Research On Key Technologies Of Indoor Localization Based On Machine Learning

With the fast development of mobile internet and Internet of Things(IoT),Location-based Services(LBS)have received considerable attention from all walks of life.The LBS are widely used in outdoor localization fields,such as medical rescue,logistics tracking,tourist recreation etc.In recent years,with the speed development of wireless communication techniques,microelectronics computing techniques and embedded computing techniques,smart mobile devices are widely used,thereby promoting the extension of LBS from outdoor to indoor.Many large indoor scenes have high application demands for LBS,such as airport,hospital,museum,shop center,etc.In complex and variant large indoor LBS application scenes,the Global Navigation Satellite System(GNSS)that is widely used in outdoor localization cannot be applied to indoor localization,due to the variant indoor environment,the complex pedestrian's movement,the low-cost equipment embedded in smart devices and the high demand for positioning accuracy.It becomes the essential issue of indoor localization to compute accurate location information of targets and provide efficient services.With the improvement of processing capacity of smart mobile devices,various sensors are embedded into smart mobile devices.It has great practical significance and broad application prospect for us to reach the fusion of multiple sensor signals and propose convenient and efficient indoor localization techniques.This paper analyses the indoor localization of smart mobile devices in depth and introduces the machine learning methods into the indoor localization of smart mobile devices.This paper also improves and replenishes the deficiency of existing researches.The research results have good theoretic significance and practical application value.The main contributions of this dissertation are concluded as follows:(1)Focusing on the accuracy problem of K-Nearest Neighbor(KNN)based instantaneous fingerprint matching indoor localization algorithm for WiFi network confronted with the fluctuation or jamming of signals,the WiFi fingerprint matching indoor localization algorithm based on time-series is proposed.The WiFi signal features in localization area are converted to the time-series fingerprints according to the sequence of sampling.The time-series RSS fingerprints are converted to the sliding window based features for increasing the fingerprint features.For the time-series fingerprint features,two fingerprint matching algorithms are proposed,namely,the dynamic time warping(DTW)based fingerprint matching algorithm and the longest common subsequence(LCS)based fingerprint matching algorithm.Therefore,the accuracy of WiFi fingerprint matching is improved.The results show that the proposed indoor localization algorithm is better than existing instantaneous fingerprints localization algorithms.(2)Focusing on the problem of accumulated localization errors in pedestrian deadreckoning(PDR)indoor localization caused by low-cost noisy sensors and complicated human movements,this paper presents a novel PDR indoor localization algorithm combined with online sequential extreme learning machine(OS-ELM).By analyzing the process of PDR localization,the proposed algorithm formulates the process of PDR localization as an approximation function.Then,a sliding-window based scheme is designed to preprocess the obtained inertial sensor data and thus to generate the feature dataset.At last,the OS-ELM algorithm is improved and introduced into PDR for addressing the localization problem of pedestrians.Due to the fact of universal approximation capability and extreme learning speed within OS-ELM,our algorithm can adapt to localization environment dynamically and reduce the localization errors to a low scale.In addition,by taking the movement habits of pedestrian into the process of extreme learning,our algorithm can predict the position of pedestrian regardless of holding postures.To evaluate the performance of the proposed algorithm,this paper implements OS-ELM-based PDR on a real android-based smart device and compares it with the state-of-the-art PDR localization approaches.Extensive experiment results demonstrate the effectiveness of the proposed algorithm in various different postures and the practicability in indoor localization.(3)Focusing on the problem of large localization errors and the complicated parameter adjustment in indoor localization of fusing WiFi and inertial sensors caused by low-cost noisy sensors,fluctuation of WiFi signals and complicated human movements,this paper presents a novel indoor localization fusion algorithm that fuses WiFi localization with PDR using the long short-term memory(LSTM)networks.By analyzing the process of indoor localization fusion,this paper first formulates the process of indoor localization fusion as a recursive functional approximation process.Then,a sliding-window based scheme is designed to preprocess the received signal strength(RSS)data and PDR data,thereby generating the time-series based feature dataset.At last,the LSTM networks based localization algorithm is proposed to address the problem of indoor localization fusion.By utilizing the powerful prediction ability of LSTM networks to time-series,the proposed algorithm can fuse multiple localization techniques to predict the position of pedestrians accurately and avoid complicated parameter adjustment.To evaluate the performance of the proposed algorithm,this paper implements LSTM based indoor localization algorithm and compares it with the state-of-the-art filter-based approaches.Extensive experiment results demonstrate the accuracy and effectiveness of the proposed algorithm in indoor localization.