CN113747385B - Indoor positioning method, device, equipment and computer readable storage medium - Google Patents
Indoor positioning method, device, equipment and computer readable storage medium Download PDFInfo
- Publication number
- CN113747385B CN113747385B CN202111000063.3A CN202111000063A CN113747385B CN 113747385 B CN113747385 B CN 113747385B CN 202111000063 A CN202111000063 A CN 202111000063A CN 113747385 B CN113747385 B CN 113747385B
- Authority
- CN
- China
- Prior art keywords
- data
- training
- determining
- prediction model
- prediction
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W4/00—Services specially adapted for wireless communication networks; Facilities therefor
- H04W4/30—Services specially adapted for particular environments, situations or purposes
- H04W4/33—Services specially adapted for particular environments, situations or purposes for indoor environments, e.g. buildings
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W4/00—Services specially adapted for wireless communication networks; Facilities therefor
- H04W4/02—Services making use of location information
- H04W4/023—Services making use of location information using mutual or relative location information between multiple location based services [LBS] targets or of distance thresholds
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses an indoor positioning method, an indoor positioning device, indoor positioning equipment and a computer readable storage medium, wherein the indoor positioning method comprises the following steps: collecting position fingerprint data in an indoor environment, and determining labeled data and non-labeled data according to position information in the position fingerprint data; inputting the label-free data into a preset generation countermeasure network for training to obtain a pre-training parameter; inputting the labeled data and the pre-training parameters into a preset prediction model for training to obtain prediction parameters, and determining the target position according to the prediction parameters. The invention improves the accuracy of indoor space positioning.
Description
Technical Field
The present invention relates to the field of positioning technologies, and in particular, to an indoor positioning method, apparatus, device, and computer-readable storage medium.
Background
At present, machine learning methods are mostly adopted in indoor positioning technologies based on wifi fingerprint databases, and the indoor positioning technologies can be generally divided into two modes of clustering and regression. The clustering mode comprises a K-nearest neighbor method, a support vector machine algorithm and the like, the discrete position of the equipment can be estimated by the mode, the operation is simple, but the positioning accuracy is low due to the discreteness. In a regression method such as a radial basis function (radial basis function) and a multi-layer Perceptron (MLP), signal intensity data is directly input to a trained shallow network to obtain a position coordinate, but the shallow network has poor fitting capability of a shallow nerve, and has poor capability of expressing characteristics of nonlinear Signal Strength indicator (RSSI) data in an indoor space, and high-precision positioning is difficult to achieve.
Disclosure of Invention
The invention mainly aims to provide an indoor positioning method, an indoor positioning device, indoor positioning equipment and a computer readable storage medium, and aims to solve the technical problem of how to improve the accuracy of indoor space positioning.
In order to achieve the above object, the present invention provides an indoor positioning method, comprising the steps of:
collecting position fingerprint data in an indoor environment, and determining labeled data and non-labeled data according to position information in the position fingerprint data;
inputting the label-free data into a preset generation countermeasure network for training to obtain a pre-training parameter;
inputting the labeled data and the pre-training parameters into a preset prediction model for training to obtain prediction parameters, and determining the target position according to the prediction parameters.
Optionally, the step of inputting the labeled data and the pre-training parameters into a preset prediction model for training to obtain prediction parameters includes:
taking the pre-training parameters as network initial values in a preset prediction model, and inputting the labeled data to an input layer in the prediction model;
transmitting the labeled data to a convolutional layer in the prediction model through an input layer in the prediction model to perform feature extraction, so as to obtain a plurality of initial features;
and transmitting each initial feature to a pooling layer in the prediction model through the convolutional layer for filtering to obtain important features, and determining prediction parameters according to the important features and the network initial values.
Optionally, the step of determining a prediction parameter according to the important feature and the initial network value includes:
determining an RNN hidden layer in the prediction model, and transferring the important features from the pooling layer to the RNN hidden layer;
if a plurality of important features exist, calculating each important feature according to a hidden layer calculation formula in the RNN hidden layer to obtain a time sequence arrangement relation between each important feature;
sequencing all the important features according to the time sequence arrangement relationship, and performing Gaussian mixture probability distribution calculation on the sequenced important features to obtain an initial predicted value;
and determining a target predicted value in the initial predicted values according to the network initial value, and taking the target predicted value as a prediction parameter.
Optionally, the step of inputting the label-free data into a preset generation countermeasure network for model training to obtain a pre-training parameter includes:
and determining a preset L2 loss function in the generated countermeasure network, and inputting the label-free data into the L2 loss function for training calculation to obtain a pre-training parameter.
Optionally, the step of collecting location fingerprint data in an indoor environment comprises:
determining all reference points in the indoor environment, acquiring signal data sent by all the reference points, and determining signal intensity in all the signal data and position information corresponding to all the signal data;
and using the signal strength and the position information as position fingerprint data.
Optionally, the step of determining all reference points in the indoor environment comprises:
and determining a data acquisition area in the indoor environment, and performing grid division on the data acquisition area to obtain reference points, wherein at least one signal source is arranged in each reference point.
Optionally, the step of determining tagged data and untagged data according to the location information in the location fingerprint data includes:
determining position information and signal intensity in the position fingerprint data, and performing normalization processing on the signal intensity to obtain processed signal intensity;
storing the processed signal intensity serving as non-tag data into a preset first fingerprint database;
and storing the processed signal intensity and the position information into a second fingerprint database as labeled data.
In addition, to achieve the above object, the present invention also provides an indoor positioning device, including:
the system comprises an acquisition module, a storage module and a processing module, wherein the acquisition module is used for acquiring position fingerprint data in an indoor environment and determining labeled data and non-labeled data according to position information in the position fingerprint data;
the training module is used for inputting the label-free data into a preset generation countermeasure network for training to obtain a pre-training parameter;
and the determining module is used for inputting the labeled data and the pre-training parameters into a preset prediction model for training to obtain prediction parameters, and determining the target position according to the prediction parameters.
In addition, to achieve the above object, the present invention further provides an indoor positioning device, which includes a memory, a processor, and an indoor positioning program stored in the memory and capable of running on the processor, wherein when the indoor positioning program is executed by the processor, the steps of the indoor positioning method are implemented.
In addition, to achieve the above object, the present invention further provides a computer readable storage medium, on which an indoor positioning program is stored, and the indoor positioning program, when executed by a processor, implements the steps of the indoor positioning method as described above.
According to the invention, the position fingerprint data in the indoor environment is collected, the labeled data and the unlabeled data in the position fingerprint data are determined, the unlabeled data are trained according to the generated countermeasure network to obtain the pre-training parameters, the labeled data and the pre-training parameters are input into the prediction model together for training to obtain the prediction parameters, and the target position is determined according to the prediction parameters, so that the phenomenon that the positioning accuracy is low due to the fact that the position is determined only according to the position coordinates in the prior art is avoided, the signal intensity and the position are comprehensively considered, and the accuracy of indoor space positioning is improved.
Drawings
Fig. 1 is a schematic diagram of a terminal \ device structure of a hardware operating environment according to an embodiment of the present invention;
fig. 2 is a schematic flow chart illustrating a first embodiment of an indoor positioning method according to the present invention;
FIG. 3 is a schematic view of an apparatus module of the indoor positioning apparatus according to the present invention;
FIG. 4 is a schematic flow chart of an indoor positioning method according to the present invention;
FIG. 5 is a schematic diagram of generation of a countermeasure network in the indoor positioning method of the present invention;
FIG. 6 is a diagram illustrating a prediction model in the indoor positioning method according to the present invention;
FIG. 7 is a schematic diagram showing a second embodiment of a prediction model of the indoor positioning method according to the present invention;
fig. 8 is a schematic diagram of data collected in the indoor positioning method of the present invention.
The objects, features and advantages of the present invention will be further explained with reference to the accompanying drawings.
Detailed Description
It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
As shown in fig. 1, fig. 1 is a schematic terminal structure diagram of a hardware operating environment according to an embodiment of the present invention.
The terminal of the embodiment of the invention is indoor positioning equipment.
As shown in fig. 1, the terminal may include: a processor 1001, such as a CPU, a network interface 1004, a user interface 1003, a memory 1005, a communication bus 1002. Wherein a communication bus 1002 is used to enable connective communication between these components. The user interface 1003 may include a Display screen (Display), an input unit such as a Keyboard (Keyboard), and the optional user interface 1003 may also include a standard wired interface, a wireless interface. The network interface 1004 may optionally include a standard wired interface, a wireless interface (e.g., a WI-FI interface). The memory 1005 may be a high-speed RAM memory or a non-volatile memory (e.g., a magnetic disk memory). The memory 1005 may alternatively be a storage device separate from the processor 1001.
Optionally, the terminal may further include a camera, a Radio Frequency (RF) circuit, a sensor, an audio circuit, a WiFi module, and the like. Such as light sensors, motion sensors, and other sensors, among others. Specifically, the light sensor may include an ambient light sensor that adjusts the brightness of the display screen according to the brightness of ambient light, and a proximity sensor that turns off the display screen and/or the backlight when the terminal device is moved to the ear. Of course, the terminal device may also be configured with other sensors such as a gyroscope, a barometer, a hygrometer, a thermometer, and an infrared sensor, which are not described herein again.
Those skilled in the art will appreciate that the terminal structure shown in fig. 1 is not intended to be limiting and may include more or fewer components than those shown, or some components may be combined, or a different arrangement of components.
As shown in fig. 1, a memory 1005, which is one type of computer storage medium, may include therein an operating system, a network communication module, a user interface module, and an indoor positioning program.
In the terminal shown in fig. 1, the network interface 1004 is mainly used for connecting to a backend server and performing data communication with the backend server; the user interface 1003 is mainly used for connecting a client (user side) and performing data communication with the client; and the processor 1001 may be configured to invoke the indoor positioning program stored in the memory 1005 and perform the following operations:
referring to fig. 2, the present invention provides an indoor positioning method, in a first embodiment of which, the indoor positioning method includes the steps of:
step S10, collecting position fingerprint data in an indoor environment, and determining labeled data and non-labeled data according to position information in the position fingerprint data;
because the current indoor positioning mode is a clustering mode and a regression mode, and the clustering mode comprises a K neighbor method, a support vector machine algorithm and the like, the method can estimate the discrete position of the equipment, and the operation is simple, but the positioning precision is low due to the discreteness. The regression mode such as radial basis function and MLP directly inputs signal intensity data into a trained shallow layer network to obtain position coordinates, but the shallow layer neural fitting capability is poor, the characteristic expression capability of nonlinear RSSI data in indoor space is not strong, and high-precision positioning is difficult to realize. In a deep learning model, the stacked type drying self-encoder has strong generalization capability, effectively overcomes the noise interference of data, enables the data characteristics obtained by automatic learning to have more robustness, but has lower training precision and still has insufficient information acquisition. Therefore, in the embodiment, in order to avoid the phenomenon that the RSSI of the WiFi is easily affected by the indoor environmental noise and has strong time-varying property due to the signal fading effect and the multipath effect, and the common deep learning network is not enough to overcome the difficulty, the position is positioned by using the mode of generating the model + the prediction model.
In this embodiment, data acquisition and processing are performed first to obtain training samples and test samples. Training samples are further classified into labeled (i.e., labeled data) and unlabeled (unlabeled data). The method comprises the steps of inputting label-free training samples into a generation model to perform network confrontation training to obtain training references, adding the training references into a prediction model as pre-training parameters of the prediction model, inputting labeled data into the prediction model to perform training to obtain prediction parameters, and determining the target position to be positioned according to the prediction parameters. For example, as shown in fig. 4, a fingerprint sample is collected, and a fingerprint database is constructed according to the collected fingerprint sample, wherein the fingerprint database includes a fingerprint database with no label and a fingerprint database with a label. And inputting the data in the fingerprint database without the label into a generation model for training to obtain a pre-training parameter. And inputting the data in the fingerprint database with the label into a prediction model for model training, and positioning the target position according to the training result. Before training, the prediction model needs to obtain pre-training parameters, and the pre-training parameters are used as initial network values.
In this embodiment, signal detection may be performed through application software that can detect WiFi signal strength on a mobile phone or other smart terminals, so as to collect signal data of each area in an indoor environment, record a position coordinate where the signal data is located and signal strength received by the signal data, and use the position coordinate as position fingerprint data. And after acquiring the position fingerprint data, performing fingerprint data processing, such as removing processing, on each position fingerprint data, normalizing the signal intensity, and then storing the fingerprint data into a database, wherein the database comprises a supervised database (namely, a database with tags) and an unsupervised database (namely, a database without tags). The supervised database is a rectangle, the first two columns are X coordinates and Y coordinates, and the rest are signal intensity. Randomly extracting 1/4 of data at each training point as a test set, and the rest being training sets, aiming at enabling the training sets and the test sets to cover all the regions so as to train out the characteristics of all the regions. The unsupervised database has only signal intensity and no XY coordinates. XY coordinates are the labels of the data samples herein. Because the acquisition of the labels is troublesome and laborious, a database without labels is additionally collected to prepare a subsequent generation network.
And then acquiring labeled data in the labeled database, and acquiring unlabeled data in the unlabeled database, wherein the signal intensity in the labeled data can be the same as the signal intensity in the unlabeled data.
S20, inputting the label-free data into a preset generation countermeasure network for training to obtain a pre-training parameter;
in this embodiment, after determining the non-tag data, the generated countermeasure network GAN may be used to pre-train the non-tag data, and the effective features may be summarized, extracted, and located from the non-tag data. As shown in fig. 5, m = N × b (N reference points and b training points), the input X is a one-dimensional array of m RSSI [ RSSI1, RSSI2, RSSI3.. RSSIm ], each RSSI having its coordinates [ X, y ], and each RSSI developing a wifi signal value set collected at this coordinate, e.g., RSSI1= [ -30, -37., -56], with N = a × k elements inside. And X is a two-dimensional array [ m, n ], one group being the number of positions and one group being the eigenvalues. GAN does not use coordinate values. The generation countermeasure network comprises a discriminator and a generator, the generator is used for learning probability distribution and characteristics of real data x, the discriminator is used for judging whether a data source is the real data or the generated data, and parameter optimization is achieved through mutual countermeasure of the data source and the generated data.
The real data is { x 1 ,x 2 ,x 3 ,...,x m H, distribution of P data (x) The distribution generated by the generator is P G (x; θ) in order to find the parameter θ such that P G Is closer to P data . Using a random variable z, another distribution P is generated by G (z) = x G' Is allowed to react with P data The comparison corrects θ. The GAN formula is:
by fixing G, maxV (G, D) means P G And P data The target function of GAN is:
in addition, in this embodiment, to remove noise interference, a pooling layer is introduced between the discriminator and the generator, and part of noise is filtered by the pooling layer. In addition, the generative countermeasure network in the present embodiment uses an L2 least squares error loss function. Thus, when unlabeled data is input to generate the countermeasure network for training, K results are generated, and the L2 distance minimum is calculated for each result. Therefore, all predicted positions can be covered, and positioning is more accurate. In addition, in this embodiment, the parameters trained by the generated confrontation network are used as initial network values of the prediction network, that is, pre-training parameters.
And S30, inputting the labeled data and the pre-training parameters into a preset prediction model for training to obtain prediction parameters, and determining the target position according to the prediction parameters.
In this embodiment, a prediction model is also built, and as shown in fig. 6, the prediction model includes three parts, i.e., an input layer, a convolutional layer, a pooling layer, a flatten layer, an RNN hidden layer, a parameter, and an output Y. Wherein the first part comprises a convolutional layer, a pooling layer and a flatten layer. The second part includes an RNN (Recurrent Neural Network) hidden layer. The third portion includes hidden layers and parameters. And inputting the pre-training parameters into the prediction model in advance as the initial network values of the prediction model. And then, receiving label data in an input layer of the prediction model, sequentially performing model training of the first part, the second part and the third part to obtain a training result, namely a prediction parameter, and taking the position contained in the prediction parameter as a target position to be positioned.
In addition, in order to prevent the prediction model from easily falling into local optimum, gradient disappearance, overfitting and the like in the training process. Therefore, in this embodiment, the learning rate auto-adjustment mode is set using the pre-training parameters of the generative model. For gradient vanishing problems, a ReLU (Rectified Linear Unit) activation function is used here, and if the situation is severe, a residual structure can be used moderately. Batch normalization layer (BN) and Dropout layers are added to the prediction model herein for the overfitting problem. The Dropout layer randomly assigns zero weights to the neurons in the network. A ratio of 0.5 is chosen here, i.e. 50% of the neurons will be zero weighted. By doing so, the network is less sensitive to small changes in data. Therefore, it can further improve the accuracy of processing invisible data. And the BN layer is used for carrying out normalization processing among layers. With the deepening of the hierarchy, the distribution of the deep neural network may gradually shift, so that the BN is added between the convolutional layer and the active layer, the value range of information transmission between the hierarchy levels can be kept consistent, the attenuation trend of the hierarchy levels is avoided, the gradient disappearance can be prevented, and the training speed can be accelerated.
In addition, the fingerprint indoor positioning method in the embodiment can be applied to each mobile device, and can be applied to a hospital, a market and other scenes. The method has the advantages that pre-training parameters are obtained by training the GAN in an unsupervised mode, original cross entropy loss functions of the GAN are abandoned, and the L2 loss functions are used for covering all results, so that the positioning accuracy is improved. And a pooling layer is added between the generator and the arbiter to filter out environmental noise. In conclusion, the generative model can reduce the overfitting influence of the prediction model and simultaneously accelerate the training speed. And training the predictive model using supervised approaches. The characteristic extraction characteristic of the convolutional nerves, the middle state recording characteristic of the recurrent nerves and the mixed Gaussian probability distribution of the MDK are mixed to generate a novel prediction model framework CRMNN, the characteristics can be learned more comprehensively, the robustness and the accuracy of an algorithm are improved, and the accuracy of WIFI fingerprint positioning is improved.
In the embodiment, the position fingerprint data in the indoor environment is collected, the labeled data and the unlabeled data in the position fingerprint data are determined, the unlabeled data are trained according to the generated countermeasure network to obtain the pre-training parameters, the labeled data and the pre-training parameters are input into the prediction model together for training to obtain the prediction parameters, and the target position is determined according to the prediction parameters, so that the phenomenon that the positioning accuracy is low due to the fact that the position is determined only according to the position coordinates in the prior art is avoided, the signal intensity and the position are comprehensively considered, and the accuracy of indoor space positioning is improved.
Further, based on the first embodiment of the present invention, a second embodiment of the indoor positioning method of the present invention is provided, in this embodiment, step S30 of the above embodiment inputs the labeled data and the pre-training parameters into a preset prediction model for training, and refinements of the step of obtaining the prediction parameters include:
step a, taking the pre-training parameters as network initial values in a preset prediction model, and inputting the labeled data to an input layer in the prediction model;
in this embodiment, when performing position prediction by using a prediction model, it is necessary to first obtain a pre-training parameter obtained through training a generated confrontation network, and use the pre-training parameter as a network initial value in the prediction model, that is, an initial parameter in the prediction model. And after the parameters in the prediction model are set, the label data is received, namely the input layer in the prediction model receives the label data.
B, transmitting the labeled data to a convolutional layer in the prediction model through the input layer for feature extraction to obtain a plurality of initial features;
and c, transmitting each initial characteristic to a pooling layer in the prediction model through the convolution layer for filtering to obtain important characteristics, and determining a prediction parameter according to the important characteristics and the network initial value.
And (4) transmitting the labeled data to a convolution layer in the prediction model through the input layer for feature extraction to obtain each feature, namely the initial feature. And filtering the initial characteristics through the pooling layer to obtain important characteristics. In the first part of the prediction model, label data X [ m, n ] is input, training labels Y [ m, X, Y ] are input, and the input layer mainly uses a convolution layer and a pooling layer in the early stage of the prediction model. After the sample is input, the sample first passes through a plurality of layers of one-dimensional convolutional layers (1D Conv) in order to extract features. First, two 1D Conv with M filters are passed, so that M different characteristics X M, M can be obtained in the training of this layer. Then the maximum pooling layer filtering feature is used, important parameters are preserved, interference parameters are removed, and the complexity of the output can be reduced, at this time X M, M/2. Then two 1D Conv with 2M filters are passed so that higher level features can be learned, at this time X M, 2M. Then averaged over pooling layers to further reduce the effect of overfitting, at which time X M, M. In summary, the first part has the effect of extracting useful features from simple patterns, generating more complex patterns in higher-level layers, preserving important parameters, reducing non-important parameters, and reducing the amount of computation to reduce the overfitting effect. After the important features are determined, the important features can be directly screened according to the initial value of the network in the prediction model to obtain prediction parameters.
In this embodiment, the pre-training parameters are used as initial network values of the prediction model, feature extraction is performed on the labeled data through the convolution layer of the prediction model to obtain initial features, the initial features are filtered through the pooling layer to obtain important features, and then the prediction parameters are determined according to the important features and the initial network values, so that the accuracy of the obtained prediction parameters is guaranteed.
Specifically, the step of determining the prediction parameters according to the important features and the initial network value includes:
d, determining an RNN hidden layer in the prediction model, and transmitting the important features from the pooling layer to the RNN hidden layer;
in this embodiment, the prediction model further includes a second part, namely, an RNN hidden layer, and after the first part of the prediction model extracts the important features, the hidden layer can transfer the important features to the RNN hidden layer in the second part.
Step e, if a plurality of important features exist, calculating each important feature according to a hidden layer calculation formula in the RNN hidden layer to obtain a time sequence arrangement relation between each important feature;
step f, sequencing all the important features according to the time sequence arrangement relationship, and performing Gaussian mixture probability distribution calculation on the sequenced important features to obtain an initial predicted value;
and g, determining a target predicted value in the initial predicted values according to the network initial values, and taking the target predicted value as a prediction parameter.
When a plurality of important features are found in the RNN hidden layer, the first part of the prediction model can only extract the features to obtain each important feature, but the time sequence relation of each important feature is not determined, so that the hidden layer calculation formula in the RNN hidden layer is determined first, and then calculation is carried out to obtain the time sequence arrangement relation among the important features. The operation principle of the RNN hidden layer is shown in fig. 7, and includes an output layer, a hidden layer, an input layer, and a loop layer. Where x is the input, y is the output, h is the hidden layer unit, σ h And
for the activation function, U is the weight matrix from the input layer to the hidden layerW is the weight matrix from the previous hidden layer to the next hidden layer, V is the weight matrix from the hidden layer to the output layer, b h To hide layer deviations, b y Is the output offset. Calculating the hidden layer and the output of the RNN at t time according to a hidden layer calculation formula as follows:
h t ＝σ h (Ux t +Wh t-1 +b h )
furthermore, RNN is a loop in time, and each loop uses the result of the previous calculation, so the last output of RNN contains information at all previous times. The second part uses RNN to learn the lower dimension features from the first part, and supplements the intermediate state information. The input and output of the RNN hidden layer are both X (m, n).
When the time sequence arrangement relation of each important feature is determined and the important features are reordered, gaussian mixture probability distribution calculation can be carried out, namely the output layer is the combination of M mu, sigma and w. The loss function of the network is
When the neural network is used for prediction, possible values of all y can be obtained, the training precision is greatly improved, and the method is suitable for prediction tasks. At this time Y [ m, 1]]The position, i.e., the initial predicted value, is indicated in 1. And determining a target predicted value in each initial predicted value according to the network initial value, and taking the target predicted value as a prediction parameter. Namely, the position fingerprint data corresponding to each initial predicted value can be determined and used as the target position fingerprint data, and then the final initial predicted value, namely the target predicted value, is determined according to the size of the network initial value corresponding to the target position fingerprint data.
In the embodiment, important features are transmitted from the pooling layer to the RNN hidden layer to determine a time sequence arrangement relation, the sequences are reordered according to the time sequence arrangement relation, gaussian mixture probability distribution calculation is performed to obtain an initial predicted value, and a prediction parameter is determined according to the network initial value and the initial predicted value, so that the accuracy of the obtained prediction parameter is guaranteed.
Further, the step of inputting the label-free data into a preset generation countermeasure network for model training to obtain a pre-training parameter includes:
and h, determining a preset L2 loss function in the generated countermeasure network, and inputting the label-free data into the L2 loss function for training calculation to obtain a pre-training parameter.
In this embodiment, when the generation of the countermeasure network is used for model training, the generation of the countermeasure network needs to be set first by a model algorithm, and in this embodiment, an L2 minimum square error loss function, that is, an L2 loss function, may be set. After the L2 loss function in the countermeasure network is determined to be generated, label-free data can be directly input into the L2 loss function for training calculation, and final pre-training parameters are obtained.
In this embodiment, the pre-training parameters are obtained by inputting the label-free data to the L2 loss function in the generated countermeasure network for training calculation, so that the accuracy of the obtained pre-training parameters is ensured.
Further, the step of collecting location fingerprint data in an indoor environment comprises:
k, determining all reference points in the indoor environment, acquiring signal data sent by all the reference points, and determining the signal intensity in all the signal data and the position information corresponding to all the signal data;
and m, taking the signal intensity and the position information as position fingerprint data.
In this embodiment, an indoor environment, such as an indoor experimental area, may be selected first, and a number of signals AP (i.e., signal sources) with fixed positions may be placed on the area, where the number is denoted by a. The experimental area is divided into N × N blocks, the number being denoted by N. The center point of each square grid is taken as a reference point. And (3) performing signal detection by using mobile phone software capable of detecting the WiFi signal strength on the mobile phone, and recording the coordinate (namely position information) and the received signal strength. An example is shown in fig. 8. The set of reference points is R, R = { R1, R2,. ·, rj,.., RN }, rj is the jth reference point, j =1,2,. And N. And the signal strength and location information as location fingerprint data.
In this embodiment, by determining all reference points in the indoor environment, acquiring the signal data sent by each reference point, and using the signal strength and the location information in the signal data as the location fingerprint data, the accuracy of the acquired location fingerprint data is ensured.
In particular, the step of determining all reference points in the indoor environment comprises:
and n, determining a data acquisition area in the indoor environment, and performing grid division on the data acquisition area to obtain reference points, wherein at least one signal source is arranged in each reference point.
In this embodiment, an indoor environment, such as a mall, a hospital, etc., is determined first. And determining a data acquisition area of the indoor environment, performing corresponding grid division to obtain each grid, and taking the central point of each grid as a reference point. And one AP (i.e., signal source) at the first reference point, R1= A1, two APs at the first reference point, R1= [ A1, A2], a APs at the first reference point, R1= [ A1, A2, A3.. Aa ]. Then an RSSI value formed by a APs at the jth reference point can be expressed as: rj = [ A1, A2, A3.. Aa ], j =1, 2.. N. The wifi signal set of the jth reference point at the ith AP is as follows:
B training points are set for each reference point, and then the wifi signal set of the jth reference point at the ith AP is as follows:
the set of wifi signal strength information for the N reference points is RSSI = [ RBA1, RBA 2., RBAN ]
In the embodiment, the reference point is obtained by performing grid division on the data acquisition area in the indoor environment, so that the effectiveness of the obtained reference point is guaranteed.
Further, the step of determining tagged data and untagged data according to the location information in the location fingerprint data includes:
step x, determining position information and signal intensity in the position fingerprint data, and performing normalization processing on the signal intensity to obtain processed signal intensity;
step y, storing the processed signal intensity serving as non-tag data into a preset first fingerprint database;
and z, storing the processed signal intensity and the position information into a second fingerprint database as labeled data.
In this embodiment, after the position fingerprint data is acquired, the rejection processing is required, the signal strength is normalized, each RSSI is increased by 90 and then divided by 70, so that the maximum difference value of the signal strength is 70dB, that is, the specific RSSI value is between-30 dB and-100 dB, and the purpose is to reduce the complexity of calculation. And data collection is recorded into a database. The supervised database (i.e. the second fingerprint database) is a rectangle with the first two columns being X and Y coordinates and the remainder being signal strength. Randomly extracting 1/4 of data at each training point as a test set, and the rest being training sets, aiming at enabling the training sets and the test sets to cover all the regions so as to train out the characteristics of all the regions. The unsupervised database (i.e. the first fingerprint database) has only signal strength, no XY coordinates. XY coordinates are the labels of the data samples herein. Since the acquisition of the tags is laborious and time consuming, another database without tags is collected to prepare for the subsequent generation of the network. Example run results: the input values are: -45, the predicted coordinates are [7,15], the actual coordinates are [7,13], and the error is 2; the input values are: -30, predicted coordinates [9,7], actual coordinates [11,11], and error 4.47.
In this embodiment, the signal intensity in the position fingerprint data is normalized to obtain the processed signal intensity, the processed signal intensity is stored as the non-tag data in the first fingerprint database, and the processed signal intensity and the position information are stored in the second fingerprint database together, so that the accuracy of determining the tagged data and the non-tag data is ensured.
In addition, referring to fig. 3, an embodiment of the present invention further provides an indoor positioning device, including:
the system comprises an acquisition module A10, a storage module and a processing module, wherein the acquisition module A is used for acquiring position fingerprint data in an indoor environment and determining labeled data and non-labeled data according to position information in the position fingerprint data;
the training module A20 is used for inputting the label-free data into a preset generation countermeasure network for training to obtain a pre-training parameter;
and the determining module A30 is used for inputting the labeled data and the pre-training parameters into a preset prediction model for training to obtain prediction parameters, and determining the target position according to the prediction parameters.
Optionally, the determining module a30 is configured to:
taking the pre-training parameters as network initial values in a preset prediction model, and inputting the labeled data to an input layer in the prediction model;
transmitting the labeled data to a convolutional layer in the prediction model through an input layer in the prediction model to perform feature extraction, so as to obtain a plurality of initial features;
and transmitting each initial characteristic to a pooling layer in the prediction model through the convolution layer for filtering processing to obtain important characteristics, and determining a prediction parameter according to the important characteristics and the network initial value.
Optionally, the determining module a30 is configured to:
determining an RNN hidden layer in the prediction model, and transferring the important features from the pooling layer to the RNN hidden layer;
if a plurality of important features exist, calculating each important feature according to a hidden layer calculation formula in the RNN hidden layer to obtain a time sequence arrangement relation between each important feature;
sequencing all the important features according to the time sequence arrangement relation, and calculating the mixed Gaussian probability distribution of each sequenced important feature to obtain an initial predicted value;
and determining a target predicted value in the initial predicted values according to the network initial values, and taking the target predicted value as a prediction parameter.
Optionally, the training module a20 is configured to:
and determining a preset L2 loss function in the generation countermeasure network, and inputting the label-free data into the L2 loss function for training calculation to obtain a pre-training parameter.
Optionally, the acquisition module a10 is configured to:
determining all reference points in the indoor environment, acquiring signal data sent by all the reference points, and determining signal intensity in all the signal data and position information corresponding to all the signal data;
and using the signal strength and the position information as position fingerprint data.
Optionally, the acquisition module a10 is configured to:
and determining a data acquisition area in the indoor environment, and performing grid division on the data acquisition area to obtain reference points, wherein at least one signal source is arranged in each reference point.
Optionally, the acquisition module a10 is configured to:
determining position information and signal strength in the position fingerprint data, and performing normalization processing on the signal strength to obtain processed signal strength;
storing the processed signal intensity serving as label-free data to a preset first fingerprint database;
and storing the processed signal intensity and the position information into a second fingerprint database as labeled data.
The steps implemented by each functional module of the indoor positioning device may refer to each embodiment of the indoor positioning method of the present invention, and are not described herein again.
In addition, the present invention also provides an indoor positioning apparatus, including: a memory, a processor, and an indoor positioning program stored on the memory; the processor is configured to execute the indoor positioning program to implement the steps of the embodiments of the indoor positioning method.
The present invention also provides a computer readable storage medium storing one or more programs, which are also executable by one or more processors for implementing the steps of the embodiments of the indoor positioning method described above.
The specific implementation of the computer-readable storage medium of the present invention is substantially the same as the embodiments of the indoor positioning method, and is not described herein again.
It should be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or system. Without further limitation, an element defined by the phrases "comprising one of 8230; \8230;" 8230; "does not exclude the presence of additional like elements in a process, method, article, or system that comprises the element.
The above-mentioned serial numbers of the embodiments of the present invention are merely for description and do not represent the merits of the embodiments.
Through the above description of the embodiments, those skilled in the art will clearly understand that the method of the above embodiments can be implemented by software plus a necessary general hardware platform, and certainly can also be implemented by hardware, but in many cases, the former is a better implementation manner. Based on such understanding, the technical solution of the present invention may be embodied in the form of a software product, which is stored in a storage medium (e.g., ROM/RAM, magnetic disk, optical disk) as described above and includes instructions for enabling a terminal device (e.g., a mobile phone, a computer, a server, an air conditioner, or a network device) to execute the method according to the embodiments of the present invention.
The above description is only a preferred embodiment of the present invention, and not intended to limit the scope of the present invention, and all modifications of equivalent structures and equivalent processes, which are made by using the contents of the present specification and the accompanying drawings, or directly or indirectly applied to other related technical fields, are included in the scope of the present invention.
Claims (6)
1. An indoor positioning method, characterized in that the indoor positioning method comprises the following steps:
collecting position fingerprint data in an indoor environment, and determining labeled data and non-labeled data according to position information in the position fingerprint data; wherein the step of determining tagged data and non-tagged data according to location information in the location fingerprint data comprises: determining position information and signal intensity in the position fingerprint data, and performing normalization processing on the signal intensity to obtain processed signal intensity; storing the processed signal intensity serving as non-tag data into a preset first fingerprint database; storing the processed signal intensity and the position information into a second fingerprint database as tagged data;
inputting the label-free data into a preset generation countermeasure network for training to obtain a pre-training parameter; the step of inputting the label-free data into a preset generation countermeasure network for model training to obtain a pre-training parameter comprises the following steps: determining a preset L2 loss function in a generated countermeasure network, and inputting the label-free data into the L2 loss function for training calculation to obtain a pre-training parameter;
inputting the labeled data and the pre-training parameters into a preset prediction model for training to obtain prediction parameters, and determining a target position according to the prediction parameters; inputting the labeled data and the pre-training parameters into a preset prediction model for training to obtain prediction parameters, wherein the step of obtaining the prediction parameters comprises the following steps:
taking the pre-training parameters as network initial values in a preset prediction model, and inputting the labeled data to an input layer in the prediction model;
transmitting the labeled data to a convolutional layer in the prediction model through an input layer in the prediction model to perform feature extraction, so as to obtain a plurality of initial features;
transmitting each initial feature to a pooling layer in the prediction model through the convolutional layer for filtering processing to obtain important features, and determining a prediction parameter according to the important features and the network initial value; wherein, the step of determining the prediction parameters according to the important features and the initial network values comprises the following steps:
determining an RNN hidden layer in the prediction model, and transferring the important features from the pooling layer to the RNN hidden layer;
if a plurality of important features exist, calculating each important feature according to a hidden layer calculation formula in the RNN hidden layer to obtain a time sequence arrangement relation between each important feature;
sequencing all the important features according to the time sequence arrangement relation, and calculating the mixed Gaussian probability distribution of each sequenced important feature to obtain an initial predicted value;
and determining a target predicted value in the initial predicted values according to the network initial values, and taking the target predicted value as a prediction parameter.
2. The indoor positioning method of claim 1, wherein the step of collecting location fingerprint data in an indoor environment comprises:
determining all reference points in the indoor environment, acquiring signal data sent by all the reference points, and determining signal intensity in all the signal data and position information corresponding to all the signal data;
and using the signal strength and the position information as position fingerprint data.
3. The indoor positioning method of claim 2, wherein the step of determining all reference points in the indoor environment comprises:
and determining a data acquisition area in the indoor environment, and performing grid division on the data acquisition area to obtain reference points, wherein at least one signal source is arranged in each reference point.
4. An indoor positioning device, characterized in that, indoor positioning device includes:
the system comprises an acquisition module, a storage module and a processing module, wherein the acquisition module is used for acquiring position fingerprint data in an indoor environment and determining labeled data and non-labeled data according to position information in the position fingerprint data; wherein the step of determining tagged data and non-tagged data according to location information in the location fingerprint data comprises: determining position information and signal strength in the position fingerprint data, and performing normalization processing on the signal strength to obtain processed signal strength; storing the processed signal intensity serving as non-tag data into a preset first fingerprint database; storing the processed signal intensity and the position information into a second fingerprint database as labeled data;
the training module is used for inputting the label-free data into a preset generation countermeasure network for training to obtain a pre-training parameter; the step of inputting the label-free data into a preset generation countermeasure network for model training to obtain a pre-training parameter comprises the following steps: determining a preset L2 loss function in a generated countermeasure network, and inputting the label-free data into the L2 loss function for training calculation to obtain a pre-training parameter;
the determining module is used for inputting the labeled data and the pre-training parameters into a preset prediction model for training to obtain prediction parameters, and determining a target position according to the prediction parameters; inputting the labeled data and the pre-training parameters into a preset prediction model for training to obtain prediction parameters, wherein the step of obtaining the prediction parameters comprises the following steps: taking the pre-training parameters as network initial values in a preset prediction model, and inputting the labeled data to an input layer in the prediction model; transmitting the labeled data to a convolutional layer in the prediction model through an input layer in the prediction model to perform feature extraction, so as to obtain a plurality of initial features; transmitting each initial feature to a pooling layer in the prediction model through the convolutional layer for filtering processing to obtain important features, and determining a prediction parameter according to the important features and the network initial value; wherein, the step of determining the prediction parameters according to the important characteristics and the network initial values comprises the following steps: determining an RNN hidden layer in the prediction model, and transferring the important features from the pooling layer to the RNN hidden layer; if a plurality of important features exist, calculating each important feature according to a hidden layer calculation formula in the RNN hidden layer to obtain a time sequence arrangement relation between each important feature; sequencing all the important features according to the time sequence arrangement relation, and calculating the mixed Gaussian probability distribution of each sequenced important feature to obtain an initial predicted value; and determining a target predicted value in the initial predicted values according to the network initial values, and taking the target predicted value as a prediction parameter.
5. An indoor positioning apparatus, characterized in that the indoor positioning apparatus includes: memory, a processor and an indoor positioning program stored on the memory and executable on the processor, the indoor positioning program when executed by the processor implementing the steps of the indoor positioning method of any one of claims 1 to 3.
6. A computer-readable storage medium, characterized in that the computer-readable storage medium has stored thereon an indoor positioning program, which when executed by a processor implements the steps of the indoor positioning method of any one of claims 1 to 3.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111000063.3A CN113747385B (en) | 2021-08-27 | 2021-08-27 | Indoor positioning method, device, equipment and computer readable storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111000063.3A CN113747385B (en) | 2021-08-27 | 2021-08-27 | Indoor positioning method, device, equipment and computer readable storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113747385A CN113747385A (en) | 2021-12-03 |
| CN113747385B true CN113747385B (en) | 2022-12-16 |
Family
ID=78733635
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111000063.3A Active CN113747385B (en) | 2021-08-27 | 2021-08-27 | Indoor positioning method, device, equipment and computer readable storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113747385B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115482836A (en) * | 2022-08-05 | 2022-12-16 | 深圳市景创科技电子股份有限公司 | Infant AI monitoring method and device |
| CN115426710A (en) * | 2022-08-15 | 2022-12-02 | 浙江工业大学 | Indoor fingerprint positioning-oriented sparse feature completion method |
| CN116761249B (en) * | 2022-09-23 | 2024-04-12 | 荣耀终端有限公司 | Indoor positioning method, fingerprint library construction method, electronic equipment and storage medium |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110110745A (en) * | 2019-03-29 | 2019-08-09 | 上海海事大学 | Based on the semi-supervised x-ray image automatic marking for generating confrontation network |
| CN111930992A (en) * | 2020-08-14 | 2020-11-13 | 腾讯科技（深圳）有限公司 | Neural network training method and device and electronic equipment |
| US20210241118A1 (en) * | 2020-01-30 | 2021-08-05 | Visa International Service Association | System, Method, and Computer Program Product for Implementing a Generative Adversarial Network to Determine Activations |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11751057B2 (en) * | 2019-09-19 | 2023-09-05 | Colorado State University Research Foundation | Security-enhanced deep learning fingerprint-based indoor localization |
-
2021
- 2021-08-27 CN CN202111000063.3A patent/CN113747385B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110110745A (en) * | 2019-03-29 | 2019-08-09 | 上海海事大学 | Based on the semi-supervised x-ray image automatic marking for generating confrontation network |
| US20210241118A1 (en) * | 2020-01-30 | 2021-08-05 | Visa International Service Association | System, Method, and Computer Program Product for Implementing a Generative Adversarial Network to Determine Activations |
| CN111930992A (en) * | 2020-08-14 | 2020-11-13 | 腾讯科技（深圳）有限公司 | Neural network training method and device and electronic equipment |
Non-Patent Citations (1)
| Title |
|---|
| 基于改进支持向量回归的室内定位算法;姚英彪等;《仪器仪表学报》;20170915(第09期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113747385A (en) | 2021-12-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113747385B (en) | Indoor positioning method, device, equipment and computer readable storage medium | |
| CN110927706B (en) | Convolutional neural network-based radar interference detection and identification method | |
| US20210174212A1 (en) | Operating Machine-Learning Models on Different Platforms | |
| CN112036433B (en) | CNN-based Wi-Move behavior sensing method | |
| CN108924736A (en) | A kind of passive indoor occupant condition detection method based on PCA-Kalman | |
| CN113221787A (en) | Pedestrian multi-target tracking method based on multivariate difference fusion | |
| CN113095370A (en) | Image recognition method and device, electronic equipment and storage medium | |
| CN112764003A (en) | Radar radiation source signal time-frequency feature identification method and device and storage medium | |
| JP2022545969A (en) | System and method for event recognition | |
| CN111901028B (en) | Human body behavior identification method based on CSI (channel State information) on multiple antennas | |
| CN114781463A (en) | Cross-scene robust indoor tumble wireless detection method and related equipment | |
| CN114114363A (en) | Opportunistic signal sensing method and system based on time frequency and convolutional neural network and opportunistic signal positioning method | |
| CN114091554A (en) | Training set processing method and device | |
| CN115294615A (en) | Radio frequency fingerprint identification method based on machine learning | |
| CN110363075B (en) | Suspicious ill face detection system based on big data server | |
| CN107358246A (en) | A kind of method for being finely divided class to object using convolutional neural networks | |
| CN118094371A (en) | Rolling bearing fault diagnosis method based on multi-scale residual error network and improved GRU | |
| CN114155489A (en) | Multi-device cooperative unmanned aerial vehicle flyer detection method, device and storage medium | |
| CN117221816A (en) | Multi-building floor positioning method based on Wavelet-CNN | |
| Shi et al. | WiFi-based activity recognition using activity filter and enhanced correlation with deep learning | |
| CN114724245B (en) | Incremental learning human body action recognition method based on CSI | |
| US20230016455A1 (en) | Decomposing a deconvolution into multiple convolutions | |
| Sonny et al. | Fingerprint Image-Based Multi-Building 3D Indoor Wi-Fi Localization Using Convolutional Neural Networks | |
| CN114926745A (en) | Small-sample SAR target identification method based on domain feature mapping | |
| CN113726457A (en) | Human body action recognition method and device, storage medium and network equipment |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |