Research On Wlan Indoor Localization Based On Location Fingerprint

With the development of wireless communications and pervasive computing, various location-based services have sprung up, which can offer accurate localization information for object localization, emergency rescue, traffic management etc. Outdoor localization systems are limited in indoor environments owing to signal attenuation and complex radio propagation, so numerous indoor localization systems have been developed. Currently, because received signal strength (RSS) samples are easily collected from pervasively deployed access points by commonly used wireless local area network (WLAN) mobile terminals without additional hardware being required, location fingerprint-based indoor localization using WLAN is specially preferred and extensively researched in location sensing of pervasive computing. However, localization errors are caused by various reasons in complex indoor radio propagation environments as well as some key techniques in location fingerprint-based WLAN localization also need further research, so offering accurate and real-time localization results for indoor users is still a challenging task.Based on the research on location fingerprint-based WLAN localization and analyses on its development in China and abroad given in this dissertation, the problems existed in location fingerprint-based WLAN localization are summarized as follows. First, usually multiple on-line RSS samples are collected at one location, but most of fingerprinting algorithms use RSS mean samples for localization, which fails to make full use of all the available on-line RSS samples. Additionally, regarding mobile user localization, through employing information like spatial proximities of consecutive localization results, filtering algorithms can process localization results of fingerprinting algorithms for accuracy improvement. But existing linear filtering algorithms have limited accuracy and nonlinear filtering algorithms are computationally expensive. Finally, the conventional establishment process of radio-map is burdensome and time-consuming and the established one also needs to be updated according to variations of indoor radio propagation. Thus, based on the problems mentioned above, the main work and creative points of this dissertation are listed as follows.First, based on the analyses and comparisons of existing indoor localization systems, the reasons why location fingerprint-based WLAN localization has been widely researched and its characteristics are illustrated. Then the development and drawbacks of several key components of location fingerprint-based WLAN localization are researched. Finally, classic fingerprinting algorithms and filtering algorithms that are used for indoor localization are introduced and analyzed. Second, in the view of the fact the most of fingerprinting algorithms compute localization results with RSS mean samples, which fails to make full use of all the available on-line RSS samples, an RSS matrix correlation fingerprinting algorithm is proposed to make use of all the on-line RSS samples and incorporate RSS variations of reference points as weights into correlation coefficient computations for accurate neighbor selection. So the proposed algorithm outperforms the classic neighbor selection algorithms. Moreover, two fast matrix correlation algorithms based on off-line and on-line computations are proposed to compute correlation coefficients for different application conditions. They are able to effectively reduce the computational complexity and therefore more suitable for practical applications.Third, the existing Kalman filtering that is computationally efficient predicts user state with a linear model, which limits its performance. Thus, a map matching-based Kalman filtering is proposed. Using indoor map information, a map matching algorithm is developed. The building structure, human activity area and a link model that represents human walkways are recognized with a created map image matrix and then unreasonable localization results are corrected to the link model. With the integration of the map matching algorithm into Kalman filtering, localization results are corrected using indoor map information and spatial proximities of consecutive localization results, so the localization performance is greatly increased.Last but not least, because radio-map establishment is a burdensome and time-consuming process needed for collecting off-line RSS data at reference points and the radio-map also needs to be updated according to variations of radio propagation, an artificial neural network-based error correction algorithm is proposed to solve this problem. Some newly collected RSS samples with location information are used to compute localization results and errors with neighbor selection-based fingerprinting algorithms for the artificial neural network training. Then the RSS samples and localization results are fused by the artificial neural network as its inputs and nonlinear relationship between the inputs and its outputs localization errors is modeled. The network is also optimized by genetic and backpropagation algorithms. In the on-line phase, localization errors estimated by the artificial neural network are used to correct localization results. Then negative influence of static radio-map and reference point distribution can be effectively reduced and therefore localization accuracy is increased. Meanwhile, the RSS data for training the artificial neural network is easily collected with reduced effort, which can avoid updating the whole radio-map.