based on urban street view and uses them to complete the mapping of digital images to geographic space so as to achieve accurate spatial location.The main research contents of this paper are as follows:In view of the open characteristics of the current online street view,the urban street view data source and main parameter requirements are determined,and a street view download scheme based on the Baidu map development interface is studied.The Street View data download work in the study area was completed,and the grayscale image preprocessing was performed on the original Street View data.Analyze the street view image characteristics and discuss deeply the selection criteria of street view image description features.SIFT features have good scale,illumination,rotation,and translation invariance characteristics.The paper uses SIFT feature as a feature description method to finish the feature extraction experiments on street view images.Selecting the two structures of classical K-d tree and hierarchical K-means tree,feature index is built on SIFT feature vector,and fast and efficient feature search is implemented based on nearest neighbor search.Comparing the acceleration effect and the actual precision achieved by the two methods on the standard test data set,the experiment proved that the hierarchical K-means tree algorithm can search faster and the accuracy reduce slightly.Because there are a large number of identical structures on the surface of urban buildings,the existing feature point misjudgment judgment method will eliminate a large number of correct matches.This paper proposes a feature matching optimization algorithm considering geographical location,and the optimization algorithm can retain more correct matching.Finally,according to the main steps of visual positioning,this paper based on MATLAB language to achieve the visual positioning of urban street view image prototype system.The experimental results show that using the hierarchical K-means tree structure to construct the feature index has a faster positioning speed,the improved feature matching algorithm has relatively higher accuracy,and the operation of the visual positioning system also achieves the desired effect.