Research On Indoor Fingerprinting Localization Techniques Based On Artificial Intelligence

Along with the development of wireless communication techniques and the increas-ing demand for the indoor location based service (LBS), indoor wireless localization technology has received more extensive attention in recent years. The quality of ser-vice (QoS) of the indoor LBS depends on the high performance of the wireless local-ization techniques, including the high positioning accuracy, high real-time performance, low computational complexity and low development cost. Among other existing posi-tioning techniques, the received signal strength (RSS) of the wireless local area network (WLAN) based fingerprinting localization technique has become the first choice for the indoor LBS applications. The main reason is because of the meter-level localization ac-curacy, the low cost of development on the smart devices and the seamless coverage of the WLAN in indoor environments.The fingerprinting localization system has some challenges to be resolved. In the of-fline phase, the system needs to establish a fingerprint database (or called radio map), and also needs to update the database to guarantee the validity of the database for position-ing when the environment has changed. However, the construction and updating of the database is a time and labour-consuming work which limits its application, especially for the large positioning scenarios. In the online phase, The time-variation of the RSS, which is caused by the multi-path and shadowing effects, will result in large location estimation error. Hence, the designed fingerprint pattern matching algorithm should deal with such problem to achieve high localization accuracy. In addition, the location estimation algo-rithm should have low computational complexity to improve the real-time performance. In view of these problems, this dissertation adopts the artificial intelligence techniques and optimization techniques to improve the overall performance of the fingerprinting lo-calization system. The innovations of this dissertation are as follows:1) In the offline phase, an automatically fingerprint database reconstruction technique based on affinity propagation clustering is proposed. The technique first observes the clustering fading character of the RSS in different sub-regions of the indoor testing area based on the area partitioning algorithm used in the research of cluster based channel model, then proposes a regional propagation model (RPM) to pre- dict the RSS of the indoor wireless signal. To realize the database reconstruction automatically, the technique adopts the affinity propagation clustering algorithm to divide the fingerprints pre-collected at the sparsely distributed reference points (RPs) into a certain number of clusters, and then partitioning the whole indoor area into sub-areas according to the clustering results. Subsequently, the RPM models are established based on the observations, and are used to predict the fingerprints at other RPs in the testing area to reconstruct the whole radio map. The numeri-cal experimental results show that RSS prediction accuracy of the proposed RPM model is higher than other existing path-loss propagation models. Meanwhile, the proposed positioning system can achieve high localization accuracy when cutting workload of fingerprint calibration by more than50%in the offline phase.2) Access point (AP) selection method can reduce the feature dimension of the pro-cessed signal in the fingerprinting localization system. Hence, AP selection is not only useful to reduce the computational complexity to improve the real-time perfor-mance of positioning, but also helpful to reduce the memory cost. The dissertation analyzes the Cramer-Rao lower bound (CRLB) of the location estimation error, and then fuses the signal strength, signal discrimination and signal stability to propose a hybrid AP selection method. This proposed method can effectively maximize the location discrimination ability of the AP signal to improve the localization ac-curacy, and can also reduce the computational complexity at the sametime. The numerical experimental results show that the proposed hybrid AP selection method achieves better localization accuracy than other algorithms, and the average posi-tioning accuracy can reach up to1～2m.3) In the online phase, a particle swarm optimization (PSO) and Kalman filter (PSOKF) based positioning and tracking system is proposed. The dissertation de-scribes the positioning system using PSO model in detail, analyzes the effects of the particle swarm initialization on the location estimation error, and also analyzes the computational complexity of the proposed PSO based positioning system. In the design of dynamic location estimation algorithm, the initialization of the PSO method is modified to improve the convergence rate and the global convergence ability. Hence, the modified PSO algorithm can effectively achieve higher localiza-tion accuracy. In addition, the tracking system adopts the Kalman filter to update and smooth the estimation error to track the trail of the mobile terminal. The the-oretical analysis and the numerical experimental results demonstrate the validity of the proposed PSOKF positioning and tracking system. As a result, the largest estimation error of the PSOKF system is less than1.5m.4) All of the experiments in this dissertation are carried out by the developed posi-tioning platform in two real-world indoor environments. Furthermore, all the re-sults effectively verify the validity, practicality and utility of the proposed methods. Hence, the proposed algorithms are practical to promote the commercialization of the indoor LBSs.