| With the popularity of various smart wearable devices,the demand for locationbased services is increasing.Among them,the WiFi fingerprint-based indoor positioning system has become one of the most promising indoor positioning solutions due to high positioning accuracy and wide deployment of WiFi routers.The existing WiFi fingerprint positioning systems mainly use the on-site survey to build an offline fingerprint database and then realize indoor positioning.However,to meet the positioning accuracy requirements,an intensive offline survey is needed in the positioning scene,which requires high labor costs.How to build a fingerprint database with high positioning accuracy at a low cost is one of the key problems of the system.In addition,the dynamic environment makes the offline fingerprint database gradually unable to describe the real-time fingerprint distribution,resulting in a continuous decline in positioning performance.And it also requires high labor costs to update the offline fingerprint database by re-exploration the environment.Therefore,how to update the fingerprint database with low cost and high accuracy is also an urgent problem to be solved.Because of the above problems,this dissertation considers the construction and updating of a WiFi fingerprint indoor positioning system under low labor cost,and completes the following two tasks:(1)Research on the construction algorithm of fingerprint database based on unsupervised super-resolution.Aiming at the construction of an offline fingerprint database,the unsupervised super-resolution algorithm is used to realize fingerprint augmentation,and the sparse samples are augmented to construct a dense offline fingerprint database,which meets the positioning accuracy requirements.The proposed algorithm only needs sparse samples to complete the training of the super-resolution network,realizing the construction of a dense offline fingerprint database with a low cost.The experimental results in a real scene and two simulation scenes show that the fingerprint augmentation error of the proposed algorithm is lower than IDW(Inverse Distance Weighted)and GPR(Gaussian Process Regression)methods.In the above scenarios,the positioning performance of the constructed fingerprint database is improved by more than 10.75%compared with the sparse fingerprint database.(2)Research on fingerprint database update algorithm based on super-resolution network transfer.Aiming at the problem that the offline fingerprint database fails over time,considering that the super-resolution network at the historical time contains the mapping relationship from sparse fingerprints to dense fingerprints,the historical fingerprint database information is reused by transferring the mapping to the current time,to reduce the training cost and improve the performance of the super-resolution network at the current time,and the fingerprint database is updated with the sparse samples at the current time.In the process of transfer,a model-based transfer strategy is adopted,and the transfer of the super-resolution network model is realized by freezing and finetuning methods,to effectively use the known information.The experimental results in the measured and simulation scenarios show that the proposed transfer algorithm reduces the network training cost by 93.3%,and the positioning performance of the updated fingerprint database is improved by more than 10.57%compared with that before the update.Aiming at the problem of fingerprint database construction and updating under sparse samples,this dissertation proposes methods based on unsupervised super-resolution and super-resolution network transfer,respectively.The experimental results show that the proposed algorithm is feasible and effective in constructing and updating fingerprint database under sparse samples.The performance of the proposed two algorithms is improved compared with the existing methods,which provides a new idea for solving the construction and maintenance of the indoor WiFi fingerprint positioning system. |
