Research On Image Landmark Feature Constructing Of Indoor Location

Vision is an indispensable way for most organisms to obtain information.There is no doubt that People from visual information to understand the environment.With the developments of economy,science and technology,the level of social digital information is improving increasingly.Humans have made great breakthroughs in visual sensor technology,artificial intelligence algorithms,wireless multimedia technologies,machine vision,which make the image processing technology can be a complement to various positioning navigation technologies in industry,life,medicine,aerospace,military and other fields.Image processing technology greatly improves the efficiency and the precision of the positioning system.Using the technology of machine vision to locate and navigate,it has the advantages of high efficiency,low price,great flexibility and no direct contact.In daily complex indoor environment such as: building,airport,warehouse,underground parking,the technology can collect the visual image of surrounding environment meanwhile compare and map with existing map library to give the user accurate location feedback.In this paper,we describe the theory of spatial scales that mimics the observation of human eyes firstly.The relevant theoretical knowledge of image features includes the classification of image feature,the important measurement features of image features and the description of common image features.Through the analysis of the features,comparison experiment and the understanding of the application scenarios of the system,we selected the image features which are suit for this system.When we combine the features,we discuss the normalization of the internal of features and between features separately.And we change the SIFT algorithm which only be used in processing grayscale images can also use in processing color images.The original 128-dimensional grayscale feature description has been expanded to a 160-dimension(128Gray+32Color)feature descriptor.The error matching of images is reduced meanwhile the time efficiency of the original algorithm is not greatly affected.In this paper,we also compared the proposed algorithm with some common related algorithms in terms of rotation invariance,brightness change,scaling change,and anti-noise performance.The comprehensive test verified that the proposed algorithm has better comprehensive performance.At the same time,according to the applicable scene of this system,we performed a performance comparison of the relevant characteristics of common content-based retrieval system.And in order to overcome the disadvantages of long time and low efficiency in retrieving large scale of image data,we add two kinds of techniques,unsupervised learning clustering technology and supervised learning classification technology.By classifying or clustering the image features in the image feature database,the retrieval efficiency of the retrieval system for the image data is greatly improved.Finally,a positioning system based on image landmarks is proposed.The system is composed of the improved content-based retrieval and the improved SIFT-based precision matching modules.The retrieval module is mainly responsible for quick and coarse retrieval in the large-scale image database,narrowing the matching range of precision matching module.The matching module is responsible for the accurate matching and feedback of image location information.The positioning system is divided into two parts: client and server.The client's function is to transmit query image to the server and to receive map information with location annotation from the server side.The function of the server side is to receive the query image,to realize the relevant algorithms and functions of the two-level module and to transmit the positioning information to the client.