CN110225460B - Indoor positioning method and device based on deep neural network - Google Patents

Indoor positioning method and device based on deep neural network Download PDF

Info

Publication number
CN110225460B
CN110225460B CN201910488845.2A CN201910488845A CN110225460B CN 110225460 B CN110225460 B CN 110225460B CN 201910488845 A CN201910488845 A CN 201910488845A CN 110225460 B CN110225460 B CN 110225460B
Authority
CN
China
Prior art keywords
neural network
deep neural
layer
rss
indoor positioning
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
Application number
CN201910488845.2A
Other languages
Chinese (zh)
Other versions
CN110225460A (en
Inventor
张佳琪
刘胜利
钱国良
余官定
汪运平
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Zhejiang University ZJU
Sunwave Communications Co Ltd
Original Assignee
Zhejiang University ZJU
Sunwave Communications Co Ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Zhejiang University ZJU, Sunwave Communications Co Ltd filed Critical Zhejiang University ZJU
Priority to CN201910488845.2A priority Critical patent/CN110225460B/en
Publication of CN110225460A publication Critical patent/CN110225460A/en
Application granted granted Critical
Publication of CN110225460B publication Critical patent/CN110225460B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • 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
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • 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
    • H04W4/021—Services related to particular areas, e.g. point of interest [POI] services, venue services or geofences
    • 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

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)
  • Position Fixing By Use Of Radio Waves (AREA)

Abstract

The invention discloses an indoor positioning method based on a deep neural network, which comprises the following steps: acquiring RSS reference data of the Wi-Fi access point, filtering the acquired RSS reference data, and constructing an RSS reference database of the Wi-Fi access point based on the RSS reference data obtained after processing; constructing a deep neural network, and training by using the RSS reference database as a training sample to obtain a deep neural network model; and preprocessing RRS data of the to-be-positioned points, and inputting the preprocessed RRS data into the deep neural network model obtained by training to obtain position estimation coordinates of the to-be-positioned points. Compared with a plurality of indoor positioning methods which need people to carry the sensor, the indoor positioning method does not need to carry a special sensor, does not need to add extra equipment, can realize indoor positioning only by a smart phone, has great convenience and is convenient to popularize.

Description

Indoor positioning method and device based on deep neural network
Technical Field
The invention relates to the field of indoor wireless positioning, in particular to an indoor positioning method and device based on a deep neural network.
Background
With the rapid development of the internet and wireless technologies, the new form of intelligent economic development based on the internet changes the life style of people. The intelligent terminal almost permeates every corner of people's life, and various applications with endless emergence are generated, wherein the applications comprise positioning technology. At present, people can realize accurate positioning with the accuracy of one meter outdoors by means of satellite positioning technologies such as a global positioning system, Beidou and the like. Under the background of a highly developed information age, the demand of mobile users for position information is not limited to outdoor rough positioning, and the demand of accurate real-time positioning is increasing day by day. However, due to the shielding of building materials such as building walls and buildings, the satellite signals in the room are greatly attenuated, which results in serious misalignment of the positioning and reduced accuracy. Currently, an indoor positioning technology which can be widely used, is mature and has high precision is still lacked. Therefore, research on indoor positioning technology is becoming a current research focus.
The existing indoor positioning technologies mainly include Bluetooth (Bluetooth) technology, Infrared (Infrared) technology, RFID technology, and Wi-Fi technology. Among the above indoor positioning technologies, the bluetooth technology requires a small size of the device, but it has disadvantages of short transmission distance and poor stability in a complex environment; the infrared technology requires visibility between detectors and between targets to be positioned, and is difficult to realize in a complex indoor environment; the RFID technology has high positioning accuracy, but the anti-interference capability is poor. Furthermore, in addition to Wi-Fi technology, the rest of indoor positioning technology does not have a well-established suitable infrastructure in existing public places, and a significant time and economic investment is required to establish such infrastructure. With the Wi-Fi signals covering various places in life of people, such as shopping malls, airports, office places and the like, the Wi-Fi signals have inherent advantages when being applied to indoor positioning, and become an important research field in the indoor positioning technology.
The Wi-Fi-based indoor positioning method mainly comprises two categories, namely a Received Signal Strength (RSS) -based positioning method and a ranging-based positioning model, wherein the positioning algorithm based on the ranging model needs to estimate an indoor channel environment in advance and establish a channel model, the indoor positioning method based on the ranging model cannot be suitable for a new environment due to the characteristic of complex and changeable indoor environment, and the RSS-based indoor positioning algorithm has outstanding advantages due to the fact that the indoor positioning method can be suitable for the complex indoor environment.
The traditional Wi-Fi indoor positioning algorithm mainly comprises: K-Nearest Neighbor (K-Nearest Neighbor, KNN) and weighted K-Nearest Neighbor (Weight K-Nearest Neighbor, WKNN). Compared with other Wi-Fi-based indoor positioning algorithms, the algorithm has the advantages of low calculation complexity, high running speed and easiness in implementation. However, due to the popularization of Wi-Fi technology, the number of Wi-Fi hotspots in a large site is gradually increased, sometimes reaching hundreds of numbers, and the number of positioning reference points is also increased, so that the positioning efficiency of the locations is greatly influenced. Meanwhile, the WKNN algorithm cannot adaptively acquire the effective K value of the real-time positioning point in the positioning stage, and thus manual setting is required, and the positioning accuracy cannot be guaranteed.
Therefore, a solution that can shorten the positioning time and ensure the positioning efficiency when the amount of RSS data of Wi-Fi becomes huge is urgently needed.
Disclosure of Invention
In order to solve the problems in the prior art, the invention provides an indoor positioning method based on a deep neural network. On the premise of ensuring the positioning accuracy, the deep learning algorithm is applied to the indoor positioning technology, so that the indoor positioning system can shorten the positioning time when processing massive Wi-Fi RSS data.
In order to achieve the purpose, the invention provides the following technical scheme:
an indoor positioning method based on a deep neural network comprises the following steps:
s1: the method comprises the steps of acquiring RSS reference data of a Wi-Fi access point, preprocessing the acquired RSS reference data, and constructing an RSS reference database of the Wi-Fi access point based on the preprocessed RSS reference data;
s2: constructing a deep neural network model, and training by using the RSS reference database as a training sample to obtain the deep neural network model;
s3: and preprocessing RRS data of the to-be-positioned point, inputting the preprocessed RRS data into the deep neural network model obtained through training, and outputting to obtain position estimation coordinates of the to-be-positioned point.
Further, the step S1 specifically includes the following steps:
step S11, dividing the room to be positioned into L areas along the transverse direction and the longitudinal direction, wherein the transverse direction is divided into L1 areas along the maximum length direction, the longitudinal direction is divided into L2 areas along the maximum length direction, L is L1 multiplied by L2, and the center of each area is taken as the position of the area; if the region is a non-centrosymmetric region, the intersection point of the perpendicular bisectors of the transverse sidelines and the longitudinal sidelines of the region is taken as the center of the region.
Step S12, uniformly setting N reference points in each area, and acquiring the position information of the N reference points, wherein N is a positive integer;
step S13, obtaining received RSS data of M Wi-Fi access points corresponding to N reference point positions, wherein M is a positive integer;
step S14, performing filtering processing on the RSS data obtained in the previous step to remove background noise, and constructing a Wi-Fi access point RSS reference database of N reference points based on the RSS reference data obtained after the filtering processing.
Further, the deep neural network adopts a back propagation algorithm to adjust the connection weight between the neurons in the neural network, specifically refers to a prediction of a target signal obtained by propagating to the last layer according to a forward response, so as to further obtain an error between the target signal and a reference signal, and continuously adjusts the connection weight of the neural network according to the obtained error value, wherein the learning process of the back propagation algorithm can be divided into the following four stages:
(1) forward propagation: sample set { (x)(1),y(1)),...,(x(m),y(m)) The process of layer-by-layer propagation from the input layer to the output layer through the hidden layer;
(2) error inverse propagation: the difference between the expected output and the actual output of the neural network, namely the error signal is propagated from the output layer to the input layer by layer through the hidden layer, and the process of connecting the weights is corrected;
(3) memory training: the memory training process is carried out by repeatedly and alternately carrying out forward propagation and error reverse propagation;
(4) and (3) convergence in learning: a process of network global error tending to a minimum value;
the neurons in each layer in the deep neural network are in a fully-connected relation, no connection exists among the neurons in each layer, y is recorded as a given target value, L (·) is recorded as an objective function, any set of input modes (x (i), y (i)), i ═ 1, 2.
Figure BDA0002085545510000031
The sum is averaged over N data and let:
Figure BDA0002085545510000032
wherein w is a weight value in the neural network neuron, b is a bias value, and θ is other parameters;
Figure BDA0002085545510000033
is a weight decay term;
obtaining a target function value through feedforward calculation, then adopting a gradient descent algorithm according to the criterion of a minimum loss function, and reversely propagating a gradient value, wherein the weight learning rule is as follows:
Figure BDA0002085545510000034
Figure BDA0002085545510000041
ε represents the learning rate, defining the magnitude of each parameter update.
Further, the step S3 further includes determining the area to which the point to be located belongs, and acquiring the AP to be locatediThe location information of (1).
Further, in step S2, the deep neural network adapts a self-adaptive learning rate.
Furthermore, the adaptive adjustment learning rate adopts an Adam algorithm, and the learning rate of each parameter is dynamically adjusted according to the gradient first moment estimation and the second moment estimation of each parameter in the neural network by the loss function.
An indoor positioning device based on a deep neural network, comprising:
the position information acquisition module is used for acquiring received RSS data of M Wi-Fi access points corresponding to the positions of N reference points, wherein M is a positive integer; the module is preferably a smart terminal including, but not limited to, a smart communication device, such as a smart phone, a tablet computer, and the like.
The RSS reference database construction module is used for preprocessing the RSS reference data of the Wi-Fi access point acquired by the position information acquisition module and constructing an RSS reference database based on the RSS reference data acquired after preprocessing;
the deep neural network model construction and position coordinate positioning module is used for training by taking the RSS reference database as a training sample to obtain a deep neural network model; and preprocessing RRS data of the to-be-positioned point, inputting the preprocessed RRS data into a deep neural network model obtained through training, and outputting to obtain position estimation coordinates of the to-be-positioned point.
Further, the preprocessing is filtering to remove background noise.
Further, the preprocessing is to remove background noise by using low-pass filtering. Preferably, the low-pass filtering is implemented by a butterworth low-pass filter.
Compared with the prior art, the invention has the beneficial effects that:
1. compared with a plurality of indoor positioning methods which need people to carry the sensor, the indoor positioning method has the advantages that no additional equipment is added, the indoor positioning can be realized only by the smart phone or the tablet personal computer, the convenience is high, and the popularization is facilitated.
2. A butterworth filter is applied to the RSS signal to filter out high frequency background noise.
3. Machine learning is applied to an indoor positioning technology, an accurate mathematical model is established to predict the indoor position of a person, and the indoor positioning precision is effectively improved.
4. The deep learning algorithm is applied to the indoor positioning technology, and under the condition of ensuring the positioning efficiency and accuracy, the indoor positioning system can shorten the positioning time when having massive Wi-Fi RSS data.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below. It is obvious that the drawings in the following description are only some embodiments of the invention, and that for a person skilled in the art, other drawings can be derived from them without inventive effort.
FIG. 1 is a block diagram of the system architecture of the present invention;
FIG. 2 is a diagram of the location distribution of an experimental site and Wi-Fi access points AP of the present invention;
FIG. 3 is a block diagram of the deep neural network of the present invention;
fig. 4 is a graph of the error accumulation distribution of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be further described in detail with reference to the accompanying drawings and examples. It should be understood that the detailed description and specific examples, while indicating the scope of the invention, are intended for purposes of illustration only and are not intended to limit the scope of the invention.
Fig. 1 is a flowchart of an indoor positioning method according to an embodiment of the present invention, and as shown in fig. 1, in an embodiment of the present invention, the indoor positioning method based on a deep neural network according to the present invention includes:
and step S1, acquiring RSS reference data of the Wi-Fi access point, filtering the acquired RSS reference data, and constructing a Wi-Fi access point RSS reference database based on the processed RSS reference data.
Wherein the Wi-Fi access point RSS reference data refers to Wi-Fi access point RSS data received at a reference point location. The reference point may be one or more, and the Wi-Fi access point AP may be one or more. In one embodiment of the present invention, a room to be located is divided into L regions in the transverse direction and the longitudinal direction, wherein the transverse direction is divided into L1 regions in the maximum length direction, the longitudinal direction is divided into L2 regions in the maximum length direction, L is L1 × L2, and the center of each region is taken as the position of the region; if the area is a non-centrosymmetric area, the intersection point of the perpendicular bisectors of the transverse sideline and the longitudinal sideline of the area is taken as the center of the area.
In this embodiment, the room to be positioned is uniformly divided into L areas of L1 × L2, each area uniformly selects N reference points to collect RSS data of a Wi-Fi access point AP, the experimental site and the surrounding areas thereof are provided with M WiFi access points AP, and the coverage area of the WiFi access points arranged around the experimental site should cover the experimental site. The location distribution of the experimental site and WiFi access points is shown in fig. 2.
In this embodiment, the step S1 further includes:
step S11, setting L multiplied by N reference points in total, and acquiring the position information of each reference point;
step S12, obtaining RSS data of the Wi-Fi access point correspondingly received at the position of the LxN reference points;
in step S13, the background noise is removed by filtering with a low-pass filter, preferably a butterworth filter. The frequency of the reflected signal of the walking person is much lower than that of the background noise, so the present embodiment adopts a butterworth low-pass filter to remove the background noise.
In an embodiment of the present invention, the RSS reference database of the belonging Wi-Fi access point may be as shown in table 1, and the information of the RSS reference data of each Wi-Fi access point is represented by RSSI:
TABLE 1 Wi-Fi Access Point RSS reference database
Figure BDA0002085545510000061
In the table, the label represents the region number to which the reference point belongs, denoted as l, and in the present embodiment, l (l ∈ [1,42 ]])。lRPjIs shown in the regionThe jth reference point in the region with field number l, j ═ 1 … N, N denotes the number of reference points, (x) is set toj,yj) Indicating reference pointslRPjPosition coordinates of, APiDenotes the ith Wi-Fi access point, i-1 … M, M denotes the number of Wi-Fi access points,lRSSIijindicating the jth reference point within the region with region number llRPjIs received from the ith Wi-Fi access point APiRSS reference data, i.e. reference pointslRPjMay be expressed as: [ l, (x)j,yj)，RSSI1j，…,RSSIij,…,RSSIMj]。
And step S2, constructing a deep neural network model, and training by using a database as a training sample to obtain the deep neural network model.
The structure of the deep neural network model in an embodiment of the present invention is shown in fig. 3, which has an input layer, an output layer and a plurality of hidden layers. The deep neural network adopts a back propagation Algorithm (BP) to adjust the connection weight between the neurons in the neural network. Specifically, the difference between the target signal and the reference signal is further obtained according to the prediction of the target signal obtained by the forward response propagation to the last layer, and the difference is an error needing to be propagated back. And according to the obtained error value, the connection weight of the neural network can be continuously adjusted. The learning process of the back propagation algorithm can be divided into the following four stages:
(1) forward propagation: sample set { (x)(1),y(1)),...,(x(m),y(m)) And (4) propagating from the input layer to the output layer by layer through the hidden layer.
(2) Error inverse propagation: the difference between the expected output and the actual output of the neural network, namely the error signal, is propagated from the output layer to the input layer by layer through the hidden layer, and the process of correcting the connection weight is carried out.
(3) Memory training: and the forward propagation and the error reverse propagation are repeatedly and alternately used for carrying out a memory training process.
(4) And (3) convergence in learning: the global error of the network tends to the minimum value.
The neurons in each layer in the BP neural network are in a full connection relation, and the neurons in each layer are not connected. Let y denote a given target value, L (-) denote an objective function, and given any set of input patterns (x (i), y (i)), i ═ 1, 2.., N, its loss function can be expressed as:
Figure BDA0002085545510000071
the sum is averaged over N data and let:
Figure BDA0002085545510000072
wherein w is a weight value in the neural network neuron, b is a bias value, and θ is other parameters;
Figure BDA0002085545510000081
is a weight decay term whose purpose is to reduce the magnitude of the weight and prevent overfitting, and the weight decay parameter is used to control the relative importance of the two terms in the formula.
Obtaining a target function value through feedforward calculation, then adopting a gradient descent algorithm according to the criterion of a minimum loss function, and reversely propagating a gradient value, wherein the weight learning rule is as follows:
Figure BDA0002085545510000082
Figure BDA0002085545510000083
epsilon represents the learning rate and defines the magnitude of each parameter update, and in one embodiment of the invention, a method of adaptively adjusting the learning rate is adopted. The algorithm specifically adopted is an adam (adaptive motion estimation) algorithm, and the learning rate for each parameter is dynamically adjusted according to the first Moment estimation and the second Moment estimation of the gradient of each parameter in the neural network by the loss function. Adam is also a gradient descent-based method, but the learning step length of the parameters in each iteration has a certain range, so that a large learning step length cannot be caused by a large gradient, the values of the parameters are stable, and the method is suitable for large data sets and high-dimensional spaces.
And step S3, when the point to be positioned is positioned, the RSS data of the point to be positioned is preprocessed and then input into the trained neural network model, and the position estimation coordinates of the point to be positioned can be obtained.
In this embodiment, the step S3 further includes:
step S31, determining AP to be positionediBelonging area, and obtaining AP of point to be locatediThe location information of (a);
in an embodiment of the present invention, an AP to be locatediCan be expressed as (x)i,yi) And the area is marked as l.
Step S32, at the AP point to be positionediMeasuring an RSS from the Wi-Fi access point;
in one embodiment of the invention, the number of the Wi-Fi access points is M, and the access points to be positioned are the access points to be positionediThe RSS measured from the M access points can be expressed as RSSI'1i，RSSI’2i，……，RSSI’Mi]。
Step S33, AP to be positionediFiltering the RSS of the Wi-Fi access point to obtain RSS data of a to-be-positioned point;
wherein the filtering uses the same filter processing as mentioned above for the Wi-Fi access point RSS reference data to remove background noise.
Step S34, outputting the point P to be positioned by the deep neural networkiIs noted as (x)i’,yi’)。
The accuracy rate is the proportion of all test data in which the label classification is correct, and can be expressed as:
Figure BDA0002085545510000091
the average positioning error represents the distance between the predicted position and the actual position and can be expressed as:
Figure BDA0002085545510000092
as can be seen from the CDF positioning error cumulative distribution function in fig. 4, the positioning performance has a probability positioning error of more than 60% which can be smaller than 2m, and a probability positioning error of about 80% which is smaller than 3 m.

Claims (10)

1. An indoor positioning method based on a deep neural network comprises the following steps:
s1: the method comprises the steps of acquiring RSS reference data of a Wi-Fi access point, preprocessing the acquired RSS reference data, and constructing an RSS reference database of the Wi-Fi access point based on the preprocessed RSS reference data;
s2: constructing a deep neural network model, and training by using the RSS reference database as a training sample to obtain the deep neural network model;
s3: RRS data of a point to be positioned is input into a deep neural network model obtained by training after being preprocessed, position estimation coordinates of the point to be positioned are output and obtained, wherein,
the deep neural network adopts a back propagation algorithm to adjust the connection weight between the neurons in the neural network, specifically refers to the prediction of a target signal obtained by transmitting to the last layer according to the forward response propagation, so as to further obtain the error between the target signal and a reference signal, and continuously adjusts the connection weight of the neural network according to the obtained error value, wherein the learning process of the back propagation algorithm can be divided into the following four stages:
(1) forward propagation: sample set { (x)(1),y(1)),...,(x(m),y(m)) The process of layer-by-layer propagation from the input layer to the output layer through the hidden layer;
(2) error inverse propagation: the difference between the expected output and the actual output of the neural network, namely the error signal is propagated from the output layer to the input layer by layer through the hidden layer, and the process of connecting the weights is corrected;
(3) memory training: the memory training process is carried out by repeatedly and alternately carrying out forward propagation and error reverse propagation;
(4) and (3) convergence in learning: a process of network global error tending to a minimum value;
the neurons in each layer in the deep neural network are in a fully-connected relation, no connection exists among the neurons in each layer, y is recorded as a given target value, L (·) is recorded as an objective function, any set of input modes (x (i), y (i)), i ═ 1, 2.
Figure FDA0002893711750000011
The sum is averaged over N data and let:
Figure FDA0002893711750000012
wherein w is a weight value in the neural network neuron, b is a bias value, and θ is other parameters;
Figure FDA0002893711750000013
is a weight decay term;
obtaining a target function value through feedforward calculation, then adopting a gradient descent algorithm according to the criterion of a minimum loss function, and reversely propagating a gradient value, wherein the weight learning rule is as follows:
Figure FDA0002893711750000021
Figure FDA0002893711750000022
ε represents the learning rate, defining the magnitude of each parameter update.
2. The deep neural network-based indoor positioning method of claim 1, wherein: the step S1 specifically includes the following steps:
step S11, dividing the room to be positioned into L areas along the transverse direction and the longitudinal direction, wherein the transverse direction is divided into L1 areas along the maximum length direction, the longitudinal direction is divided into L2 areas along the maximum length direction, L is L1 multiplied by L2, and the center of each area is taken as the position of the area; if the area is a non-centrosymmetric area, taking the intersection point of the perpendicular bisectors of the transverse sideline and the longitudinal sideline of the area as the center;
step S12, uniformly setting N reference points in each area, and acquiring the position information of the N reference points, wherein N is a positive integer;
step S13, obtaining received RSS data of M Wi-Fi access points corresponding to N reference point positions, wherein M is a positive integer;
step S14, performing filtering processing on the RSS data obtained in the previous step to remove background noise, and constructing a Wi-Fi access point RSS reference database of N reference points based on the RSS reference data obtained after the filtering processing.
3. The deep neural network-based indoor positioning method of claim 1, wherein: the step S3 further includes determining the area to which the point to be located belongs, and acquiring the AP to be locatediThe location information of (1).
4. The deep neural network-based indoor positioning method of claim 1, wherein: in step S2, the deep neural network adaptively adjusts a learning rate.
5. The deep neural network-based indoor positioning method of claim 4, wherein: the self-adaptive adjustment learning rate adopts an Adam algorithm, and the learning rate of each parameter is dynamically adjusted according to the gradient first moment estimation and the second moment estimation of each parameter in the neural network by the loss function.
6. An indoor positioning device based on a deep neural network, comprising:
the position information acquisition module is used for acquiring received RSS data of M Wi-Fi access points corresponding to the positions of N reference points, wherein N, M is a positive integer;
the RSS reference database construction module is used for preprocessing the RSS reference data of the Wi-Fi access point acquired by the position information acquisition module and constructing an RSS reference database based on the RSS reference data acquired after preprocessing;
the deep neural network model construction and position coordinate positioning module is used for training by taking the RSS reference database as a training sample to obtain a deep neural network model; the method comprises the steps that RRS data of a to-be-positioned point are input into a deep neural network model obtained through training after being preprocessed, and position estimation coordinates of the to-be-positioned point are output and obtained, wherein the deep neural network adjusts connection weights among neurons in the neural network by adopting a back propagation algorithm, specifically, the connection weights are transmitted to the last layer according to forward response propagation to obtain target signal prediction, so that errors between the target signal prediction and reference signals are further obtained, the connection weights of the neural network are continuously adjusted according to obtained error values, and the learning process of the back propagation algorithm can be divided into the following four stages:
(1) forward propagation: sample set { (x)(1),y(1)),...,(x(m),y(m)) The process of layer-by-layer propagation from the input layer to the output layer through the hidden layer;
(2) error inverse propagation: the difference between the expected output and the actual output of the neural network, namely the error signal is propagated from the output layer to the input layer by layer through the hidden layer, and the process of connecting the weights is corrected;
(3) memory training: the memory training process is carried out by repeatedly and alternately carrying out forward propagation and error reverse propagation;
(4) and (3) convergence in learning: a process of network global error tending to a minimum value;
the neurons in each layer in the deep neural network are in a fully-connected relation, no connection exists among the neurons in each layer, y is recorded as a given target value, L (·) is recorded as an objective function, any set of input modes (x (i), y (i)), i ═ 1, 2.
Figure FDA0002893711750000031
The sum is averaged over N data and let:
Figure FDA0002893711750000032
wherein w is a weight value in the neural network neuron, b is a bias value, and θ is other parameters;
Figure FDA0002893711750000033
is a weight decay term;
obtaining a target function value through feedforward calculation, then adopting a gradient descent algorithm according to the criterion of a minimum loss function, and reversely propagating a gradient value, wherein the weight learning rule is as follows:
Figure FDA0002893711750000041
Figure FDA0002893711750000042
ε represents the learning rate, defining the magnitude of each parameter update.
7. The deep neural network-based indoor positioning apparatus of claim 6, wherein: the position information acquisition module is an intelligent terminal which comprises intelligent communication equipment.
8. The deep neural network-based indoor positioning apparatus of claim 6, wherein: the preprocessing is filtering to remove background noise.
9. The deep neural network-based indoor positioning apparatus of claim 6, wherein: the preprocessing is to remove background noise by low-pass filtering.
10. The deep neural network-based indoor positioning apparatus of claim 9, wherein: the low-pass filtering is realized by a Butterworth filtering low-pass filter.
CN201910488845.2A 2019-06-05 2019-06-05 Indoor positioning method and device based on deep neural network Active CN110225460B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910488845.2A CN110225460B (en) 2019-06-05 2019-06-05 Indoor positioning method and device based on deep neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910488845.2A CN110225460B (en) 2019-06-05 2019-06-05 Indoor positioning method and device based on deep neural network

Publications (2)

Publication Number Publication Date
CN110225460A CN110225460A (en) 2019-09-10
CN110225460B true CN110225460B (en) 2021-03-23

Family

ID=67819493

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910488845.2A Active CN110225460B (en) 2019-06-05 2019-06-05 Indoor positioning method and device based on deep neural network

Country Status (1)

Country Link
CN (1) CN110225460B (en)

Families Citing this family (15)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110632554A (en) * 2019-09-20 2019-12-31 深圳前海微众银行股份有限公司 Indoor positioning method, device, terminal equipment and medium based on federal learning
US11363416B2 (en) * 2019-10-04 2022-06-14 Samsung Electronics Co., Ltd. System and method for WiFi-based indoor localization via unsupervised domain adaptation
CN110856100B (en) * 2019-10-21 2021-04-23 深圳数位传媒科技有限公司 Method and device for terminal positioning and positioning model construction based on 5G signal
CN110972056B (en) * 2019-11-08 2020-09-29 宁波大学 UWB indoor positioning method based on machine learning
CN110837806B (en) * 2019-11-11 2020-07-28 辽宁科技学院 Indoor positioning method based on deep convolutional memory network
CN111031477A (en) * 2019-12-25 2020-04-17 中山大学 Millimeter wave indoor positioning method based on deep learning
CN113051976A (en) * 2019-12-27 2021-06-29 广东博智林机器人有限公司 Fingerprint positioning method and device, electronic equipment and storage medium
CN111123340B (en) * 2020-02-10 2024-05-28 腾讯科技（深圳）有限公司 Logistics distribution navigation method and system, near-field positioning navigation device and storage medium
CN111654910A (en) * 2020-04-15 2020-09-11 深圳新贝奥科技有限公司 Indoor positioning calculation method based on neural network
CN113543305A (en) * 2020-04-22 2021-10-22 维沃移动通信有限公司 Positioning method, communication equipment and network equipment
CN111523667B (en) * 2020-04-30 2023-06-27 天津大学 RFID positioning method based on neural network
CN111929674B (en) * 2020-07-10 2022-10-04 西安电子科技大学 Intelligent amplitude comparison angle measurement method based on neural network, storage medium and equipment
CN112188388A (en) * 2020-08-25 2021-01-05 海盐南原电力工程有限责任公司 Hybrid indoor positioning method based on machine learning
CN113596724A (en) * 2021-08-05 2021-11-02 广东工业大学 Indoor positioning method, device, equipment and medium based on transfer learning
WO2023097634A1 (en) * 2021-12-03 2023-06-08 Oppo广东移动通信有限公司 Positioning method, model training method, and device

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105101408A (en) * 2015-07-23 2015-11-25 常熟理工学院 Indoor positioning method based on distributed AP selection strategy
CN106658422A (en) * 2017-01-09 2017-05-10 深圳市智开科技有限公司 Network side positioning method and network side positioning system for aiming at high-sparse WiFi data
CN108303672A (en) * 2017-12-26 2018-07-20 武汉创驰蓝天信息科技有限公司 WLAN indoor positionings error correcting method based on location fingerprint and system
CN109447244A (en) * 2018-10-11 2019-03-08 中山大学 A kind of advertisement recommended method of combination gating cycle unit neural network

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11537845B2 (en) * 2017-04-12 2022-12-27 Yodlee, Inc. Neural networks for information extraction from transaction data

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105101408A (en) * 2015-07-23 2015-11-25 常熟理工学院 Indoor positioning method based on distributed AP selection strategy
CN106658422A (en) * 2017-01-09 2017-05-10 深圳市智开科技有限公司 Network side positioning method and network side positioning system for aiming at high-sparse WiFi data
CN108303672A (en) * 2017-12-26 2018-07-20 武汉创驰蓝天信息科技有限公司 WLAN indoor positionings error correcting method based on location fingerprint and system
CN109447244A (en) * 2018-10-11 2019-03-08 中山大学 A kind of advertisement recommended method of combination gating cycle unit neural network

Also Published As

Publication number Publication date
CN110225460A (en) 2019-09-10

Similar Documents

Publication Publication Date Title
CN110225460B (en) Indoor positioning method and device based on deep neural network
CN101778399B (en) Method for optimizing WLAN (Wireless Local Area Network) indoor ANN (Artificial Neural Network) positioning based on FCM (fuzzy C-mean) and least-squares curve surface fitting methods
CN106851573A (en) Joint weighting k nearest neighbor indoor orientation method based on log path loss model
Wang et al. Research on indoor positioning algorithm based on SAGA-BP neural network
CN107801147A (en) One kind is based on the adaptive indoor orientation method of the improved multizone of RSSI rangings
CN107576330A (en) A kind of localization method of the indoor dynamic sensing strategy based on WLAN fingerprints
Zhang et al. Weighted adaptive KNN algorithm with historical information fusion for fingerprint positioning
WO2016079656A1 (en) Zero-calibration accurate rf-based localization system for realistic environments
CN105866732B (en) The mixing indoor orientation method that a kind of improvement MK models and WKNN algorithms are combined
CN110187336A (en) A kind of positioning of multistation radar site and combined tracking method based on distributed PHD
Yang Indoor localization system using dual-frequency bands and interpolation algorithm
Yan et al. A fuzzy decision based WSN localization algorithm for wise healthcare
CN111988844A (en) Indoor room-level positioning method based on block chain and mobile crowd sensing
Sun et al. A novel GCN based indoor localization system with multiple access points
Huan et al. Indoor location fingerprinting algorithm based on path loss parameter estimation and bayesian inference
CN110990500A (en) Propagation path model map establishing method and path loss determining method
CN106707232A (en) WLAN propagation model positioning method based on crowd sensing
CN111505566A (en) Ultrahigh frequency radio frequency signal DOA estimation method
CN117241215A (en) Wireless sensor network distributed node cooperative positioning method based on graph neural network
Huang et al. QSFDEW: a fingerprint positioning method based on quadtree search and fractal direction entropy weighting
CN109581280A (en) The adaptive tuning on-line method, system and device of terminal
Zhou et al. An improved MCB localization algorithm based on weighted RSSI and motion prediction
Wang et al. Indoor fingerprint positioning method based on real 5G signals
KR20150060456A (en) Apparatus and method for generating radio fingerprint map
Oh et al. WiFi Positioning by Optimal k-NN in 3GPP Indoor Office Environment

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