An indoor positioning method, device and system

Abstract

The present invention discloses an indoor positioning method, device and system. The method includes obtaining an offline coordinate fingerprint library corresponding to different positioning regions, where each offline coordinate fingerprint library includes reference point coordinates and RSSI values corresponding to each reference point; respectively training a DAE model and an RNN model based on the data in each offline coordinate fingerprint library; determining the positioning region where the point to be located is located, and screening out the trained DAE model and RNN model corresponding to this positioning region; obtaining the RSSI value of the point to be located, and sequentially sending it to the screened DAE model and RNN model to calculate the coordinates of the point to be located. The present invention first performs regional positioning, and then uses the combination of fingerprint positioning and neural network for accurate position positioning, which can effectively resist environmental interference.

Description

Technical Field

[0001] The present invention belongs to the technical field of indoor positioning, and particularly relates to an indoor positioning method, device and system. Background Art

[0002] With the improvement of the economic level, the penetration rate of mobile terminals and mobile Internet services in China has reached an unprecedented height, and more and more personalized needs have emerged, among which the demand for indoor positioning is becoming stronger and stronger. Satellite positioning technology can provide positioning services well in outdoor environments. However, when positioning indoors, since the satellite signal is weak after reaching the ground and does not have penetration ability, and the floor information cannot be well judged, indoor positioning can hardly use satellite positioning. In order to solve the indoor positioning problem well, many types of solutions have emerged at home and abroad. Currently, several mainstream indoor positioning technologies include WiFi fingerprint positioning, UWB positioning, LED positioning, etc.

[0003] With the development of technology, mobile phones integrate more and more functions and have better signal reception, so more data can be obtained to process indoor positioning problems. However, the current commonly used positioning methods all have some defects more or less. WiFi fingerprint positioning is restricted by the size of the indoor space, and too large a space will lead to a decrease in positioning accuracy; although UWB ultra-wideband technology is centimeter-level, it is expensive and not economically applicable; LED positioning technology is interfered by indoor light sources and is not suitable for large-area laying. Summary of the Invention

[0004] In view of the above problems, the present invention proposes an indoor positioning method, device and system, which combines fingerprint positioning with a neural network and can effectively resist environmental interference.

[0005] In order to achieve the above technical purpose and reach the above technical effect, the present invention is realized through the following technical solutions:

[0006] In a first aspect, the present invention provides an indoor positioning method, including:

[0007] Obtain offline coordinate fingerprint libraries corresponding to different positioning regions, where each offline coordinate fingerprint library includes reference point coordinates and RSSI values corresponding to each reference point;

[0008] Train a DAE model and an RNN model respectively based on the data in each offline coordinate fingerprint library;

[0009] Determine the positioning region where the point to be located is located, and screen out the trained DAE model and RNN model corresponding to this positioning region;

[0010] Obtain the RSSI value of the point to be located and send it to the selected DAE model and RNN model in sequence to calculate the coordinates of the point to be located.

[0011] Optionally, the method for obtaining the offline coordinate fingerprint library corresponding to different positioning areas includes:

[0012] Divide the indoor area into several positioning areas;

[0013] For each positioning area, perform the following steps to obtain the offline coordinate fingerprint library corresponding to different positioning areas:

[0014] Record the RSSI value of each reference point at continuous times;

[0015] Based on the reference point coordinates and the corresponding RSSI values, establish an offline coordinate fingerprint library.

[0016] Optionally, the method for determining the positioning area where the point to be located is located includes:

[0017] Based on the light source frequency of the light source signal received by the point to be located and the corresponding relationship between the positioning area and different light source frequencies, determine the positioning area where the point to be located is located.

[0018] Optionally, the training method of the RNN model includes:

[0019] Take the RSSI values of each AP connected to the same reference point in the offline coordinate fingerprint library at the same moment as a group of inputs to obtain several groups of RSSI values set in chronological order;

[0020] Input the first group of RSSI values into the first layer of the RNN model, and the first layer performs weighted calculation on the first group of RSSI values, and the calculated result is coordinate C 1 ;

[0021] Input the second group of RSSI values into the second layer of the RNN model, and the second layer performs weighted calculation on the second group of RSSI values, and calculates the corresponding coordinate C 2 , which is coordinate C 1 and C 2 are given weights, and the calculated coordinate C' 2 ;

[0022] ……

[0023] Input the Nth group of RSSI values into the Nth layer of the RNN model, and the Nth layer performs weighted calculation on the Nth group of RSSI values, and calculates the corresponding coordinate C n , which is coordinate C n-1 and C n are given weights, and the calculated coordinate C' n , which is the coordinate of the point to be located;

[0024] Modify the weights of the input RSSI values and the coordinates C through backpropagation n-1 and C n weights until an ideal RNN model is obtained.

[0025] Optionally, the training method of the DAE model includes:

[0026] Obtain RSSI values in the case of no one in the room and establish a fingerprint database Ψ 0 ;

[0027] Obtain RSSI values in the case of normal movement of people in the room and establish a fingerprint database Ψ 1 ;

[0028] Using the fingerprint database Ψ 1 as the input layer and the fingerprint database Ψ 0 as the output layer, continuously adjust the DAE model parameters until the best DAE model is obtained.

[0029] In a second aspect, the present invention provides an indoor positioning device, including:

[0030] An acquisition module, configured to acquire an offline coordinate fingerprint database corresponding to different positioning regions, where each offline coordinate fingerprint database includes reference point coordinates and corresponding RSSI values;

[0031] A training module, configured to train a DAE model and an RNN model respectively based on the data in each offline coordinate fingerprint database;

[0032] A screening module, configured to determine the positioning region where the point to be located is located, and screen out the trained DAE model and RNN model corresponding to the positioning region;

[0033] A positioning module, configured to acquire the RSSI value of the point to be located and send it to the screened DAE model and RNN model in sequence, and calculate the coordinates of the point to be located.

[0034] Optionally, the acquisition module includes:

[0035] A division sub-module, configured to divide the indoor area into several positioning regions;

[0036] An offline coordinate fingerprint database establishment sub-module, configured to perform the following steps for each positioning region to obtain an offline coordinate fingerprint database corresponding to different positioning regions:

[0037] Record the RSSI value of each reference point at continuous times;

[0038] Based on the reference point coordinates and the corresponding RSSI values, establish an offline coordinate fingerprint database.

[0039] Optionally, the method for determining the positioning area where the point to be positioned is located includes:

[0040] Based on the light source frequency of the light source signal received by the point to be positioned and the corresponding relationship between the positioning area and different light source frequencies, determine the positioning area where the point to be positioned is located.

[0041] Optionally, the training method of the RNN model includes:

[0042] Take the RSSI values of each AP connected to the same reference point in the offline coordinate fingerprint database at the same moment as a set of inputs, and obtain several sets of RSSI values arranged in chronological order;

[0043] Input the first set of RSSI values into the first layer of the RNN model, and the first layer performs weighted calculation on the first set of RSSI values. The calculated result is coordinate C 1 ;

[0044] Input the second set of RSSI values into the second layer of the RNN model, and the second layer performs weighted calculation on the second set of RSSI values to calculate the corresponding coordinate C 2 , which is coordinate C 1 and C 2 are assigned weights to calculate the coordinate C′ 2 ;

[0045] ……

[0046] Input the Nth set of RSSI values into the Nth layer of the RNN model, and the Nth layer performs weighted calculation on the Nth set of RSSI values to calculate the corresponding coordinate C n , which is coordinate C n-1 and C n are assigned weights to calculate the coordinate C′ n , which is the coordinate of the point to be positioned;

[0047] Modify the weights of the input RSSI values and the weights of coordinates C n-1 and C n through backpropagation until an ideal RNN model is obtained.

[0048] Optionally, the training method of the DAE model includes:

[0049] Obtain the RSSI values in the case of no one in the room and establish the fingerprint database Ψ 0 ;

[0050] Obtain the RSSI values in the case of normal movement of people in the room and establish the fingerprint database Ψ 1 ;

[0051] Using the fingerprint database Ψ 1 as the input layer, the fingerprint database Ψ 0As the output layer, it continuously adjusts the parameters of the DAE model until the optimal DAE model is obtained.

[0052] In a third aspect, the present invention provides an indoor positioning device, including: a plurality of light sources, a signal transmitter, a signal receiver, and a positioning unit;

[0053] Each light source is installed in a different positioning area. The light source frequencies of the light sources in the same positioning area are the same, and the light source frequencies of the light sources in different positioning areas are different;

[0054] A reference point and an AP point are provided in each positioning area. The signal receiver is arranged at the reference point and the point to be located, and the signal transmitter is arranged at the AP point;

[0055] The positioning unit stores trained DAE models and RNN models corresponding to different positioning areas; the trained DAE models and RNN models are obtained by training based on the data in the offline coordinate fingerprint database corresponding to different positioning areas;

[0056] The positioning unit is connected to each signal receiver, and based on the light source frequency of the light source signal received by the signal receiver at the point to be located, determines the area where the point to be located is located, and filters out the trained DAE model and RNN model corresponding to this positioning area;

[0057] The positioning unit is connected to each signal transmitter, and obtains the RSSI value of the point to be located from each signal transmitter in the positioning area where the point to be located is located, and sequentially sends it to the filtered DAE model and RNN model to calculate the coordinates of the point to be located.

[0058] In a fourth aspect, the present invention provides an indoor positioning system, including: a storage medium and a processor;

[0059] The storage medium is used to store instructions;

[0060] The processor is used to operate according to the instructions to execute the method according to any one of the methods in the first aspect.

[0061] Compared with the prior art, the beneficial effects of the present invention are:

[0062] The present invention first divides a large indoor area into smaller positioning areas, uses light sources to distinguish each positioning area, establishes an offline coordinate fingerprint database, and uses the data in the offline coordinate fingerprint database to train DAE models and RNN models. In the actual positioning process, first, the area where the point to be located is located is judged by the light source frequency at the point to be located, and then the RSSI value of the point to be located obtained by the AP in this positioning area is uploaded to the trained DAE model for noise reduction processing, and the processed data is further handed over to the RNN model to calculate the coordinates of the point to be located.

[0063] It can be seen that the method of the present invention combines traditional WiFi fingerprint positioning with a neural network, and can effectively resist environmental interference. The method of the present invention divides a large area into smaller positioning areas, which can reduce the problem of large fluctuations in AP signals at long distances and further improve the positioning accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] In order to make the content of the present invention easier to be clearly understood, the present invention will be further described in detail below with reference to specific embodiments and the accompanying drawings, where:

[0065] Figure 1 is a flowchart of an indoor positioning method according to an embodiment of the present invention;

[0066] Figure 2 is a floor plan of a hypothetical shopping mall;

[0067] Figure 3 is a structure diagram of a recurrent neural network model;

[0068] Figure 4 is a diagram of the training process of a denoising autoencoder. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0069] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the protection scope of the present invention.

[0070] The application principle of the present invention will be described in detail below with reference to the accompanying drawings.

[0071] WiFi fingerprint positioning actually establishes the relationship between the indoor space and the received signal strength, and estimates the coordinates of any point to be located based on the given received signal strength (Received Signal Strength Indication, RSSI) value. The neural network is essentially a bridge to establish this relationship. Through a large amount of accurate data information training, the bridge is continuously rebuilt until the most suitable bridge is built. The neural network is an improved form of the feedforward neural network, which needs to access the previous information in the current iteration, and the calculations in the model will take into account the historical information. In addition, considering the complexity of the indoor environment, such as the interference caused by personnel flow to the signal, the present invention selects to add a denoising autoencoder at the front end of the recurrent neural network to preprocess the real-time signal and ensure the accuracy of the signal strength value as much as possible. The present invention can achieve the solution of the positioning accuracy problem and the solution of the slow calculation speed of the traditional WiFi positioning under the condition of being economical and applicable.

[0072] Embodiment 1

[0073] The present invention provides an indoor positioning method, as Figure 1 shown, which specifically includes the following steps:

[0074] Step (1): Obtain offline coordinate fingerprint libraries corresponding to different positioning regions, where each offline coordinate fingerprint library includes reference point coordinates and corresponding RSSI values;

[0075] Step (2): Train a DAE model and an RNN model respectively based on the data in each offline coordinate fingerprint library;

[0076] Step (3): Determine the positioning region where the point to be located is located, and screen out the trained DAE model and RNN model corresponding to this positioning region;

[0077] Step (4): Obtain the RSSI value of the point to be located, and sequentially send it to the screened DAE model and RNN model to calculate the coordinates of the point to be located.

[0078] In a specific implementation manner of the embodiment of the present invention, the method for obtaining the offline coordinate fingerprint libraries corresponding to different positioning regions includes:

[0079] Divide the indoor area into several positioning regions, where each positioning region has U reference points and V APs (Access Points);

[0080] For each positioning region, perform the following steps to obtain the offline coordinate fingerprint libraries corresponding to different positioning regions:

[0081] Record the RSSI value M times at each reference point in continuous time;

[0082] Based on the reference point coordinates and the corresponding RSSI values, establish an offline coordinate fingerprint library.

[0083] In the embodiment of the present invention, each positioning region corresponds to a different light source frequency. Based on this, the method for determining the positioning region where the point to be located is located includes:

[0084] Receive the light source signal of the point to be located, and based on the light source frequency of the light source signal received by the point to be located and the corresponding relationship between the positioning region and different light source frequencies, determine the positioning region where the point to be located is located.

[0085] In a specific implementation manner of the embodiment of the present invention, the training method of the RNN (Recurrent Neural Network) model includes:

[0086] Use the RSSI values of the V APs connected to the same reference point in the offline coordinate fingerprint library at the same moment as a group of inputs to obtain several groups of RSSI values arranged in chronological order;

[0087] Input the first group of RSSI values into the first layer of the RNN model. The first layer performs weighted calculation on the first group of RSSI values, and the calculated result is coordinate C 1 ;

[0088] Input the second group of RSSI values into the second layer of the RNN model. The second layer performs weighted calculation on the second group of RSSI values, and calculates the corresponding coordinate C 2 , which is coordinate C 1 and C 2 are assigned weights, and the calculated coordinate C' 2 ;

[0089] ……

[0090] Input the Nth group of RSSI values into the Nth layer of the RNN model. The Nth layer performs weighted calculation on the Nth group of RSSI values, and calculates the corresponding coordinate C n , which is coordinate C n-1 and C n are assigned weights, and the calculated coordinate C' n , which is the coordinate of the point to be located;

[0091] The weights of the input RSSI values and the coordinates C n-1 and C n are corrected through backpropagation until an ideal RNN model is obtained.

[0092] In a specific implementation manner of the embodiment of the present invention, the training method of the DAE model includes:

[0093] Obtain the RSSI values in the case of no one in the room, and establish a fingerprint database Ψ 0 ;

[0094] Obtain the RSSI values in the case of normal movement of people in the room, and establish a fingerprint database Ψ 1 ;

[0095] Using the fingerprint database Ψ 1 as the input layer and the fingerprint database Ψ 0 as the output layer, continuously adjust the DAE model parameters until the best DAE model is obtained.

[0096] The following combines the indoor scene as shown in Figure 2 and the attached Figure 1 to elaborate in detail on the method in the embodiment of the present invention.

[0097] Step 1: Establish a partitioned offline coordinate fingerprint database; specifically including the following sub-steps:

[0098] Step 11: Manually divide the large indoor area into positioning areas of appropriate sizes; for example, asFigure 2 As shown, a large area of 100×100m 2 is divided into 4 positioning areas of 50×50m 2 .

[0099] Step 12: Install LED light sources that emit a unified frequency in each positioning area. In specific applications, the light sources can also be other forms, which can be specifically set according to the actual situation;

[0100] Install LED lights in all four positioning areas. Using the frequency division multiplexing method, the microprocessor generates PWM wave signals of different frequencies. After being isolated by an optocoupler, the LED lights in the same positioning area are controlled to generate light of the same frequency, while the change frequencies of the LED lights in different positioning areas are different. For example, in area a, the light source frequency is adjusted to 2000Hz, in area b it is 2500Hz, in area c it is 3000Hz, and in area d it is 3500Hz. In this way, a relatively large indoor area can be artificially divided into small positioning areas of appropriate size, which is conducive to the subsequent neural network's processing of data.

[0101] Step 13: Assume there are U reference points and V AP points in the area. Set a signal receiver for each reference point and a signal transmitter for each AP point. Record the RSSI value vectors M times at each reference point in continuous time, and organize them as follows:

[0102]

[0103] Among them, i takes integer values in [1, U], and r(M, V) represents the Mth RSSI value collected by the Vth AP at the ith reference point. Here, it is assumed that within M seconds, the RSSI value is collected once per second, so a total of M RSSI values are collected. Then, the column vector of the matrix represents the M RSSI values collected by a reference point at the same AP within M seconds; the row vector represents the V RSSI values collected by a reference point at the same moment at V APs.

[0104] According to the size of the large area divided in Step 11, assume that there are three APs distributed in a small area (positioning area), that is, V = 3. The signal acquisition time is set to 5 seconds, so the same reference point and the same AP sample the RSSI value 5 times, that is, M = 5. Assume that a reference point is set every two meters, then there are a total of 625 reference points, that is, U = 625. Then the R at this time i can be expressed as:

[0105]

[0106] Step 14: Establish an offline coordinate fingerprint database from the reference point coordinates and the corresponding RSSI values.

[0107] Step 2: Train the Recurrent Neural Network model (RNN model). The model of the recurrent neural network is shown in Figure 3 , and the method for training the recurrent neural network includes:

[0108] Step 21: Use the RSSI value vectors of 3 APs at the same reference point at the same moment as the input, that is, use the row vectors of the R i matrix as the input, corresponding to x in the model 1 ;

[0109] Step 22: Perform weighted calculation on the 3 values. The weight row vector is a x , which corresponds one by one to each element (RSSI value) in x 1 , and the result obtained is the coordinate C 1 , which is h in the model 1 ;

[0110] Step 23: Present h 1 to the second layer of the next RNN, and at the same time input the next group of RSSI value vectors x 2 , and calculate the corresponding coordinate C 2 through x 2 ;

[0111] Step 24: Assign the weight W to the coordinate C 1 , and then calculate with C 2 to obtain the coordinate C' 2 , that is, h 2 ; C' 2 = C 1 +W*C 2 ;

[0112] Step 25: Repeat the above steps until the 5th group of RSSI value vectors are processed, and obtain the coordinate C' 5 , and output C' 5 , which is the coordinate of the point to be located. Thus, the expression for the forward propagation of the RNN can be obtained as:

[0113] h t = φ(a x x t +Wh t-1 +b)

[0114] Since indoor positioning does not care about the positioning results in the middle, and only requires the output of the final estimated coordinates, so the "many-to-one" model is adopted in this model. Then the expression for the final output y is:

[0115] y = h 5 = φ(a x x 5 +Wh 4 )

[0116] where a x is the weight vector of the input x t , W is the weight for coordinate calculation. b is the bias of the neuron. φ is the activation function, and generally the Sigmoid function is selected, as shown below:

[0117] Sigmoid activation function:

[0118]

[0119] Step 26: Correct a x vector by backpropagation for the weighted sum of the x t vector and the weights of the two fitted coordinates. For the RNN, since there is a loss function at each layer of the RNN, the final loss function L is:

[0120]

[0121] As can be seen from the RNN model, during backpropagation, the gradient loss at a certain position t is jointly determined by the gradient loss corresponding to the output at the current position and the gradient loss at the sequence index position t+1. Therefore, for the gradient loss of W at a certain sequence position t, it needs to be calculated step by step through backpropagation. Define the gradient of the hidden state at the sequence index t position as:

[0122]

[0123] From this, δ t+1 can be recursively derived to δ t :

[0124]

[0125] Expression for calculating the gradient of the weight W:

[0126]

[0127] Weight a x Expression for calculating the gradient of:

[0128]

[0129] Expression for calculating the gradient of the neuron bias b:

[0130]

[0131] Step 27: Iterate round by round through gradient descent and repeatedly train until an ideal RNN model is obtained.

[0132] Step 3: Train the denoising autoencoder model, i.e., the DAE (Denoising AutoEncoder) model;

[0133] The training of the denoising autoencoder is only related to the amount of data involved in the training. Therefore, when establishing the fingerprint database, only one AP is needed to obtain a large amount of RSSI data at a certain reference point. The method for training the denoising autoencoder includes:

[0134] Step 31: In the case of no one in the room and no obvious interference to the signal, collect the RSSI values of U reference points to form a vector r j and establish the fingerprint database Ψ 0 ; r j is in the form of:

[0135] r j =[r 1 r 2 … r U , j = 1, 2, …, U

[0136] Step 32: In the case of normal movement of people in the room, collect the RSSI values of U reference points to form a vector r′ j , r′ j has a one-to-one correspondence with the elements of r j and establish the fingerprint database Ψ 1 , r′ j is in the form of:

[0137] r′ j =[r′ 1 r′ 2 … r′ U , j = 1, 2, …, U

[0138] Step 33: Use the fingerprint database Ψ 1 as the input layer and the fingerprint database Ψ 0 as the output layer to train the DAE model.

[0139] Multiple groups of RSSI data can be used to train the DAE model and perform backpropagation, and the model is adjusted until the best DAE model is obtained. The training process of the denoising autoencoder is shown in Figure 4 , and the denoising autoencoder is trained to reconstruct the clean data point x from the corrupted version . Among them, the sample is the data of the fingerprint database Ψ 1 , and the sample x is the data of the fingerprint database Ψ 0 . h is the trained DAE model, and L is the loss function during training. Minimizing it can obtain the most suitable DAE model h.

[0140] Step 4: Determine the indoor area range where the point to be located is located.

[0141] Step 41: When a pedestrian enters the indoor area, the mobile phone camera (i.e., the signal receiver) receives the LED light source, and determines the positioning area where the pedestrian is located according to the frequency of the LED light source.

[0142] Step 5: Collect the signals at the point to be located and estimate the coordinate position according to the trained model.

[0143] Step 51: The signal transmitter at the AP in the area obtains the RSSI information of the mobile phone carried by the pedestrian and forms a matrix R i ;

[0144] Step 52: Upload the matrix R i to the corresponding DAE model for noise reduction processing to obtain the processed matrix R' i ;

[0145] Step 53: Present the matrix R' i to the corresponding RNN model, and after calculation by the RNN, obtain the coordinates of the point to be located.

[0146] Embodiment 2

[0147] Based on the same inventive concept as Embodiment 1, an indoor positioning device provided by an embodiment of the present invention includes:

[0148] An acquisition module, configured to acquire offline coordinate fingerprint libraries corresponding to different positioning areas, where each offline coordinate fingerprint library includes reference point coordinates and corresponding RSSI values;

[0149] A training module, configured to train a DAE model and an RNN model respectively based on the data in each offline coordinate fingerprint library;

[0150] A screening module, configured to determine the positioning area where the point to be located is located, and screen out the trained DAE model and RNN model corresponding to the positioning area;

[0151] A positioning module, configured to acquire the RSSI value of the point to be located, and sequentially send it to the screened DAE model and RNN model to calculate the coordinates of the point to be located.

[0152] Optionally, the acquisition module includes:

[0153] A sub-module for dividing, configured to divide the indoor area into several positioning areas, where each positioning area has U reference points and V APs;

[0154] A sub-module for establishing an offline coordinate fingerprint library, configured to perform the following steps for each positioning area to obtain offline coordinate fingerprint libraries corresponding to different positioning areas:

[0155] Record the RSSI value M times at each reference point at continuous time;

[0156] Based on the reference point coordinates and the corresponding RSSI values, an offline coordinate fingerprint database is established.

[0157] Optionally, each positioning area corresponds to a different light source frequency, and the method for determining the positioning area where the point to be located is located includes:

[0158] Based on the light source frequency of the light source signal received by the point to be located, determine the positioning area where the point to be located is located.

[0159] Optionally, the training method of the RNN model includes:

[0160] Take the RSSI values of V APs connected to the same reference point in the offline coordinate fingerprint database at the same moment as a group of inputs, and obtain several groups of RSSI values set in chronological order;

[0161] Input the first group of RSSI values into the first layer of the RNN model, and the first layer performs weighted calculation on the first group of RSSI values, and the calculated result is coordinate C 1 ;

[0162] Input the second group of RSSI values into the second layer of the RNN model, and the second layer performs weighted calculation on the second group of RSSI values, and calculate the corresponding coordinate C 2 , which is coordinate C 1 and C 2 are assigned weights, and the calculated coordinate C' 2 ;

[0163] ……

[0164] Input the Nth group of RSSI values into the Nth layer of the RNN model, and the Nth layer performs weighted calculation on the Nth group of RSSI values, and calculate the corresponding coordinate C n , which is coordinate C n-1 and C n are assigned weights, and the calculated coordinate C' n , which is the coordinate of the point to be located;

[0165] Modify the weights of the input RSSI values and the coordinates C n-1 and C n through backpropagation until an ideal RNN model is obtained.

[0166] Optionally, the training method of the DAE model includes:

[0167] Obtain the RSSI values in the case of no one in the room and establish a fingerprint database Ψ 0 ;

[0168] Obtain the RSSI values in the case of normal movement of people in the room and establish a fingerprint database Ψ 1 ;

[0169] With the fingerprint database Ψ 1 as the input layer and the fingerprint database Ψ 0 as the output layer, continuously adjust the DAE model parameters until the optimal DAE model is obtained.

[0170] Embodiment 3

[0171] Based on the same inventive concept as in Embodiment 1, an indoor positioning device is provided in an embodiment of the present invention, including: a plurality of light sources, a signal transmitter, a signal receiver, and a positioning unit;

[0172] Each light source is installed in a different positioning area. The light source frequencies of the light sources located in the same positioning area are the same, and the light source frequencies of the light sources in different positioning areas are different;

[0173] A reference point and an AP point are provided in each positioning area. The signal receiver is arranged at the reference point and the point to be positioned, and the signal transmitter is arranged at the AP point;

[0174] The positioning unit stores trained DAE models and RNN models corresponding to different positioning areas; the trained DAE models and RNN models are obtained by training based on the data in the offline coordinate fingerprint database corresponding to different positioning areas;

[0175] The positioning unit is connected to each signal receiver, and based on the light source frequency of the light source signal received by the signal receiver at the point to be positioned, determines the area where the point to be positioned is located, and filters out the trained DAE model and RNN model corresponding to this positioning area;

[0176] The positioning unit is connected to each signal transmitter, and obtains the RSSI value of the point to be positioned from each signal transmitter in the positioning area where the point to be positioned is located, and sequentially sends the filtered DAE model and RNN model to calculate the coordinates of the point to be positioned.

[0177] Embodiment 4

[0178] Based on the same inventive concept as in Embodiment 1, an indoor positioning system is provided in an embodiment of the present invention, including: a storage medium and a processor;

[0179] The storage medium is used to store instructions;

[0180] The processor is used to operate according to the instructions to execute the method according to any one of the methods in Embodiment 1.

[0181] Those skilled in the art should understand that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0182] The present application is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowchart and / or block diagram, as well as the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0183] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including instruction means that implement the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0184] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0185] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: modifications or equivalent replacements can still be made to the specific embodiments of the present invention, and any modifications or equivalent replacements that do not depart from the spirit and scope of the present invention should be covered by the protection scope of the claims of the present invention.

Claims

1. An indoor positioning method, characterized in that, it includes: Obtain offline coordinate fingerprint libraries corresponding to different positioning areas, where each offline coordinate fingerprint library includes reference point coordinates and RSSI values corresponding to each reference point; Train a DAE model and an RNN model respectively based on the data in each offline coordinate fingerprint library; Determine the positioning area where the point to be located is located, and filter out the trained DAE model and RNN model corresponding to this positioning area; Obtain the RSSI value of the point to be located, and send it to the filtered DAE model and RNN model in sequence to calculate the coordinates of the point to be located; The training method of the RNN model includes: Take the RSSI values of each AP connected to the same reference point in the offline coordinate fingerprint library at the same moment as a group of inputs to obtain several groups of RSSI values set in chronological order; Input the first set of RSSI values into the first layer of the RNN model. The first layer performs weighted calculations on the first set of RSSI values, and the calculated result is coordinate C 1 ; Input the second set of RSSI values into the second layer of the RNN model. The second layer performs weighted calculations on the second set of RSSI values to calculate the corresponding coordinate C 2 , which is the coordinate C 1 and C 2 are weighted, and the coordinate C ′ 2 is calculated; ……； Input the Nth group of RSSI values into the Nth layer of the RNN model, and the Nth layer performs weighted calculation on the Nth group of RSSI values to calculate the corresponding coordinate C n , which is the coordinate C n-1 and C n are weighted, and the coordinate C ′ n is calculated, which is the coordinate of the point to be located; Modify the weights of the input RSSI values and the coordinates C through backpropagation n-1 and C n weights until an ideal RNN model is obtained; The training method of the DAE model includes: Obtain the RSSI value when there is no one in the room and establish the fingerprint database Ψ 0 ; Obtain the RSSI values under normal indoor personnel movement conditions and establish a fingerprint database Ψ 1 ; With the fingerprint database Ψ 1 as the input layer and the fingerprint database Ψ 0 as the output layer, continuously adjust the DAE model parameters until the optimal DAE model is obtained.

2. The indoor positioning method according to claim 1, characterized in that, The method for obtaining the offline coordinate fingerprint library corresponding to different positioning areas includes: Divide the indoor area to form several positioning areas; For each positioning area, perform the following steps to obtain the offline coordinate fingerprint library corresponding to different positioning areas: Record the RSSI value of each reference point continuously in time; Based on the reference point coordinates and the corresponding RSSI values, establish an offline coordinate fingerprint library.

3. The indoor positioning method according to claim 2, characterized in that: The method for judging the positioning area where the point to be located is located includes: Based on the light source frequency of the light source signal received by the point to be located and the corresponding relationship between the positioning area and different light source frequencies, judge the positioning area where the point to be located is located.

4. An indoor positioning device, characterized in that, it includes: An acquisition module for obtaining offline coordinate fingerprint libraries corresponding to different positioning areas, where each offline coordinate fingerprint library includes reference point coordinates and RSSI values corresponding to each reference point; A training module for training a DAE model and an RNN model respectively based on the data in each offline coordinate fingerprint library; A screening module for determining the positioning area where the point to be located is located and screening out the trained DAE model and RNN model corresponding to this positioning area; A positioning module for obtaining the RSSI value of the point to be located and sending it to the filtered DAE model and RNN model in sequence to calculate the coordinates of the point to be located; The training method of the RNN model includes: Take the RSSI values of each AP connected to the same reference point in the offline coordinate fingerprint library at the same moment as a group of inputs to obtain several groups of RSSI values set in chronological order; Input the first group of RSSI values into the first layer of the RNN model, and the first layer performs weighted calculation on the first group of RSSI values. The calculated result is the coordinate C 1 ; Input the second group of RSSI values into the second layer of the RNN model, and the second layer performs weighted calculation on the second group of RSSI values to calculate the corresponding coordinate C 2 , which is the coordinate C 1 and C 2 are given weights, and the coordinate C ′ 2 ; ……； Input the Nth group of RSSI values into the Nth layer of the RNN model. The Nth layer performs weighted calculation on the Nth group of RSSI values to calculate the corresponding coordinate C n , which is the coordinate C n-1 and C n are weighted, and the calculated coordinate C ′ n is the coordinate of the point to be located; Modify the weights of the input RSSI values and the coordinates C through backpropagation n-1 and C n weights until an ideal RNN model is obtained; The training method of the DAE model includes: Obtain the RSSI value when there is no one in the room and establish the fingerprint database Ψ 0 ; Obtain the RSSI values under normal indoor personnel movement conditions and establish a fingerprint database Ψ 1 ; With the fingerprint database Ψ 1 as the input layer, and the fingerprint database Ψ 0 as the output layer, continuously adjust the DAE model parameters until the optimal DAE model is obtained.

5. The indoor positioning device according to claim 4, characterized in that, The acquisition module includes: A sub-module for dividing the indoor area to form several positioning areas; An offline coordinate fingerprint library establishment sub-module for performing the following steps for each positioning area to obtain the offline coordinate fingerprint library corresponding to different positioning areas: Record the RSSI value of each reference point continuously in time; An offline coordinate fingerprint database is established based on the reference point coordinates and the corresponding RSSI values.

6. An indoor positioning device according to claim 4, wherein, the method for judging the positioning area where the point to be positioned is located includes: Based on the light source frequency of the light source signal received by the point to be positioned and the corresponding relationship between the positioning area and different light source frequencies, the positioning area where the point to be positioned is located is judged.

7. An indoor positioning device, wherein, comprises: a plurality of light sources, signal transmitters, signal receivers, and a positioning unit; Each light source is installed in a different positioning area, the light source frequencies of the light sources located in the same positioning area are the same, and the light source frequencies of the light sources in different positioning areas are different; A reference point and an AP point are provided in each positioning area, the signal receiver is arranged at the reference point and the point to be positioned, and the signal transmitter is arranged at the AP point; The positioning unit stores trained DAE models and RNN models corresponding to different positioning areas; the trained DAE models and RNN models are obtained by training based on the data in the offline coordinate fingerprint database corresponding to different positioning areas; The positioning unit is connected to each signal receiver, and based on the light source frequency of the light source signal received by the signal receiver at the point to be positioned, the area where the point to be positioned is located is judged, and the trained DAE model and RNN model corresponding to this positioning area are selected; The positioning unit is connected to each signal transmitter, and obtains the RSSI value of the point to be positioned from each signal transmitter in the positioning area where the point to be positioned is located, and sequentially sends it to the selected DAE model and RNN model to calculate the coordinates of the point to be positioned; The training method of the RNN model includes: Taking the RSSI values of each AP connected to the same reference point in the offline coordinate fingerprint database at the same moment as a group of inputs, and obtaining several groups of RSSI values set in chronological order; Input the first set of RSSI values into the first layer of the RNN model, and the first layer performs weighted calculations on the first set of RSSI values. The calculated result is coordinate C 1 ; Input the second group of RSSI values into the second layer of the RNN model, and the second layer performs weighted calculation on the second group of RSSI values to calculate the corresponding coordinate C 2 , which is the coordinate C 1 and C 2 are assigned weights, and the coordinate C ′ 2 is calculated; ……； Input the Nth group of RSSI values into the Nth layer of the RNN model. The Nth layer performs weighted calculations on the Nth group of RSSI values to calculate the corresponding coordinate C n , for coordinate C n-1 and C n are weighted, and the calculated coordinate C ′ n is the coordinate of the point to be located; Correct the weights of the input RSSI values and the coordinates C through backpropagation n-1 and C n weights until an ideal RNN model is obtained; The training method of the DAE model includes: Obtain the RSSI value when there is no one in the room and establish the fingerprint database Ψ 0 ; Obtain the RSSI values under normal indoor personnel movement conditions and establish a fingerprint database Ψ 1 ; With the fingerprint database Ψ 1 as the input layer and the fingerprint database Ψ 0 as the output layer, continuously adjust the DAE model parameters until the optimal DAE model is obtained.

8. An indoor positioning system, wherein, comprises: including a storage medium and a processor; The storage medium is used for storing instructions; The processor is used to operate according to the instructions to execute the method according to any one of claims 1 to 3.

Images