Research On Indoor Localization Methods And Its Optimizations For End-edge Collaborative Computing

With the widespread coverage of wireless networks and the increasing popularity of smart mobile devices,Location-based Services(LBS)have been widely used in large exhibition halls,parking lots,smart factories,underground mines,emergency fire rescue and other scenes.For outdoor scenarios,positioning and navigation systems such as GPS can already provide people with relatively accurate positioning results;however,for indoor scenarios,due to the complex and changeable environment and the instability of wireless signals in the process of propagation,the wireless indoor localization system can provide barely satisfactory locating results due to the challenges.In addition,problems such as heterogeneous terminal data fusion,limited system scalability and reliability,insufficient system real-time response capabilities and so on,remains as challenges to practical indoor locating systems as well.Addressing the large-scale real-time indoor localization application scenarios with high reliability requirements,this dissertation proposes a scalable end-to-edge collaborative distributed indoor localization computing architecture,using heterogeneous multi-core FPGAs as edge computing gateways and wearable smart devices as positioning terminals,with a set of end-to-edge collaborative indoor localization algorithms that can integrate multiple indoor localization technologies designed and implemented on top of the proposed hardware.By taking the full advantage of hardware architecture of the FPGA hardware platform,the operation efficiency of the positioning system is optimized.In this end-to-edge collaborative computing mode,the location computing task is completed on the edge computing gateway,which can effectively shorten the data transmission delay and improve the response efficiency of positioning requests.The calculation results of the edge computing gateway will be fed back to the terminal node through the wireless network for further LBS,and the relevant global location information and positioning results will be also uploaded to the cloud data center for subsequent data analysis and other processing.The main research contents and innovations of this paper are summarized as follows:1.Aiming at the problem of heterogeneous terminal data fusion in multiple wireless networks,a highly reliable indoor localization mechanism based on multi-terminal coordination is proposed.Based on the virtual grid division of the area to be located,and according to the locatable ability of each wireless network in each grid,a feature grid fused with the global network is constructed.By matching weights for each wireless network to obtain the best global positioning combination scheme,efficient and high-precision location information is provided for end users.In order to deal with the problem of probabilistic failure of anchor nodes in practice,an anchor-confidence verification mechanism is designed to extract the reliability characteristics of the positioning capability of heterogeneous wireless networks,and the weights of each wireless network is dynamically adjusted based on the mechanism.In such a way,the reliability and accuracy of the results are improved.The experimental results show that the proposed reliable cooperative indoor localization mechanism based on multi-terminal has achieved good location accuracy,spatial stability and anti-fault ability.2.Addressing at the limited robustness and scalability of indoor localization algorithms in large-scale scenarios,a deep learning indoor localization algorithm based on FPGA acceleration is proposed.The algorithm adopts a scalable deep neural network with noise reduction function,which can fully extract its hidden feature representation from the noisy input fingerprint,thereby improving the accuracy and robustness of the positioning system.In order to improve the operation efficiency,an FPGA-optimized acceleration strategy for the proposed deep learning-based indoor localization algorithm is presented,aiming to provide a scalable and dynamically configurable hardware implementation scheme.The experimental results show that the deep learning indoor localization algorithm based on FPGA acceleration has good accuracy and system scalability,and the execution efficiency of the algorithm is 5.8 times higher than that of the version without hardware acceleration.3.In order to improve the real-time response capability of indoor localization system under the condition of high concurrent positioning request,an optimization of multiregional indoor localization method based on Compressed Sensing(CS)theory is proposed.Firstly,based on the CS theory,the fingerprint of Wireless Sensor Network(WSN)measured by the smart terminal is converted from the time domain to the frequency domain by constructing a discrete Fourier orthogonal change matrix.Then,the sparse feature data in the frequency domain is analyzed and processed.Finally,data reconstruction is performed on the frequency domain sparse vector to obtain target location information.Using the advantages of CS theory can effectively reduce the network communication bandwidth under the condition of high concurrent positioning requests,which is conducive to reducing network congestion and improving the real-time response efficiency of the positioning system.In addition,an anchor node’s reliability verification mechanism is introduced into the optimization method,which enables it to discover and reduce the damaging effect of potential malicious anchor nodes on the system positioning function.A complete indoor localization system deployed in an office scene is taken as the experimental setup and the results show that the location accuracy of the proposed CS-based method is about 20%-60% higher than that of the KNN fingerprint positioning algorithm,under the condition of 50% communication bandwidth reduction.4.Based on Xilinx’s Zynq MPSo C series FPGA programmable devices,an edge computing gateway with complete computing power,storage resources and network access bandwidth is implemented,with all the aforementioned indoor localization mechanisms deployed on it,along with other supporting hardware drivers and software libraries,which demonstrates a full hardware and software stack on the platform and servers as an full integration of all the work of this dissertation.In addition,combined with a low-cost,multi-protocol,and wearable smart terminal as the locating client,an indoor localization application system with simple deployment and good scalability is constructed.The actual operation results show that the average delay of the proposed system,which uses FPGA for parallel computing optimization and hardware acceleration,is significantly lower than that without hardware acceleration;and with the increase of the number of concurrent requests,the system still maintains a small response delay.This dissertation integrates the proposed indoor localization methods into a unified edge computing platform,and designs and develops an end-to-edge collaborative indoor localization system with high location accuracy,low system response delay,and strong load scalability.And it will be a great significance to the practical application of indoor localization technology in large-scale and dense scenes.