Research On Visual Image Matching Method For Indoor Positioning Based On SIFT

In the vision-based indoor positioning method,image matching technology is the most important technical link.Image matching is to use the query image obtained online by the user to match the image in the offline database to obtain the location information of the online photo.The rate,accuracy and robustness of image matching directly affect the rate,accuracy and robustness of positioning.In addition,due to the interference and influence of various noises during the generation or transmission of the image,the quality of the image will be reduced,which will eventually lead to the problem of low image matching rate.This article preprocesses the images used,based on the SIFT algorithm,It focuses on the clustering,retrieval,and matching of image feature points,and applies them to visual positioning.An improved image matching algorithm based on SIFT is proposed.Due to the large number of database images,it is mainly used for the SIFT algorithm to extract the feature points of the image.The large number of feature points in each picture requires a large amount of calculation when searching for matching,which has a greater impact on the image matching speed.The BOVW model is used to process the pictures,combined with the TF-IDF algorithm,to assign a weight to each visual word,so that each picture in the database can be represented by a visual vector.In the positioning stage,the query picture is preprocessed the same as the first stage,and then the query picture is matched with the picture in the database,and the most similar pictures are retrieved,and then the voting mechanism KNN and the reclassification algorithm SVM are used to perform the picture exact match.Experimental results show that the proposed improved algorithm is more stable in robustness and faster in time performance.In terms of visual indoor positioning estimation,in different scenarios,it has good results in both positioning efficiency and positioning error.In the case of selecting an appropriate value of k and the size of the image subset,the positioning efficiency and positioning accuracy can be well balanced.