Research On Indoor Localization Techniques In Complex Scenarios Without Human Intervention

With the rapid development of wireless network technology and mobile internet,more and more services,which cover all aspects of business,industry and military,are designed based on the locations of users,making the study of indoor localization techniques a research hotspot nowadays.Indoor localization generally contains two phases:system deployment and application.In deployment phase,localization devices are deployed and tested,and preparations for the next phase are made.In application phase,the position of the target is calculated by localization algorithms according to the data received.As many algorithms depend on lots of labeled data to evaluate the system performance,or need a large quantity of specialized knowledge and data to model the property of wireless signal and environment,which increase the system cost and affect the system efficiency,relying too much on human intervention is a major challenge in deployment phase.In application phase,the performances of most existing localization algorithms confront with a severe decrease in real-world environment because Gaussian noise,NLOS(none-line-of-sight)and outliers occur concurrently.These types of noise are mainly caused by the disturbance of signal transmission in closed,complex and dynamic changing indoor scenarios.How to deal with data noise should be carefully considered in this phase.To address the above problems,i.e.,the demand for human intervention in de-ployment phase and the existence of complex noise in application phase,we propose two novel algorithms in two key domains of these phases:base station quality evalu-ation and TDOA(Time Difference of Arrival)localization,and apply them to a com-mercial localization system.The main contributions of this paper can be summarized as follows:· We propose a fast base station quality evaluation method based on topology learning and feature selection.It takes the process of base station quality evalua-tion as an unsupervised feature selection problem.First,unlabeled data is collected and topology learning is adopted on them to reduce the data size,preserving the intrinsic information at the same time.Then data embedding and LI-regularization learning are used to get the importance representations of all features.At last,the evaluation of base stations is achieved by ranking the features.Compared with traditional methods,the proposed method is free of human intervention and much faster.It should be noted that this method is an unsupervised feature selection al-gorithm as well,which performs well on general datasets.· We propose a robust method for TDOA localization based on multistage de-noising and hierarchical solving.It contains three well-designed steps:multi-stage data denoising,Taylor-WLS algorithm and restricted particle filter,which are designed for eliminating outliers and NLOS,estimating the position and lowering the error caused by heavy Gaussian noise respectively.The combination of these strategies guarantees a precise,stable and smooth result in complex indoor envi-ronment.What's more,no human intervention is need so that high universality is achieved.· On the basis of the above research,we apply these two methods to a commercial high-accuracy UWB indoor localization system.In order to promote the domes-tication of localization devices and techniques,the system is developed based on domestic UWB base stations and tags,and achieves satisfying performance.The success of this system shows the effectiveness and practicability of our methods.