Research On Indoor Localization Model And Trajectory Adjustment Based On WiFi Fingerprints

With the increasing development of various intelligent applications,obtaining accurate location information becomes more and more important.Due to the complexity of the indoor environment,how to achieve accurate indoor positioning has become a hot research topic in recent years,and researchers have proposed many different solutions.Among them,the Wi Fi-based indoor localization has become one of the most popular technologies.In this paper,we study the indoor localization techniques based on the fingerprints of Wi Fi RSS,and mainly research on two aspects:(1)indoor positioning model;and(2)Radio map construction.In terms of indoor positioning models,we propose two indoor positioning models:a)CNNEu,an indoor positioning model based on convolutional neural networks.Compared with other neural network-based localization models based on sample RSS or CSI,we propose to use the distance distribution information in Wi Fi signal space of the samples as the input of the convolutional neural network,and apply the network to extract the high level features,which is further used for the location estimation of test samples.In addition,we propose a new positioning model called CNNLoss,which uses the distance distribution information in physical space to modify and constrain the signal feature vector during the network training phase.b)RMap TAFA,a trusted positioning model based on discrete radio maps.Compared with the radio map constructed by Site Survey,we innovatively propose credibility measures of crowdsourcing fingerprints.In the crowdsourcing collection process,the collected samples usually have sampling and labeling errors,and constructing radio maps directly based on them will result in inaccurate radio maps.We propose to construct the radio map after credibility evaluation of crowdsourcing trajectory samples.Furthermore,we adjust the following trajectories based on the constructed radio map.In terms of radio map construction,we propose to adjust trajectories based on particle filters and to construct radio maps based on samples of trajectories.a)Particle filter-based trajectory adjustment algorithm.Different from the existing solutions which adjust the trajectory based on the Site Survey radio map,we propose to use the fingerprinting localization results based on the fingerprint map constructed by samples of trajectories to selectively adjust the trajectory without burdensome Site Survey.The adjusted trajectory is then used to update our radio map.b)Radio map construction based on samples of trajectories.Compared with the traditional fingerprint structure of the grid,this paper proposes a new grid structure S-F,in which we define the credibility coefficient and the reliability coefficient to measure confidence level of a sample and the importance of each element in a fingerprint,respectively.To verify the performance of the proposed indoor localization models and the constructed radio map,we collect data in different actual scenarios and perform experiments with the measured data.Experiments show that in the room-type scenario,CNNEu model and CNNLoss model proposed in this paper can achieve a localization error of about 1 meter with fewer training samples;While in the corridor-type scenarios,the proposed RMap TAFA scheme can not only construct the fingerprint database using the trajectory samples to achieve reasonable positioning accuracy,but also effectively adjust the trajectory.