CN106792553A - A kind of many floor location methods and server based on wifi - Google Patents
A kind of many floor location methods and server based on wifi Download PDFInfo
- Publication number
- CN106792553A CN106792553A CN201611046269.9A CN201611046269A CN106792553A CN 106792553 A CN106792553 A CN 106792553A CN 201611046269 A CN201611046269 A CN 201611046269A CN 106792553 A CN106792553 A CN 106792553A
- Authority
- CN
- China
- Prior art keywords
- training
- floor
- deep neural
- neural network
- type deep
- 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.)
- Pending
Links
Classifications
-
- H04W4/043—
Abstract
The invention discloses a kind of many floor location methods based on wifi, methods described includes step：Each WAP receives the signal strength data that client to be detected is signaled in S100, acquisition testing region；S200, by the signal strength data be input into the first deep neural network, the floor label of the affiliated floor in client position to be detected is calculated through first deep neural network；S300, by signal strength data input and one-to-one second deep neural network of floor label of the affiliated floor in client position to be detected；The predicted position coordinate of client position to be detected is calculated through second deep neural network.The present invention is trained by a large amount of training sample data using the deep neural network after training to deep neural network, lifts Position location accuracy and precision.
Description
Technical field
The present invention relates to wireless local area network technology field, more particularly to a kind of many floor location methods and clothes based on wifi
Business device.
Background technology
Location technology worldwide mainly has GPS location, Wi-Fi positioning, bluetooth positioning etc., GPS location at present
Outdoor is mainly used in, Wi-Fi, bluetooth positioning can be not only used for interior, it can also be used to outdoor.Because Wi-Fi positions relative maturity,
Below particular content of the invention is introduced with Wi-Fi location technologies as background.With the popularization of wireless router, current big portion
Point public domain all has been carried out more than ten or even tens WiFi signals coverings, and these routers are propagated to surrounding
While WiFi signal, the information such as its physical address and signal intensity are also ceaselessly sent, as long as in its signal cover,
Even if not knowing the password of Wi-Fi, these information can be similarly obtained.Location technology general principle based on signal intensity is root
The distance between signal receiver and signal source are calculated according to the intensity of the signal for receiving, but because indoor floor has reinforcing bar up and down
The separation layer of concrete, causing the signal of upper and lower floor has notable difference, therefore, used during for being positioned in multiple floors
The obvious precision of conventional method is not high.
The content of the invention
In order to solve the above technical problems, the present invention provides a kind of many floor location methods and server based on wifi, lead to
Cross and gather the corresponding signal strength data of each WAP and determine the position coordinates of floor where client to be detected, it is real
Now the WiFi based on deep neural network is positioned.
The technical scheme that the present invention is provided is as follows：
The invention discloses a kind of many floor location methods based on wifi, methods described includes step：S100, collection inspection
Survey each WAP in region and receive the signal strength data that client to be detected is signaled；S200, by the signal
Intensity data is input into the first deep neural network, and it is in place to be calculated client institute to be detected through first deep neural network
Put the floor label of affiliated floor；S300, by the signal strength data input and the affiliated building in client position to be detected
One-to-one second deep neural network of floor label of layer；Visitor to be detected is calculated through second deep neural network
The predicted position coordinate of family end position.
It is further preferred that also including step before the step S100：S010, training in advance classifying type depth nerve net
Network, using training after the classifying type deep neural network as first deep neural network；S020, training in advance are multiple
Fitting type deep neural network corresponding with floor label respectively, by training after the fitting type deep neural network in one
It is individual as second deep neural network.
It is further preferred that the step S010 trains first deep neural network to further include step：S011、
Multiple floors in detection zone are classified, is the corresponding floor label of each floor distribution；S012, each is gathered respectively
Each WAP receives the signal strength data of the signal for training terminal to be sent out on default training position in floor, and combines
Corresponding floor label forms the training data sample of each floor, and the training data sample generation first of all floors is trained
Data set is input into classifying type deep neural network；S013, the training sample data of each floor are input into the classifying type successively
Deep neural network, corresponding training result is exported by the deep neural network；S014, the training knot that will be exported successively
The floor label of the really corresponding described default affiliated floor in training position is compared, according to comparative result to classification moldeed depth
Degree neutral net is trained, using the classifying type deep neural network after training as first deep neural network；It is described
Second deep neural network is trained to further include step in step S020：S021, respectively each floor set up one
With floor label fitting type deep neural network correspondingly；S022, plane right-angle coordinate is set up respectively in each floor,
The training position coordinates for training is marked in the plane right-angle coordinate；S023, respectively by the training in each floor
Position coordinates and each WAP receive the letter that training terminal is sent out on each corresponding position of training position coordinates
Number signal strength data as one group of training sample data, generate the second training dataset of each floor, and send into it is every
In the corresponding fitting class deep neural network of the floor label of individual floor；S024, successively by each group of training of each floor
Signal strength data in sample data is input into the fitting type deep neural network corresponding with floor label, by the plan
Mould assembly deep neural network exports corresponding training result；S025, the corresponding training of training result that will be exported successively
Position coordinates is compared, and fitting type deep neural network is trained according to comparative result, by the fitting moldeed depth after training
Degree neutral net is used as second deep neural network.
It is further preferred that the step S025 further includes step：The training that will be exported respectively according to the following equation
The corresponding training position coordinates of result is compared：
Wherein, σ represents the error of the corresponding default training position coordinates of the training result of output；(X1,Y1) represent every
The training result of secondary training output；(X2,Y2) represent corresponding training position coordinates；According to each group of instruction that training data is concentrated
Practice sample data to be trained the fitting type deep neural network so that training is exported every time training result and the instruction
Practice the error σ after position coordinates is compared to converge in preset range.
Invention additionally discloses a kind of many floor location servers based on wifi, including：Data acquisition module, for gathering
Each WAP receives the signal strength data that client to be detected is signaled in detection zone；Sort module, is used for
The signal strength data is input into the first deep neural network, visitor to be detected is calculated through first deep neural network
The floor classification of the affiliated floor in family end position；Locating module, for the signal strength data to be input into and visitor to be detected
One-to-one second deep neural network of floor label of the affiliated floor in family end position；Through the second depth nerve net
Network is calculated the predicted position coordinate of client position to be detected.
It is further preferred that also including：First training module, for training in advance classifying type deep neural network, will instruct
The classifying type deep neural network after white silk is used as first deep neural network；Second training module, for instruction in advance
Practice multiple fitting type deep neural networks corresponding with floor label respectively, by training after the fitting type deep neural network
In one as second deep neural network.
It is further preferred that first training module is further included：Floor classification submodule, for detection zone
Interior multiple floors are classified, and are the corresponding floor label of each floor distribution；First training dataset generates submodule, uses
In the signal for gathering the signal for training terminal to be sent out on the default training position of each WAP reception in each floor respectively
Intensity data, and the training sample data of each floor are formed with reference to corresponding floor label, by the training sample of all floors
The training dataset of data genaration first is input into classifying type deep neural network；First training prediction submodule, for successively will be every
The training sample data of individual floor are input into the classifying type deep neural network, export corresponding by the deep neural network
Training result, the floor label of the corresponding described default affiliated floor in training position of the training result of output is entered successively
Row compares, and classifying type deep neural network is trained according to comparative result, by the classifying type deep neural network after training
As first deep neural network；Second training module is further included：Network setting up submodule, for being respectively
Each floor sets up one with floor label fitting type deep neural network correspondingly；Establishment of coordinate system submodule, every
Individual floor sets up plane right-angle coordinate respectively, the training position for training is marked in the plane right-angle coordinate and is sat
Mark；Second training dataset generate submodule, for respectively by the training position coordinates in each floor and each wirelessly connect
Access point receives the signal strength data of the signal that training terminal is sent out on each corresponding position of training position coordinates as one
Group training sample data, generate the second training dataset of each floor, and send into corresponding with the floor label of each floor
Fitting class deep neural network in；Second training prediction submodule, for successively by each group of training sample of each floor
Signal strength data in data is input into the fitting type deep neural network corresponding with floor label, by the fitting type
Deep neural network exports corresponding training result, successively enters the corresponding training position coordinates of the training result of output
Row compares, and fitting type deep neural network is trained according to comparative result, by the fitting type deep neural network after training
As second deep neural network.
It is further preferred that it is described second training prediction submodule according to the following equation respectively will export training result with
Its corresponding training position coordinates is compared：
Wherein, σ represents the error of the corresponding training position coordinates of the training result of output；(X1,Y1) represent instruction every time
Practice the training result of output；(X2,Y2) represent corresponding training position coordinates；According to each group of training sample that training data is concentrated
Notebook data is trained to the fitting type deep neural network so that training is exported every time training result and the training position
The error σ after coordinate is compared is put to converge in preset range.
Compared with prior art, the present invention is provided many floor location methods based on wifi and server, by collecting
The first deep neural network that the signal strength data input corresponding with each WAP of client to be measured is trained
With the second deep neural network, you can determine the affiliated floor of client to be measured and position, by using containing a large amount of training samples
The training data set pair deep neural network training of data, not only lifts the lifting of positioning precision, while can not influence fixed
The accuracy of positioning result is lifted in the case of bit rate, successfully orientation problem is dissolved into the background of big data, and effectively
The advantage using big data improve the performance of real-time location-server.
Brief description of the drawings
Below by clearly understandable mode, preferred embodiment is described with reference to the drawings, the present invention is given furtherly
It is bright.
Fig. 1 is a kind of key step schematic diagram of many floor location methods based on wifi of the present invention；
Fig. 2 is that a kind of first deep neural network training step of many floor location methods based on wifi of the present invention is illustrated
Figure；
Fig. 3 is that a kind of second deep neural network training step of many floor location methods based on wifi of the present invention is illustrated
Figure；
Fig. 4 is a kind of main composition schematic diagram of many floor location servers based on wifi of the present invention；
Fig. 5 is that a kind of present invention many floor location servers based on wifi are fully composed schematic diagram.
Reference：
100th, data acquisition module, 200, sort module, 300, locating module, the 410, first training module, 411, floor
Classification submodule, the 412, first training dataset generation submodule, the 413, first training prediction submodule, the 420, second training mould
Block, 421, network setting up submodule, 422, establishment of coordinate system submodule, the 423, second training dataset generation submodule, 424,
Second training prediction submodule.
Specific embodiment
In order to illustrate more clearly about the embodiment of the present invention or technical scheme of the prior art, control is illustrated below
Specific embodiment of the invention.It should be evident that drawings in the following description are only some embodiments of the present invention, for
For those of ordinary skill in the art, on the premise of not paying creative work, other can also be obtained according to these accompanying drawings
Accompanying drawing, and obtain other implementation methods.
To make simplified form, part related to the present invention is only schematically show in each figure, they are not represented
Its as product practical structures.In addition, so that simplified form is readily appreciated, there is identical structure or function in some figures
Part, only symbolically depicts one of those, or has only marked one of those.Herein, " one " is not only represented
" only this ", it is also possible to represent the situation of " more than one ".
Fig. 1 is a kind of key step schematic diagram of many floor location methods based on wifi of the present invention, as shown in figure 1, one
The many floor location methods based on wifi are planted, methods described includes step：Each WAP in S100, acquisition testing region
Receive the signal strength data that client to be detected is signaled；S200, by the signal strength data be input into the first depth god
Through network, the floor label of the affiliated floor in client position to be detected is calculated through first deep neural network；
S300, by the signal strength data be input into it is one-to-one with the floor label of the affiliated floor in client position to be detected
Second deep neural network；The predicted position of client position to be detected is calculated through second deep neural network
Coordinate.
Specifically, above-mentioned client to be detected (hereinafter referred to as STA) is with smart mobile phone, notebook computer or personal flat board
The intelligent terminals such as computer are carrier.
STA sends detection frame in real time within a detection region, and WAP obtains detection frame signal after receiving is strong
Degree, each WAP reports signal intensity to home server or Cloud Server, and server is according to each WAP
The RSSI field intensity message generation signal strength data for reporting.The present invention is wireless with each by obtain each WAP
The corresponding signal strength data of access point as input location model initial data, for example, the form of signal strength data is<
RSSI1, RSSI2, RSSI3, RSSI4, RSSI5>, the RSSI of the STA that wherein RSSI1 is received for AP1, RSSI2 receives for AP2
The RSSI of STA, by that analogy.
Specifically, the first deep neural network is classifying type deep neural network in the present invention, by the signal intensity number
According to the first deep neural network is input into, the probable value of floor label is exported after being computed, choose the maximum floor label of probable value
It is determined that the current affiliated floor in client position to be detected.The second deep neural network is fitting moldeed depth degree nerve in the present invention
Network, the second deep neural network is input into by the signal strength data, and the prediction bits of client to be detected are exported after being computed
Put coordinate.
Preferably, step is also included before the step S100：S010, training in advance classifying type deep neural network, will
The classifying type deep neural network after training is used as first deep neural network；S020, training in advance are multiple respectively
Fitting type deep neural network corresponding with floor label, by training after the fitting type deep neural network in one work
It is second deep neural network.
Classifying type deep neural network is trained in the present invention, wherein the training network of classifying type deep neural network
Structure it is as follows：
Data Layer->Convolutional layer 1->Convolutional layer 2->ReLU layers->Max Pooling layers->Full articulamentum 1->Full articulamentum
2->SoftMaxLoss layers
After network parameter training is completed, the network parameter of classifying type deep neural network is updated to the net after training
Network parameter, while by last layer of SoftMaxLoss layers more SoftMax layers of training network, being formed and implementing network, for conduct
Location model participates in actual location process.Wherein SoftMaxLoss layers exports training when being trained for deep neural network
Output result with reality training location tags error, and SoftMax layer be used for implement network position when, output is treated
Detect the probable value of the affiliated floor label in client position.
Training network and implement network except last layer it is different, other layers all, the net obtained by training network
Network parameter can be used directly in implementation network.
Fitting type deep neural network is trained in the present invention, wherein the training network of fitting type deep neural network
Structure it is as follows：
Data Layer->Full articulamentum 1->ReLU layers->Full articulamentum 2->Euclidean Loss layers
After network parameter training is completed, the network parameter of fitting type deep neural network is updated to the net after training
Network parameter, while by last layer of Euclidean Loss layers of removal of training network, being formed and implementing network, for as positioning mould
Type participates in actual location process.Wherein Euclidean Loss layers exports the defeated of training when being trained for deep neural network
Go out the error of result and the training location tags of reality, and implement network in positioning, directly export visitor to be detected in Internet
The predicted position coordinate of family end position.
Training network and implement network except last layer it is different (training network be Euclidean Loss layer, implementation
Network removes Loss layers of Euclidean), all, the network parameter obtained by training network can be used directly in reality to other layers
In applying network.
Fig. 2 is that a kind of first deep neural network training step of many floor location methods based on wifi of the present invention is illustrated
Figure.Preferably, as shown in Fig. 2 the step S010 trains first deep neural network to further include step：S011、
Multiple floors in detection zone are classified, is the corresponding floor label of each floor distribution；S012, each is gathered respectively
Each WAP receives the signal strength data of the signal for training terminal to be sent out on default training position in floor, and combines
Corresponding floor label forms the training data sample of each floor, and the training data sample generation first of all floors is trained
Data set is input into classifying type deep neural network；S013, the training sample data of each floor are input into the classifying type successively
Deep neural network, corresponding training result is exported by the deep neural network；S014, the training knot that will be exported successively
The floor label of the really corresponding described default affiliated floor in training position is compared, according to comparative result to classification moldeed depth
Degree neutral net is trained, using the classifying type deep neural network after training as first deep neural network.
The process that the present invention is trained to the first deep neural network is introduced with instantiation below.
1st, first it is that the floor in detection zone is classified.
Floor in detection zone is classified, each floor distribution floor label, such as certain Administrative Area is followed successively by
One has 6 floors, therefore defines 6 classes, and is one corresponding floor label of each floor distribution.
2nd, training sample data collection
Gather the letter that each WAP in each floor receives the signal for training terminal to be sent out on default training position
Number intensity data, the default training position of each floor is user's self-defining in the present invention, can root when being applied to actual location
It is configured according to WIFI precision, default training position can be multiple.During floor label is 1 floor in above-mentioned 6 floors
Default training position on send detection frame by training terminal, WAP obtains detection frame signal after receiving is strong
Degree, each WAP reports signal intensity to home server or Cloud Server.Gather each by local or Cloud Server
The RSSI of individual AP obtains signal strength data corresponding with each AP, meanwhile, by the default training in floor that floor label is 1
Position binding signal intensity data generates one group of training sample data, it is assumed that have 4 AP in detection zone, then one group of training sample
Data are expressed as：(RSSI1, RSSI2, RSSI3, RSSI4,1), is expressed as with example<- 30, -12, -14, -67,1>, wherein,
The unit of preceding four numerical value can be dBm, and last numerical value can be with dimensionless.The training sample data can represent that numbering is
The 1 corresponding RSSI of access point is -30dBm, numbering be 2 the corresponding RSSI of access point for -12dBm, numbering is 3 access point
Corresponding RSSI is -14dBm, numbering be 4 the corresponding RSSI of access point for -67dBm, the STA is located at the building that floor label is 1
In layer.
3rd, the signal strength data in training sample data is input into the first deep neural network to be calculated, is finally exported
Training result and the error for training location tags.
RSSI1 to RSSI4 in training sample data is input into from input data layer, label is used at SoftMaxLoss layers
Returned, Loss is exported at SoftMaxLoss layers by training.
4th, cause that the Loss i.e. error of whole network is minimum finally by the parameter in percentage regulation neutral net, will classify
Type deep neural network is used as the first deep neural network.
It should be noted that do not indicate design parameter in entire depth neutral net, because these parameters and specific
The number of space and AP is relevant, not in the range of this patent.
Fig. 3 is that a kind of second deep neural network training step of many floor location methods based on wifi of the present invention is illustrated
Figure.Preferably, as shown in figure 3, training second deep neural network to further include step in the step S020：
S021, it is respectively each floor and sets up one with floor label fitting type deep neural network correspondingly；S022, at each
Floor sets up plane right-angle coordinate respectively, and the training position coordinates for training is marked in the plane right-angle coordinate；
S023, the training position coordinates in each floor and each WAP received into training terminal in each training position respectively
The signal strength data of the signal sent out on the corresponding position of coordinate is put as one group of training sample data, each floor is generated
In second training dataset, and the feeding fitting class deep neural network corresponding with the floor label of each floor；S024, according to
Signal strength data in the secondary each group of training sample data by each floor is input into the fitting corresponding with floor label
Type deep neural network, corresponding training result is exported by the fitting type deep neural network；S025, successively will output
The corresponding training position coordinates of training result be compared, fitting type deep neural network is carried out according to comparative result
Training, using the fitting type deep neural network after training as second deep neural network.
Preferably, the step S025 further includes step：According to the following equation respectively will export training result with
Its corresponding training position coordinates is compared：
Wherein, σ represents the error of the corresponding default training position coordinates of the training result of output；(X1,Y1) represent every
The training result of secondary training output；(X2,Y2) represent corresponding training position coordinates；According to each group of instruction that training data is concentrated
Practice sample data to be trained the fitting type deep neural network so that training is exported every time training result and the instruction
Practice the error σ after position coordinates is compared to converge in preset range.
The process that the present invention is trained to the second deep neural network is introduced with instantiation below.
1st, first it is that a fitting moldeed depth corresponding with the floor label of floor is set up to each floor in detection zone
Degree neutral net, gathers training sample data respectively in each floor.
So that the fitting type deep neural network set up to a floor is trained as an example：Flat square is set up in floor
Coordinate system, marks the training position coordinates for training in the plane right-angle coordinate, and the unit length of X-axis and Y-axis sets
It is preset value.Such as floor inner space is a length direction, it is assumed that a length of M, a width of N, area is M*N.According to the essence of WIFI
Degree characteristic determines the lower left corner for origin using 3 meters as X-axis and the unit length of Y-axis, then the unit scales of X-axis are 3 meter one single
Position, maximum scale is M/3, and the unit scales of Y-axis are 3 meter of one unit, and maximum scale is N/3.Establishing coordinate system successively
Detection zone marks the default training position coordinates for training, such as label=<1.4,5.3>, represent this position in the building
Layer in coordinate be：X=1.4, Y=5.3.
Trained in above-mentioned floor and send detection frame, WAP by training terminal on the corresponding position of position coordinates
The signal intensity of the detection frame is obtained after receiving, each WAP reports signal intensity to home server or cloud service
Device.The RSSI field intensity message for gathering each AP by local or Cloud Server obtains signal strength data corresponding with each AP,
Meanwhile, default training position coordinates binding signal intensity data is generated into one group of training sample data, it is assumed that have 4 in detection zone
Individual AP, then one group of training sample data be expressed as：(RSSI1, RSSI2, RSSI3, RSSI4, label), is expressed as with example<-
30, -12, -14, -67,1.4,5.3>, wherein, the unit of preceding four numerical value can be dBm, and last numerical value can be with immeasurable
Guiding principle.The training sample data can represent that the corresponding RSSI of access point that numbering is 1 is -30dBm, and numbering is 2 access point pair
The RSSI for answering is -12dBm, numbering be 3 the corresponding RSSI of access point for -14dBm, numbering is the 4 corresponding RSSI of access point
It is -67dBm, coordinate is X=1.4, the position of Y=5.3 during the STA is located at floor.
As a example by floor trains corresponding fitting type deep neural network as described above, successively to the corresponding plan of each floor
Mould assembly deep neural network is trained, and obtains multiple fitting type deep neural networks corresponding with floor label.
2nd, successively by the signal strength data input in each group of training sample data of each floor and floor label pair
The fitting type deep neural network answered, corresponding training result is exported with instruction by the fitting type deep neural network
Practice the error of location tags.
RSSI1 to RSSI4 in training sample data is input into from data Layer, label is used at Euclidean Loss layers
Returned, Loss is exported at Euclidean Loss layers by training.
3rd, cause that the Loss i.e. error of whole network is minimum finally by the parameter in percentage regulation neutral net, respectively will
Multiple fitting type deep neural networks after training are used as the second deep neural network.
It should be noted that do not indicate design parameter in entire depth neutral net, because these parameters and specific
The number of space and AP is relevant, not in the range of this patent.
The first deep neural network and the second deep neural network are based on to client institute to be detected with example introduction below
It is predicted in position, it is specific as follows：
STA to be detected is placed in any one position in detection zone.The outside broadcast probe request frames of the STA, detection zone
Each AP in domain obtains the signal intensity of the detection frame after receiving, each WAP reports signal intensity to local clothes
Business device or Cloud Server, the RSSI that server is reported according to all AP obtain signal strength data corresponding with each AP, with reference to
Examples detailed above is represented by, (RSSI1, RSSI2, RSSI3, RSSI4).
The RSSI1 of the signal strength data that the STA positions to be predicted training terminal that each AP is received is signaled is arrived
RSSI4 is input into from the data Layer of the first deep neural network, and the prediction probability of 6 class floor labels is exported at SoftMax layers.
Assuming that the probable value of 6 class floor labels of output is arranged as from big to small：
Label 1：0.7
Label 2：0.2
Label 3：0.1
Label 4：0.04
Labe l5：0.01
…
…
Label1 is then chosen as last prediction, that is, predicts this STA in the floor that floor label is 1.
Then RSSI1 to the RSSI4 of signal strength data is input into from the data Layer of the second deep neural network,
Euclidean layers exports predicted position coordinates of the STA in floor.
Fig. 4 is a kind of main composition schematic diagram of many floor location servers based on wifi of the present invention, as shown in figure 4,
A kind of many floor location servers based on wifi, including：Data acquisition module 100, for each nothing in acquisition testing region
Line access point receives the signal strength data that client to be detected is signaled；Sort module 200, for by the signal intensity
The deep neural network of data input first, client position institute to be detected is calculated through first deep neural network
Belong to the floor classification of floor；Locating module 300, for the signal strength data to be input into and client position to be detected
One-to-one second deep neural network of floor label of affiliated floor；It is calculated through second deep neural network and is treated
The predicted position coordinate of detection client position.
Specifically, above-mentioned client to be detected (hereinafter referred to as STA) is with smart mobile phone, notebook computer or personal flat board
The intelligent terminals such as computer are carrier.
STA sends detection frame in real time within a detection region, and WAP obtains detection frame signal after receiving is strong
Degree, each WAP reports signal intensity to home server or Cloud Server, and server is according to each WAP
The RSSI field intensity message generation signal strength data for reporting.The present invention is wireless with each by obtain each WAP
The corresponding signal strength data of access point as input location model initial data, for example, the form of signal strength data is<
RSSI1, RSSI2, RSSI3, RSSI4, RSSI5>, the RSSI of the STA that wherein RSSI1 is received for AP1, RSSI2 receives for AP2
The RSSI of STA, by that analogy.
Specifically, the first deep neural network is classifying type deep neural network in the present invention, by the signal intensity number
According to the first deep neural network is input into, the probable value of floor label is exported after being computed, choose the maximum floor label of probable value
It is determined that the current affiliated floor in client position to be detected.The second deep neural network is fitting moldeed depth degree nerve in the present invention
Network, the second deep neural network is input into by the signal strength data, and the prediction bits of client to be detected are exported after being computed
Put coordinate.
Fig. 5 is that a kind of present invention many floor location servers based on wifi are fully composed schematic diagram.Preferably, also wrap
Include：First training module 410, for training in advance classifying type deep neural network, by training after classifying type depth god
Through network as first deep neural network；Second training module 420, for training in advance it is multiple respectively with floor label
Corresponding fitting type deep neural network, using training after the fitting type deep neural network in one as described second
Deep neural network.
Classifying type deep neural network is trained in the present invention, wherein the training network of classifying type deep neural network
Structure it is as follows：
Data Layer->Convolutional layer 1->Convolutional layer 2->ReLU layers->Max Pooling layers->Full articulamentum 1->Full articulamentum
2->SoftMaxLoss layers
After network parameter training is completed, the network parameter of classifying type deep neural network is updated to the net after training
Network parameter, while by last layer of SoftMaxLoss layers more SoftMax layers of training network, being formed and implementing network, for conduct
Location model participates in actual location process.Wherein SoftMaxLoss layers exports training when being trained for deep neural network
Output result with reality training location tags error, and SoftMax layer be used for implement network position when, output is treated
Detect the probable value of the affiliated floor label in client position.
Training network and implement network except last layer it is different (training network be SoftMaxLoss layer, implementation network
For), all, the network parameter obtained by training network can be used directly in implementation network other layers.
Fitting type deep neural network is trained in the present invention, wherein the training network of fitting type deep neural network
Structure it is as follows：
Data Layer->Full articulamentum 1->ReLU layers->Full articulamentum 2->Euclidean Loss layers
After network parameter training is completed, the network parameter of fitting type deep neural network is updated to the net after training
Network parameter, while by last layer of Euclidean Loss layers of removal of training network, being formed and implementing network, for as positioning mould
Type participates in actual location process.Wherein Euclidean Loss layers exports the defeated of training when being trained for deep neural network
Go out the error of result and the training location tags of reality, and implement network in positioning, directly export visitor to be detected in Internet
The predicted position coordinate of family end position.
Training network and implement network except last layer it is different (training network be Euclidean Loss layer, implementation
Network removes Loss layers of Euclidean), all, the network parameter obtained by training network can be used directly in reality to other layers
In applying network
Preferably, first training module 410 is further included：Floor classification submodule 411, for detection zone
Interior multiple floors are classified, and are the corresponding floor label of each floor distribution；First training dataset generates submodule
412, receive the signal for training terminal to be sent out on default training position for gathering each WAP in each floor respectively
Signal strength data, and the training sample data of each floor are formed with reference to corresponding floor label, by the instruction of all floors
Practice sample data and generate the first training dataset input classifying type deep neural network；First training prediction submodule 413, is used for
The training sample data of each floor are input into the classifying type deep neural network successively, it is defeated by the deep neural network
Go out corresponding training result, successively by the building of the corresponding described default affiliated floor in training position of the training result for exporting
Layer label is compared, and classifying type deep neural network is trained according to comparative result, by the classifying type depth after training
Neutral net is used as first deep neural network；Second training module 420 is further included：Network setting up submodule
421, one is set up with floor label fitting type deep neural network correspondingly for being respectively each floor；Coordinate system is built
Vertical submodule 422, plane right-angle coordinate is set up in each floor respectively, is marked for instructing in the plane right-angle coordinate
Experienced training position coordinates；Second training dataset generates submodule 423, for respectively sitting the training position in each floor
It is marked with and each WAP receives training terminal and trains the letter of the signal sent out on the corresponding position of position coordinates at each
Number intensity data generates the second training dataset of each floor, and send into and each floor as one group of training sample data
The corresponding fitting class deep neural network of floor label in；Second training prediction submodule 424, for successively by each building
The signal strength data input fitting moldeed depth degree nerve corresponding with floor label in each group of training sample data of layer
Network, corresponding training result is exported by the fitting type deep neural network, successively by the training result for exporting and its
Corresponding training position coordinates is compared, and fitting type deep neural network is trained according to comparative result, after training
Fitting type deep neural network as second deep neural network.
It should be noted that being referred to for the training process of the training module 420 of above-mentioned first training module 410 and second
The present invention trains the explanation of classifying type deep neural network and fitting type deep neural network for method part, no longer multiple herein
State.
Preferably, the training result that the second training prediction module 420 will be exported respectively according to the following equation is right with it
The training position coordinates answered is compared：
Wherein, σ represents the error of the corresponding training position coordinates of the training result of output；(X1,Y1) represent instruction every time
Practice the training result of output；(X2,Y2) represent corresponding training position coordinates；According to each group of training sample that training data is concentrated
Notebook data is trained to the fitting type deep neural network so that training is exported every time training result and the training position
The error σ after coordinate is compared is put to converge in preset range.
The contents such as information exchange, implementation procedure in book server between each module and above method embodiment are based on same
Design, particular content can be found in the narration in the inventive method embodiment, and here is omitted.
It should be noted that above-described embodiment can independent assortment as needed.The above is only of the invention preferred
Implementation method, it is noted that for those skilled in the art, is not departing from the premise of the principle of the invention
Under, some improvements and modifications can also be made, these improvements and modifications also should be regarded as protection scope of the present invention.
Claims (8)
1. a kind of many floor location methods based on wifi, it is characterised in that methods described includes step：
Each WAP receives the signal strength data that client to be detected is signaled in S100, acquisition testing region；
S200, by the signal strength data be input into the first deep neural network, calculated through first deep neural network
To the floor label of the affiliated floor in client position to be detected；
S300, a pair of the floor label 1 by signal strength data input and the affiliated floor in client position to be detected
The second deep neural network answered；The prediction of client position to be detected is calculated through second deep neural network
Position coordinates.
2. many floor location methods of wifi are based on as claimed in claim 1, it is characterised in that before the step S100 also
Including step：
S010, training in advance classifying type deep neural network, using training after the classifying type deep neural network as described
First deep neural network；
The multiple fitting type deep neural networks corresponding with floor label respectively of S020, training in advance, by training after the plan
One in mould assembly deep neural network is used as second deep neural network.
3. many floor location methods of wifi are based on as claimed in claim 2, it is characterised in that the step S010 trains institute
State the first deep neural network and further include step：
S011, the multiple floors in detection zone are classified, be the corresponding floor label of each floor distribution；
S012, each WAP in each floor is gathered respectively receive the signal for training terminal to be sent out on default training position
Signal strength data, and the training sample data of each floor are formed with reference to corresponding floor label, by the instruction of all floors
Practice sample data and generate the first training dataset input classifying type deep neural network；
S013, the training sample data of each floor are input into the classifying type deep neural network successively, by the depth
Neutral net exports corresponding training result；
S014, the floor label of the corresponding described default affiliated floor in training position of the training result of output carried out successively
Compare, classifying type deep neural network is trained according to comparative result, the classifying type deep neural network after training is made
It is first deep neural network；
Second deep neural network is trained to further include step in the step S020：
S021, it is respectively each floor and sets up one with floor label fitting type deep neural network correspondingly；
S022, plane right-angle coordinate is set up respectively in each floor, marked for training in the plane right-angle coordinate
Training position coordinates；
S023, the training position coordinates in each floor and each WAP are received into training terminal respectively instructed at each
Practice the signal strength data of the signal sent out on the corresponding position of position coordinates as one group of training sample data, generate each building
In second training dataset of layer, and the feeding fitting class deep neural network corresponding with the floor label of each floor；
S024, successively by each group of training sample data of each floor signal strength data input it is corresponding with floor label
The fitting type deep neural network, export corresponding training result by the fitting type deep neural network；
S025, the corresponding training position coordinates of the training result of output is compared successively, according to comparative result to intending
Mould assembly deep neural network is trained, using the fitting type deep neural network after training as the second depth nerve net
Network.
4. many floor location methods of wifi are based on as claimed in claim 3, it is characterised in that the step S025 is further
Including step：
The corresponding training position coordinates of the training result of output is compared respectively according to the following equation：
Wherein, σ represents the error of the corresponding default training position coordinates of the training result of output；(X1,Y1) represent instruction every time
Practice the training result of output；(X2,Y2) represent corresponding training position coordinates；
The fitting type deep neural network is trained according to each group of training sample data that training data is concentrated so that
Error σ after the training result of training output is compared with the training position coordinates every time is converged in preset range.
5. a kind of many floor location servers based on wifi, it is characterised in that including：
Data acquisition module, the letter that client to be detected is signaled is received for each WAP in acquisition testing region
Number intensity data；
Sort module, for the signal strength data to be input into the first deep neural network, through the first depth nerve net
Network is calculated the floor classification of the affiliated floor in client position to be detected；
Locating module, for the signal strength data to be input into the floor mark with the affiliated floor in client position to be detected
Sign one-to-one second deep neural network；It is in place client institute to be detected to be calculated through second deep neural network
The predicted position coordinate put.
6. many floor location servers of wifi are based on as claimed in claim 5, it is characterised in that also included：
First training module, for training in advance classifying type deep neural network, by training after classifying type depth nerve
Network is used as first deep neural network；
Second training module, for the multiple fitting type deep neural networks corresponding with floor label respectively of training in advance, will instruct
One in the fitting type deep neural network after white silk is used as second deep neural network.
7. many floor location servers of wifi are based on as claimed in claim 5, it is characterised in that first training module
Further include：
Floor classification submodule, is the corresponding building of each floor distribution for classifying to the multiple floors in detection zone
Layer label；
First training dataset generates submodule, and default training is received for gathering each WAP in each floor respectively
The signal strength data of the signal that training terminal is sent out on position, and the training of each floor is formed with reference to corresponding floor label
The training sample data of all floors are generated the first training dataset input classifying type deep neural network by sample data；
First training prediction submodule, for the training sample data of each floor to be input into the classifying type depth nerve successively
Network, corresponding training result is exported by the deep neural network, and the training result that will be exported successively is corresponding
The floor label of the default affiliated floor in training position is compared, and classifying type deep neural network is entered according to comparative result
Row training, using the classifying type deep neural network after training as first deep neural network；
Second training module is further included：
Network setting up submodule, for being respectively, each floor sets up one and floor label is fitted moldeed depth degree god correspondingly
Through network；
Establishment of coordinate system submodule, plane right-angle coordinate is set up in each floor respectively, in the plane right-angle coordinate
Mark the training position coordinates for training；
Second training dataset generate submodule, for respectively by the training position coordinates in each floor and each wirelessly connect
Access point receives the signal strength data of the signal that training terminal is sent out on each corresponding position of training position coordinates as one
Group training sample data, generate the second training dataset of each floor, and send into corresponding with the floor label of each floor
Fitting class deep neural network in；
Second training prediction submodule, for the signal strength data in each group of training sample data successively by each floor
The input fitting type deep neural network corresponding with floor label, it is relative by fitting type deep neural network output
, be compared for the corresponding training position coordinates of the training result of output successively, according to comparative result by the training result answered
Fitting type deep neural network is trained, using the fitting type deep neural network after training as second depth nerve
Network.
8. many floor location servers of wifi are based on as claimed in claim 7, it is characterised in that the second training prediction
Be compared for the corresponding training position coordinates of the training result of output respectively according to the following equation by submodule：
Wherein, σ represents the error of the corresponding training position coordinates of the training result of output；(X1,Y1) represent and train defeated every time
The training result for going out；(X2,Y2) represent corresponding training position coordinates；
The fitting type deep neural network is trained according to each group of training sample data that training data is concentrated so that
Error σ after the training result of training output is compared with the training position coordinates every time is converged in preset range.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611046269.9A CN106792553A (en) | 2016-11-22 | 2016-11-22 | A kind of many floor location methods and server based on wifi |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611046269.9A CN106792553A (en) | 2016-11-22 | 2016-11-22 | A kind of many floor location methods and server based on wifi |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106792553A true CN106792553A (en) | 2017-05-31 |
Family
ID=58975207
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201611046269.9A Pending CN106792553A (en) | 2016-11-22 | 2016-11-22 | A kind of many floor location methods and server based on wifi |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106792553A (en) |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107368926A (en) * | 2017-07-28 | 2017-11-21 | 中南大学 | A kind of how natural parameter sensing method for amalgamation processing of intelligent environment carrying robot identification floor |
| CN107403195A (en) * | 2017-07-28 | 2017-11-28 | 中南大学 | A kind of meteorologic parameter Intelligent Fusion processing method of carrying robot identification floor |
| WO2018095009A1 (en) * | 2016-11-22 | 2018-05-31 | 上海斐讯数据通信技术有限公司 | Multi-room positioning method based on wifi and server |
| CN109379713A (en) * | 2018-08-23 | 2019-02-22 | 南京邮电大学 | Floor prediction technique based on integrated extreme learning machine and principal component analysis |
| CN109640272A (en) * | 2018-12-24 | 2019-04-16 | 维沃移动通信有限公司 | A kind of localization method and mobile terminal |
| CN109699002A (en) * | 2018-12-06 | 2019-04-30 | 深圳市中电数通智慧安全科技股份有限公司 | A kind of interior WiFi localization method, device and terminal device |
| WO2019153600A1 (en) * | 2018-02-07 | 2019-08-15 | 平安科技（深圳）有限公司 | Electronic apparatus, floor positioning method, and computer readable storage medium |
| CN111650554A (en) * | 2020-05-29 | 2020-09-11 | 浙江商汤科技开发有限公司 | Positioning method and device, electronic equipment and storage medium |
| CN111693938A (en) * | 2020-06-10 | 2020-09-22 | 北京云迹科技有限公司 | Floor positioning method and device of robot, robot and readable storage medium |
| CN114666734A (en) * | 2022-03-22 | 2022-06-24 | 广州市梦享网络技术有限公司 | Positioning method and device for deep learning radio frequency signal fingerprint identification algorithm |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101267374A (en) * | 2008-04-18 | 2008-09-17 | 清华大学 | 2.5D location method based on neural network and wireless LAN infrastructure |
| CN102791025A (en) * | 2011-05-20 | 2012-11-21 | 盛乐信息技术（上海）有限公司 | Wireless fidelity (WIFI) based layered positioning system and implementing method |
| CN103220777A (en) * | 2013-01-05 | 2013-07-24 | 东莞市力王电池有限公司 | Mobile device positioning system |
| CN103905992A (en) * | 2014-03-04 | 2014-07-02 | 华南理工大学 | Indoor positioning method based on wireless sensor networks of fingerprint data |
| CN103916821A (en) * | 2014-04-11 | 2014-07-09 | 北京工业大学 | Floor distinguishing method based on RSSI difference between floors |
-
2016
- 2016-11-22 CN CN201611046269.9A patent/CN106792553A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101267374A (en) * | 2008-04-18 | 2008-09-17 | 清华大学 | 2.5D location method based on neural network and wireless LAN infrastructure |
| CN102791025A (en) * | 2011-05-20 | 2012-11-21 | 盛乐信息技术（上海）有限公司 | Wireless fidelity (WIFI) based layered positioning system and implementing method |
| CN103220777A (en) * | 2013-01-05 | 2013-07-24 | 东莞市力王电池有限公司 | Mobile device positioning system |
| CN103905992A (en) * | 2014-03-04 | 2014-07-02 | 华南理工大学 | Indoor positioning method based on wireless sensor networks of fingerprint data |
| CN103916821A (en) * | 2014-04-11 | 2014-07-09 | 北京工业大学 | Floor distinguishing method based on RSSI difference between floors |
Cited By (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2018095009A1 (en) * | 2016-11-22 | 2018-05-31 | 上海斐讯数据通信技术有限公司 | Multi-room positioning method based on wifi and server |
| CN107403195A (en) * | 2017-07-28 | 2017-11-28 | 中南大学 | A kind of meteorologic parameter Intelligent Fusion processing method of carrying robot identification floor |
| CN107403195B (en) * | 2017-07-28 | 2018-03-27 | 中南大学 | A kind of meteorologic parameter Intelligent Fusion processing method of carrying robot identification floor |
| CN107368926B (en) * | 2017-07-28 | 2018-07-10 | 中南大学 | A kind of more natural parameter sensing method for amalgamation processing of intelligent environment carrying robot identification floor |
| CN107368926A (en) * | 2017-07-28 | 2017-11-21 | 中南大学 | A kind of how natural parameter sensing method for amalgamation processing of intelligent environment carrying robot identification floor |
| WO2019153600A1 (en) * | 2018-02-07 | 2019-08-15 | 平安科技（深圳）有限公司 | Electronic apparatus, floor positioning method, and computer readable storage medium |
| CN109379713B (en) * | 2018-08-23 | 2020-11-03 | 南京邮电大学 | Floor prediction method based on integrated extreme learning machine and principal component analysis |
| CN109379713A (en) * | 2018-08-23 | 2019-02-22 | 南京邮电大学 | Floor prediction technique based on integrated extreme learning machine and principal component analysis |
| CN109699002A (en) * | 2018-12-06 | 2019-04-30 | 深圳市中电数通智慧安全科技股份有限公司 | A kind of interior WiFi localization method, device and terminal device |
| CN109640272A (en) * | 2018-12-24 | 2019-04-16 | 维沃移动通信有限公司 | A kind of localization method and mobile terminal |
| CN109640272B (en) * | 2018-12-24 | 2021-06-29 | 维沃移动通信有限公司 | Positioning method and mobile terminal |
| CN111650554A (en) * | 2020-05-29 | 2020-09-11 | 浙江商汤科技开发有限公司 | Positioning method and device, electronic equipment and storage medium |
| CN111693938A (en) * | 2020-06-10 | 2020-09-22 | 北京云迹科技有限公司 | Floor positioning method and device of robot, robot and readable storage medium |
| CN114666734A (en) * | 2022-03-22 | 2022-06-24 | 广州市梦享网络技术有限公司 | Positioning method and device for deep learning radio frequency signal fingerprint identification algorithm |
| CN114666734B (en) * | 2022-03-22 | 2022-11-22 | 广州市梦享网络技术有限公司 | Positioning method and device for deep learning radio frequency signal fingerprint identification algorithm |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106792553A (en) | A kind of many floor location methods and server based on wifi | |
| CN106535134A (en) | Multi-room locating method based on wifi and server | |
| CN106793067A (en) | A kind of many floor indoor orientation methods and server based on joint network | |
| CN106535326A (en) | WiFi locating method based on depth neural network and server | |
| CN106507476A (en) | A kind of WiFi localization methods and server and location model construction method | |
| CN106792769A (en) | A kind of WiFi localization methods and server and location model method for building up | |
| CN103901398B (en) | A kind of location fingerprint localization method based on combination collating sort | |
| CN103686767B (en) | A kind of Downtilt method of adjustment and device based on LTE network | |
| CN103826281A (en) | Micropower wireless communication routing algorithm and networking method based on field intensity information | |
| CN108900333A (en) | A kind of appraisal procedure and assessment device of quality of wireless network | |
| CN101951617B (en) | Mobile network communication quality evaluation method based on analytic hierarchy process | |
| CN106792507A (en) | A kind of WiFi localization methods and server based on network data | |
| CN106507475A (en) | Room area WiFi localization methods and system based on EKNN | |
| CN108242946A (en) | A kind of coal mine down-hole tunnel object localization method based on MIMO-OFDM technologies | |
| CN107809766B (en) | Method and device for generating machine learning sample for network optimization | |
| CN106327016B (en) | The special car intelligent optimization scheduling system of oil recovery factory's work production based on Internet of Things | |
| CN114641015B (en) | Network evaluation method, device, electronic equipment and storage medium | |
| CN105025525B (en) | A kind of channel loading equilibrium system and method for multichannel wireless local area networks | |
| CN107170065A (en) | Intelligent movable Work attendance method, device and system | |
| CN106792506A (en) | A kind of WiFi localization methods and server | |
| CN114374449A (en) | Interference source determination method, device, equipment and medium | |
| CN107545726A (en) | A kind of bus travel speed determines method and device | |
| CN104581949B (en) | Cell gridding method and device | |
| CN106488559A (en) | A kind of outdoor positioning method based on visibility and server | |
| CN106793066A (en) | A kind of wifi localization methods and server based on two-way signaling intensity data |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20170531 |