Research On Positioning Method Based On Image Retrieval

In view of the wide application prospect and huge market value of location service in military,commercial and civil fields,and the difficulty of satellite navigation system to achieve correct positioning in the low-signal environment such as dense building block and indoor environment,this dissertation studies the relevant positioning methods.Considering the characteristic of the indoor and outdoor image in recognition of the advantages,positioning the cost and convenience,and according to the characteristics of the smart devices and people dependent on intelligent device,this dissertation puts forward a kind of location method based on image retrieval,this method is not only applicable to the outdoor positioning can also be applied to indoor positioning.According to the differences between indoor and outdoor scenes,the research contents are divided into outdoor landmark location and indoor scene location.The main research is as follows:In this dissertation,the advantages and disadvantages of SIFT and SURF algorithms in image retrieval are discussed.In view of the convenience of outdoor positioning,in the image matching stage,the database combined with the candidate images screened out by GPS signals to match with the query images to obtain relevant location information.Experimental results show that the improved method can effectively improve the accuracy of location and reduce the running time compared with the traditional method under different feature extraction algorithms.To cope with the problem of mismatching during matching,RANSAC algorithm was used for purification and the matching effect after purification was verified to be basically unaffected under the condition of image rotation and translation.According to the characteristics of the data set,the bag model in the image retrieval technology is used to construct the visual vocabulary in the indoor scene location method.According to the experimental data,the logistic regression classification algorithm used in this dissertation is more accurate than the traditional logistic regression algorithm and related classification algorithm.An improved method for indoor scene location is also proposed.By selecting the floor information in advance,the retrieval scope is reduced.Compared with ordinary methods,the running time is reduced to ensure the accuracy of image retrieval.In the final query,the method of voting decision is used to further improve the accuracy.Finally,in order to achieve more user-friendly human-computer interaction,this dissertation built a Django network framework,and combined with HTML5 and Python language to write a human-computer interaction interface.The user uploads the captured image through the human-computer interaction interface,and the server feedback the positioning result to the user.The operation process of the whole system is simple and the response time is fast.