| As the IOT(Internet of Things)technology has found its way into people's everyday life,the successful launch of various commercial applications would not have been possible without the support of location-based services,and the demand of corresponding technology has extended from outdoor to indoor.Given that the signal is shielded by buildings,Global Navigation Satellite System(GNSS)represented by GPS plays a limited role in providing location-based services in the indoor environment,how to realize indoor positioning with low cost and high precision is becoming a research focus both at home and abroad,but with no authoritative and recognized standard scheme yet.As an indispensable tool in people's daily work and life,the smartphone with its own multi-sensor function and the continuous expansion of its users makes the technologies of Wi-Fi,magnetic and pedestrian dead reckoning(PDR)qualified for general applicability and worth popularizing in a large scale.Wi-Fi realizes the pedestrian positioning by wireless access points that are widely deployed in the public areas,while the magnetic positioning relies on the magnetic fields inside buildings,which both have the advantages of no additional infrastructure,wide coverage,high positioning accuracy and no accumulation of positioning errors over time,while the PDR based on inertial sensor has the advantages of high positioning accuracy and strong independence in short time.In this case,there is a great potential to construct low-cost and high-precision indoor positioning scheme with these three technologies.The study focuses on the research the positioning methods in terms of Wi-Fi,magnetic and PDR and multi-source fusion strategies based on smartphone from the perspective of the optimization and innovation of single-system positioning method as well as the formation of multi-source fusion positioning scheme,and carries out a number of experiments in several scenarios to fully test and verify the positioning scheme.The main contributions include:(1)In order to overcome the problems where a large number of Wi-Fi samples exist in indoor environment,and irregular fluctuation of Wi-Fi signal can be resulted by the multi-path influence,a robust and accurate Wi-Fi location recognition method based on a deep neural network(DNN)is proposed.A stacked denoising autoencoder(SDAE)is employed to extract the robust features from noisy signals without the need of complex cluster screening,and in the positioning stage,the coordinates of unknown points can be obtained through combining the weights of a posteriori probability and geometric relationship of fingerprint points.In addition,we use constrained Kalman filtering and hidden Markov models(HMM)to smooth and optimize positioning results and overcome the influence of gross error on positioning results,combined with characteristics of user movement in buildings,both dynamic and static.The experiment shows that the DNN is feasible for position recognition,and the method proposed in this dissertation is more accurate and stable than the commonly used Wi-Fi positioning methods in different scenes.(2)In regard to magnetic positioning,a continuous path acquisition method is adopted to reduce the labor consumption of data acquisition,which is also applicable to the Wi-Fi acquisition.In this dissertation,the five-dimensional magnetic features are constructed by rotational transformation of the magnetic observation in an effort to solve the problem of few magnetic signal features and big positioning error due to only magnetic modulus involved.In order to improve the efficiency of magnetic positioning in the positioning process,the lower bounding function is introduced to generate an improved DTW magnetic positioning method,which eliminates the non-optimal matching sequence in advance by calculating the lower bounding distance of the magnetic sequence before each positioning.The method effectively removes the shortcoming of magnetic positioning in its high computational cost while ensuring the positioning accuracy,with positioning time consumption under different magnetic sequence lengths reduced by 23%?72% compared with the common DTW,the positioning error of 0.47 m in the corridor and 3.58 m in the whole floor.(3)In terms of the application of inertial sensors of smartphones,a pedestrian motion pattern identification method based on extreme learning machine(ELM)is brought forward according to the different time-domain statistical characteristics of acceleration and gyroscope data in different pedestrian motion patterns,with an accuracy rate of motion pattern identification reaching 97.25%.Considering that low-pass filter will destroy the original waveform and can not identify the wave crest effectively in the process of PDR,a step frequency detection method with local threshold and simple parameters is accordingly proposed,which introduces the concept of local information,and detects the peak value of pedestrian step frequency by simply setting the wave peak threshold and the interval difference threshold.The experiment suggests that the step frequency detection error is less than 3 steps at different walking distances and different smartphone usage postures.In addition,the study attempts to adopt the PDR with pure inertial navigation system(INS),and proposes for the first time to exploit the virtual zero-velocity state in the acceleration data to correct the results of INS-PDR based on the analysis of the gait data.In this case,the result of INS calculation,which diverges to 10 km away in a few minutes,is effectively converged under the constraint of zero velocity update(ZUPT),and the trajectory of PDR is thus consistent with the real one.(4)In the multi-sources locating information fusion,a fusion positioning method based on factor graph model is developed.To mitigate the influence of gross error on the positioning result in the fusion process,an adaptive robust adjustment mechanism is introduced to achieve a real-time adjustment of the observation information weights in the fusion system,and the historical position information is exploited to correct the heading in Madgwick attitude algorithm.In the Wi-Fi/PDR fusion localization,the experiment is carried out by utilizing three typical smartphone-using attitudes:held-in-hand,dangling,and calling,with the average errors of the fusion localization less than 1.4 meters,as well as the positioning accuracy improved by 29.6%compared with the positioning results acquired without adaptive robust adjustment.(5)In the light of the gross error of Wi-Fi location resulted by the signal interference,and a certain length of magnetic sequence required for the magnetic location,the probability of magnetic mismatching in large-scale scenes;as well as the incompetence of PDR in determining the starting position and its error accumulation over time,two different positioning methods concerning Wi-Fi/magnetic/PDR fusion are proposed and comparatively analyzed,and the adaptive factor graph model and quality control mechanism are employed in both methods.Method 1 takes advantages of the results of Wi-Fi positioning to narrow the scope of magnetic matching,reducing the probability of magnetic mismatching,which is followed by the fusion of the results of Wi-Fi/magnetic positioning with PDR,while Method 2 gives priority to the localization of magnetic/PDR,aided by Wi-Fi localization in starting position,static state,and areas with no evident changes in magnetic variability.Compared with Method 1,Method 2 gives full play to the positioning performance of magnetism and PDR while avoiding the gross error interference of Wi-Fi,and realizes a positioning error of only 2.33 meters in complex large-scale scenes,28% higher of the positioning accuracy than Method 1.There are 105 figures,24 tables,and 205 references in the dissertation. |
PDF Full Text Request