Research On Indoor Positioning Technology Based On Deep Learning

In recent years,the commercial value of location-based services(ILBS)has increased,and the demand for precise locations has become higher and higher.Indoor environments are usually complex and changing.Multiple obstacles indoor results in signal fluctuations.Since GPS signals are severely damaged when penetrating obstacles and cannot complete indoor positioning,other indoor signals are applied for positions estimation.Among them,the hardware equipment of Wi Fi signal is basically available indoors,without adding additional cost,so Wi-Fi positioning has been widely studied.We use Received Signal Strength Indication(RSSI)to represent the strength of the Wi-Fi signal,and RSSI is also the main reference data for completing positioning.This paper studies indoor positioning algorithms based on deep learning,and proposes three effective algorithms for the existing problems of indoor accurate positioning.The main research results of this paper include the following aspects.Firstly,this paper proposes a fingerprint selection localization algorithm based on neural network.In order to solve the problem that the fingerprints of large positioning area are long and the quality is uneven,a lot of research has been done.The cost always increases the complexity of the algorithm,so the problem has not been fundamentally improved.This paper selects excellent fingerprint data according to the relationship between signal value and distance,and then uses artificial neural network to complete position estimation.When selecting fingerprints,we use a greedy algorithm to infer the partitions of targets,and then use the signal values collected in the partition as new fingerprint data.The actual position of the target is obtained by coordinate mapping from the output value of the neural network model.The experimental results show that the fingerprint selection and positioning algorithm based on neural network can provide more accurate positioning and occupy less memory.Second,this paper proposes a localization algorithm based on path fingerprints and convolutional neural networks.We propose this algorithm solving the problem of the poor quality of fingerprint and twins fingerprints that caused by less signal sources and large ambient noise.Twins fingerprints will lead to the failure of one-dimensional fingerprints,and accurate positioning becomes impossible.Many researches are devoted to solving this problem,but most of them need to increase hardware-assisted.Without increasing the hardware cost,this paper proposes an algorithm to construct path fingerprints and use convolutional neural networks to complete position estimation.When the path fingerprint is constructed,the fingerprints of the previous moments are combined into a twodimensional fingerprint,which basically eliminates the twins fingerprint.The data in a two-dimensional fingerprint is locally correlated,so it is suitable to use a convolutional neural network to learn the features of the fingerprint.The model filters data noise through multiple convolution,pooling,activation and other operations,and fully learns the relationship between path fingerprints and coordinates.The experimental results show that the positioning algorithm based on path fingerprint and convolutional neural network can provide more accurate positioning and more stable performance.Third,this paper proposes a residual network localization algorithm based on fingerprint shaping.In order to solve the problem that the fingerprints are far from the standard and insufficient amount,many localization methods have been proposed,but they have not fundamentally solved the problem.In this paper,we choose the attention mechanism to shape the fingerprint,and then use a model synthesized by multiple algorithms to estimate the position.During fingerprint shaping,we calculate the similarity score with the standard fingerprint,and then adjust the fingerprint data according to the score.In order to learn the characteristics of fingerprints better,this algorithm sets a residual layer,and adopts a dual-input setting.Both the shaped fingerprint and the original fingerprint are input into the model to participate in the position estimation.According to their features,different algorithms are selected to extract the features of the fingerprints.The experimental results show that the fingerprint shaped by the attention mechanism is beneficial to improve the positioning accuracy,and the residual network is of great significance for improving the positioning accuracy and positioning stability.