CN107037399A - A kind of Wi Fi indoor orientation methods based on deep learning - Google Patents
A kind of Wi Fi indoor orientation methods based on deep learning Download PDFInfo
- Publication number
- CN107037399A CN107037399A CN201710323523.3A CN201710323523A CN107037399A CN 107037399 A CN107037399 A CN 107037399A CN 201710323523 A CN201710323523 A CN 201710323523A CN 107037399 A CN107037399 A CN 107037399A
- Authority
- CN
- China
- Prior art keywords
- word bank
- deep learning
- line
- fingerprint base
- cluster heart
- 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
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S5/00—Position-fixing by co-ordinating two or more direction or position line determinations; Position-fixing by co-ordinating two or more distance determinations
- G01S5/02—Position-fixing by co-ordinating two or more direction or position line determinations; Position-fixing by co-ordinating two or more distance determinations using radio waves
- G01S5/0278—Position-fixing by co-ordinating two or more direction or position line determinations; Position-fixing by co-ordinating two or more distance determinations using radio waves involving statistical or probabilistic considerations
-
- 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
Landscapes
- Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Biophysics (AREA)
- Evolutionary Computation (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Radar, Positioning & Navigation (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Remote Sensing (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Probability & Statistics with Applications (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The present invention discloses a kind of Wi Fi indoor orientation methods based on deep learning, purpose is to solve the problem of people position in the indoor environments such as the indoor garage of large scale business synthesis, the present invention uses autocoding algorithm, and from RSS big datas, depth characteristic is extracted non-supervisoryly；Then it is used to set up fingerprint base using these depth characteristics, fingerprint base is divided into some regions according to the position of initial data, the optimal neighbouring method (K nearest neighbors, KNN) finally more commonly used with matching algorithm treats position location and carries out matching positioning.
Description
Technical field
The present invention relates to indoor positioning technologies field, large scale business synthesis indoor positioning can be applied to.
Background technology
It is largely directly different wireless access will to be determined at reference point in existing Wi-Fi indoor orientation methods
The received signal strength (RSS) of point (APs) is used to set up fingerprint base or set up after traditional principal component analysis (PCA) processing
Fingerprint base, then divides to reduce follow-up match complexity, is eventually used for of point to be determined according to a certain method to fingerprint base
With positioning.The deficiency of presence：Original RSS data is with many noises and dimension is high directly with to follow-up matching positioning
Precision influence it is very big and computation complexity is very high, and the process of feature selecting is carried out to original RSS data with traditional PCA
In, although data are served with the effect of noise reduction dimensionality reduction.But because artificial subjective factor is present, during Feature Selection,
Profound data characteristics between some initial data is possible to be filtered or lost, so as to influence the essence that back location is matched
Degree and reliability, cause position location not accurate enough, and PCA is only relatively good to linear data effect, for interior such one
Under the conditions of individual complex environment, due to the influence of shelter or wall etc. non-linear, traditional PCA is presented and uncomfortable in RSS data
Close.
The content of the invention
It is an object of the invention to provide a kind of Wi-Fi indoor orientation methods based on deep learning, to solve prior art
Middle the problem of.
To realize that the technical scheme that the object of the invention is used is such：Using the automatic volume in deep learning algorithm
Code, can be directly from original training data learning depth characteristic, and is fed back and adjusted above by back-propagation algorithm
Network parameter, the effect of dimensionality reduction has not been only reached, and ensure the characteristic validity that learns, it is to avoid some effective informations
Lose.Then fingerprint base is set up with the feature learnt, after region division, for location matches.
Specifically, a kind of Wi-Fi indoor orientation methods based on deep learning, it is characterised in that:
Set up fingerprint base：
1) sample RSS data is gathered.
1-1) n Wi-Fi WAP (APs) is disposed in the room area for needing to position.
The room area 1-2) is divided into m domain block, m is natural number.
Using the center of each domain block as sampled point Pk, determination step 1-1) deployment each APs received signal strength
RSS, each sampled point PkA position coordinates is all corresponded to, the corresponding n RSS data of a position coordinates forms multidimensional after arranging
Set of data samples R a line.K=1,2 ... m.
2) to data prediction, completion, normalization.
2-1) by step 1) obtain multidimensional data sample set R carry out missing zero padding.
2-1) normalized, obtains training set H.
3) depth characteristic learns.
The structure of own coding neutral net 3-1) is determined, sets that its is a total of q+2 layers, wherein there is 1 input layer, q hidden
Hide layer and 1 output layer.Q is natural number.
3-2) using step 2) obtained by multidimensional data sample set H be used as input layer.
The initial weight W of this layer 3-3) is set, is trained using autocoding algorithm, training result C is obtained(1)。
3-4) by training result C(1)As its high one layer of input, and previous step is repeated, obtain training result C(2)。
3-5) repeat step 3-3), 3-4) q times, obtain q+1 layers of output result C(q), training terminates.
4) fingerprint base is obtained.
The feature set C extracted with deep learning(q)Fingerprint base is set up, fingerprint base is divided into t sub- fingerprints according to region
Storehouse, several neighbouring sample points of each word bank correspondence, a line one sampled point of correspondence of each word bank, each word bank has a cluster
Heart Qi, wherein, i=1,2 ..., t, all cluster heart composition set Q.
Indoor positioning：
A) it is in a certain position P to be positioned of room area0When, collection and step 1) in matrix R often capable identical structure
One row vector of the RSS data of feature, and be saved in row vector V.
B) use and step 2) identical method, vectorial V denoisings normalized is obtained into vectorial U.
C) setting and step 3) in identical structure and the own coding neutral net of weights, and using the matrix U after processing as
The input of own coding neutral net, obtains characteristic E(q)。
D) by obtained characteristic E(q)It is updated in matching algorithm, with step 4) the cluster heart in the set Q carries out
Match somebody with somebody.
Matching obtains a cluster heart QjWherein, j ∈ { 1,2 ..., t },
Calculate characteristic E(q)With cluster heart QjThe distance of each row vector in corresponding word bank j, obtain v away from
From wherein v is cluster heart QjCorresponding word bank j line number, by v distance according to sorting from small to large, F value before taking is preserved
In matrixIn.In matrix d, each element correspondence word bank j a line, word bank j a line one sampled point of correspondence.
Calculate weight,U=1,2 ... F
Calculate point to be determined P0Position coordinates.
P0=α1P1+α2P2+…+αFPF
Wherein, P1,P2,…,PFFor the position coordinates of the word bank j corresponding sampled point of every a line.
Further, step 3) in, a depth characteristic containing multiple hidden layers is built based on autocoding algorithm and learnt
Model.
Further, step 4) described in region refer to all groups of samples in sub- fingerprint base into region, its division
Foundation is a square area centered on the cluster heart, and the big I in the region is adjusted according to actual map area size.
Further, during the indoor positioning, the matching algorithm described in step d) is KNN algorithms.
Further, methods described is applied to the positioning of parking stall in large-scale synthesis body room.The solution have the advantages that need not
Doubt, after the feature that deep learning extracts initial data, as a result substantially optimized, the precision of positioning is improved.
Illustrate that deep learning model can be from initial data focusing study to more effectively more abstract feature representation, deeper expression
The potential profound rule gone out between data characteristics.Therefore this method can be effectively used for parking lot in large scale business synthesis
Deng the indoor positioning (reverse car seeking) of environment, providing the user accurately positioning (reversely parking), offer one is more convenient and has
The method of effect.
In Fig. 5, the test data set that we employ same test point obtains respective fixed after three kinds of method processing
Position deviation accumulation probability, it can be seen that directly positioned with initial data, locating effect is excessively poor, interior can not be met at all
The requirement of positioning, and pass through deep learning method autocoding and learn to obtain the data after profound feature, for positioning, by mistake
Difference has reached 80% in 1~2m, and this is relative to data position error the improving nearly in 1~2m handled with PCA
30%, the precision of positioning is significantly improved, and the positioning result obtained by autocoding can meet present user to room
The requirement of reverse car seeking in the indoor environments such as interior parking lot.Should test result indicates that, by deep learning extract initial data
Feature after, as a result substantially optimized, the precision of positioning is improved.Illustrate that deep learning model can be from raw data set
Learning is to more effective more abstract feature representation, the deeper potential profound rule given expression between data characteristics.
Therefore this method can be effectively used in the reverse car seeking function in parking lot in large scale business synthesis, be customer reverse parking
One more convenient and effective method is provided.
Brief description of the drawings
Fig. 1 region divisions and Wi-Fi deployment.
Fig. 2 method overall structure figures.
Fig. 3 method flow diagrams.
Fig. 4 autocoding structural representations.
Fig. 5 test data position error accumulated probabilities compare；
Fig. 6 installs certain parking garage of camera to each parking stall.
Embodiment
With reference to embodiment, the invention will be further described, but should not be construed above-mentioned subject area of the invention only
It is limited to following embodiments.Without departing from the idea case in the present invention described above, according to ordinary skill knowledge and used
With means, various replacements and change are made, all should be included within the scope of the present invention.
Embodiment 1：
A kind of Wi-Fi indoor orientation methods based on deep learning:
Set up fingerprint base：
1) sample RSS data is gathered.
1-1) n Wi-Fi WAP (APs) is disposed in the room area for needing to position.
The room area 1-2) is divided into m domain block, m is natural number.
Using the center of each domain block as sampled point Pk, determination step 1-1) deployment each APs received signal strength
RSS, each sampled point PkA position coordinates is all corresponded to, the corresponding n RSS data of a position coordinates forms multidimensional after arranging
Set of data samples R a line.K=1,2 ... m.
2) to data prediction, completion, normalization.
2-1) by step 1) obtain multidimensional data sample set R carry out missing zero padding.
2-1) normalized, obtains training set H.
3) depth characteristic learns.
One depth characteristic learning model for containing multiple hidden layers, pre-training original number are built based on autocoding algorithm
According to comprising the following steps that：
The structure of own coding neutral net 3-1) is determined, sets that its is a total of q+2 layers, wherein there is 1 input layer, q hidden
Hide layer and 1 output layer.Q is natural number.
3-2) using step 2) obtained by multidimensional data sample set H be used as input layer.
The initial weight W of this layer 3-3) is set, is trained using autocoding algorithm, training result C is obtained(1)。
3-4) by training result C(1)As its high one layer of input, and previous step is repeated, obtain training result C(2)。
3-5) repeat step 3-3), 3-4) q times, obtain q+1 layers of output result C(q), training terminates.
4) fingerprint base is obtained.
The feature set C extracted with deep learning(q)Fingerprint base is set up, fingerprint base is divided into t sub- fingerprints according to region
Storehouse, several neighbouring sample points of each word bank correspondence, a line one sampled point of correspondence of each word bank, each word bank has a cluster
Heart Qi, wherein, i=1,2 ..., t, all cluster heart composition set Q.
Here region refer to all groups of samples in sub- fingerprint base into region, its partitioning standards be using the cluster heart as
One square area at center, the big I in the region is adjusted according to actual map area size, and the method for division is for example attached
Fig. 1, each big square is exactly a region here, and comprising 9 sampled point regions, the sampled point labeled as grey is cluster
The heart)
Indoor positioning：
A) it is in a certain position P to be positioned of room area0When, collection and step 1) in matrix R often capable identical structure
One row vector of the RSS data of feature, and be saved in row vector V.
B) use and step 2) identical method, vectorial V denoisings normalized is obtained into vectorial U.
C) setting and step 3) in identical structure and the own coding neutral net of weights, and using the matrix U after processing as
The input of own coding neutral net, obtains characteristic E(q)。
D) by obtained characteristic E(q)It is updated to matching algorithm KNN (optimal neighbouring method, K-nearest
Neighbors in), with step 4) the cluster heart in the set Q matched.
Matching obtains a cluster heart QjWherein, j ∈ { 1,2 ..., t },
Calculate characteristic E(q)With cluster heart QjThe distance of each row vector in corresponding word bank j, obtain v away from
From wherein v is cluster heart QjCorresponding word bank j line number, by v distance according to sorting from small to large, F value before taking is preserved
In matrixIn.In matrix d, each element correspondence word bank j a line, word bank j a line one sampled point of correspondence.
Calculate weight,U=1,2 ... F
Calculate point to be determined P0Position coordinates.
P0=α1P1+α2P2+…+αFPF
Wherein, Pl,P2,…,PFFor the position coordinates of the word bank j corresponding sampled point of every a line.
Embodiment 2：
The methods described of embodiment 1 is applied to the positioning of the parking stall of large-scale synthesis body indoor parking garage by the present embodiment.
What deserves to be explained is, for large-scale indoor parking garage, it is an insoluble problem all the time to find vehicle.
In the prior art (as shown in Figure 6), employ and carry out registration of vehicle in each parking stall installation camera and Vehicle License Plate Recognition System
Stand.Although the way solves above mentioned problem, but the cost that installation camera is spent is very high, and after camera
Phase maintenance cost and reliability are undesirable.
The present embodiment discloses a kind of method for finding storing cycle position:
Set up fingerprint base：
1) sample RSS data is gathered.
1-1) as shown in figure 1, n Wi-Fi WAP (APs) is disposed in garage indoors.
The room area 1-2) is divided into m domain block, m is natural number, one domain block of each parking stall correspondence.
Using the center of each domain block as sampled point Pk, determination step 1-1) deployment each APs received signal strength
RSS, each sampled point PkA position coordinates is all corresponded to, the corresponding n RSS data of a position coordinates forms multidimensional after arranging
Set of data samples R a line.K=1,2 ... m.
2) to data prediction, completion, normalization.
2-1) by step 1) obtain multidimensional data sample set R carry out missing zero padding.
2-1) normalized, obtains training set H.
3) depth characteristic learns.
One depth characteristic learning model for containing multiple hidden layers, pre-training original number are built based on autocoding algorithm
According to comprising the following steps that：
The structure of own coding neutral net 3-1) is determined, sets that its is a total of q+2 layers, wherein there is 1 input layer, q hidden
Hide layer and 1 output layer.Q is natural number.
3-2) using step 2) obtained by multidimensional data sample set H be used as input layer.
The initial weight W of this layer 3-3) is set, is trained using autocoding algorithm, training result C is obtained(1)。
3-4) by training result C(1)As its high one layer of input, and previous step is repeated, obtain training result C(2)。
3-5) repeat step 3-3), 3-4) q times, obtain q+1 layers of output result C(q), training terminates.
4) fingerprint base is obtained.
The feature set C extracted with deep learning(q)Fingerprint base is set up, fingerprint base is divided into t sub- fingerprints according to region
Storehouse, several neighbouring sample points of each word bank correspondence, a line one sampled point of correspondence of each word bank, each word bank has a cluster
Heart Qi, wherein, i=1,2 ..., t, all cluster heart composition set Q.
Here region refer to all groups of samples in sub- fingerprint base into region, its partitioning standards be using the cluster heart as
One square area at center, the big I in the region is adjusted according to actual map area size, and the method for division is for example attached
Fig. 1, each big square is exactly a region here, and comprising 9 sampled point regions, the sampled point labeled as grey is cluster
The heart)
Vehicle location：
A) when storing cycle is in some parking stall, the position of the parking stall is designated as P0When, collection with step 1) in matrix R
Often go identical architectural feature RSS data a row vector, and be saved in row vector V.The process can be used
The hand-held mobile terminal device (mobile phone app) of driver and conductor is completed.Afterwards, the data collected can be uploaded onto the server,
And carry out subsequent arithmetic.
B) use and step 2) identical method, vectorial V denoisings normalized is obtained into vectorial U.
C) setting and step 3) in identical structure and the own coding neutral net of weights, and using the matrix U after processing as
The input of own coding neutral net, obtains characteristic E(q)。
D) by obtained characteristic E(q)It is updated in matching algorithm KNN, with step 4) the cluster heart in the set Q enters
Row matching.
Matching obtains a cluster heart QjWherein, j ∈ { 1,2 ..., t },
Calculate characteristic E(q)With cluster heart QjThe distance of each row vector in corresponding word bank j, obtain v away from
From wherein v is cluster heart QjCorresponding word bank j line number, by v distance according to sorting from small to large, F value before taking is preserved
In matrixIn.In matrix d, each element correspondence word bank j a line, word bank j a line one sampled point of correspondence.
Calculate weight,U=1,2 ... F
Calculate point to be determined P0Position coordinates.
P0=α1P1+α2P2+…+αFPF
Wherein, Pl,P2,…,PFFor the position coordinates of the word bank j corresponding sampled point of every a line.Process is found in vehicle
In, pass through Query Location point P0Position coordinates, you can learn the position of storing cycle.
Claims (5)
1. a kind of Wi-Fi indoor orientation methods based on deep learning, it is characterised in that:
Set up fingerprint base：
1) the sample RSS data collection；
1-1) n Wi-Fi WAP (APs) is disposed in the room area for needing to position；
The room area 1-2) is divided into m domain block, m is natural number；
Using the center of each domain block as sampled point Pk, determination step 1-1) deployment each APs received signal strength RSS, each
Sampled point PkAll correspond to a position coordinates.The corresponding n RSS data of one position coordinates forms multidimensional data sample after arranging
Collect R a line；K=1,2 ... m；
2) to data prediction, completion, normalization；
2-1) by step 1) obtain multidimensional data sample set R carry out missing zero padding.
2-1) normalized, obtains training set H.
3) depth characteristic learns；
The structure of own coding neutral net 3-1) is determined, sets that its is a total of q+2 layers, wherein having 1 input layer, q hidden layer
With 1 output layer；Q is natural number；
3-2) using step 2) obtained by multidimensional data sample set H be used as input layer.
The initial weight W of this layer 3-3) is set, is trained using autocoding algorithm, training result C is obtained(1)。
3-4) by training result C(1)As its high one layer of input, and previous step is repeated, obtain training result C(2)。
3-5) repeat step 3-3), 3-4) q times, obtain q+1 layers of output result C(q), training terminates.
4) fingerprint base is obtained；
The feature set C extracted with deep learning(q)Fingerprint base is set up, fingerprint base is divided into t sub- fingerprint bases according to region, often
Several neighbouring sample points of individual word bank correspondence, a line one sampled point of correspondence of each word bank, each word bank has a cluster heart Qi,
Wherein, i=1,2 ..., t, all cluster heart composition set Q；
Indoor positioning：
A) it is in a certain position P to be positioned of room area0When, collection and step 1) in matrix R often capable identical architectural feature
One row vector of RSS data, and be saved in row vector V；
B) use and step 2) identical method, vectorial V denoisings normalized is obtained into vectorial U；
C) setting and step 3) in identical structure and the own coding neutral net of weights, and using the matrix U after processing as self-editing
The input of code neutral net, obtains characteristic E(q)；
D) by obtained characteristic E(q)It is updated in matching algorithm, with step 4) the cluster heart in the set Q matched；
Matching obtains a cluster heart QjWherein, j ∈ { 1,2 ..., t },
Calculate characteristic E(q)With cluster heart QjThe distance of each row vector in corresponding word bank j, obtains v distance, its
Middle v is cluster heart QjCorresponding word bank j line number, by v distance according to sorting from small to large, F value before taking is stored in matrixIn；In matrix d, each element correspondence word bank j a line, word bank j a line one sampled point of correspondence；
Calculate weight,U=1,2 ... F
Calculate point to be determined P0Position coordinates.
P0=α1P1+α2P2+…+αFPF
Wherein, Pl,P2,…,PFFor the position coordinates of the word bank j corresponding sampled point of every a line.
2. a kind of Wi-Fi indoor orientation methods based on deep learning according to claim 1, it is characterised in that：Step
3) in, a depth characteristic learning model for containing multiple hidden layers is built based on autocoding algorithm.
3. a kind of Wi-Fi indoor orientation methods based on deep learning according to claim 1 or 2, it is characterised in that：Step
It is rapid 4) described in region refer to all groups of samples in sub- fingerprint base into region, its partitioning standards are centered on the cluster heart
A square area, the big I in the region adjusts according to actual map area size.
4. a kind of Wi-Fi indoor orientation methods based on deep learning according to claim 1 or 3, it is characterised in that：Institute
State during indoor positioning, the matching algorithm described in step d) is KNN algorithms, formula：Wherein
caFor C(q)A line, a ∈ { 1,2 ..., m } ebFor E(q)In an element,For caIn an element.
5. a kind of Wi-Fi indoor orientation methods based on deep learning according to claim 1, it is characterised in that：It is described
Method is applied to the positioning of parking stall in large-scale synthesis body room.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710323523.3A CN107037399A (en) | 2017-05-10 | 2017-05-10 | A kind of Wi Fi indoor orientation methods based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710323523.3A CN107037399A (en) | 2017-05-10 | 2017-05-10 | A kind of Wi Fi indoor orientation methods based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107037399A true CN107037399A (en) | 2017-08-11 |
Family
ID=59538012
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710323523.3A Pending CN107037399A (en) | 2017-05-10 | 2017-05-10 | A kind of Wi Fi indoor orientation methods based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107037399A (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107529222A (en) * | 2017-09-15 | 2017-12-29 | 上海交通大学 | A kind of WiFi indoor locating systems based on deep learning |
| CN109040957A (en) * | 2018-08-14 | 2018-12-18 | 广东小天才科技有限公司 | A kind of indoor orientation method and device based on WIFI |
| CN109934597A (en) * | 2019-02-22 | 2019-06-25 | 北京航天泰坦科技股份有限公司 | External tax control tray attachment device |
| CN110536257A (en) * | 2019-08-21 | 2019-12-03 | 成都电科慧安科技有限公司 | A kind of indoor orientation method based on depth adaptive network |
| CN110536245A (en) * | 2019-08-28 | 2019-12-03 | 杭州电子科技大学 | A kind of indoor wireless positioning method and system based on deep learning |
| CN110830939A (en) * | 2019-08-29 | 2020-02-21 | 天津大学 | Positioning method based on improved CPN-WLAN fingerprint positioning database |
| CN110892760A (en) * | 2017-08-21 | 2020-03-17 | 北京嘀嘀无限科技发展有限公司 | Positioning terminal equipment based on deep learning |
| CN110933596A (en) * | 2019-12-04 | 2020-03-27 | 哈尔滨工业大学 | Fingerprint positioning method based on metric learning |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105072169A (en) * | 2015-07-30 | 2015-11-18 | 中国传媒大学 | Intelligent information display system of culture exhibition hall |
| CN105242239A (en) * | 2015-10-19 | 2016-01-13 | 华中科技大学 | Indoor subarea positioning method based on crowdsourcing fingerprint clustering and matching |
-
2017
- 2017-05-10 CN CN201710323523.3A patent/CN107037399A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105072169A (en) * | 2015-07-30 | 2015-11-18 | 中国传媒大学 | Intelligent information display system of culture exhibition hall |
| CN105242239A (en) * | 2015-10-19 | 2016-01-13 | 华中科技大学 | Indoor subarea positioning method based on crowdsourcing fingerprint clustering and matching |
Non-Patent Citations (2)
| Title |
|---|
| M.K. LEE ETC.: ""Dimensionality reduction of radio map with nonlinear autoencoder"", 《ELECTRONICS LETTERS》 * |
| 唐洋等: ""基于WiFi的指纹匹配算法在室内定位中的应用研究"", 《计算机科学》 * |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110892760A (en) * | 2017-08-21 | 2020-03-17 | 北京嘀嘀无限科技发展有限公司 | Positioning terminal equipment based on deep learning |
| CN110892760B (en) * | 2017-08-21 | 2021-11-23 | 北京嘀嘀无限科技发展有限公司 | Positioning terminal equipment based on deep learning |
| CN107529222A (en) * | 2017-09-15 | 2017-12-29 | 上海交通大学 | A kind of WiFi indoor locating systems based on deep learning |
| CN109040957A (en) * | 2018-08-14 | 2018-12-18 | 广东小天才科技有限公司 | A kind of indoor orientation method and device based on WIFI |
| CN109934597A (en) * | 2019-02-22 | 2019-06-25 | 北京航天泰坦科技股份有限公司 | External tax control tray attachment device |
| CN110536257A (en) * | 2019-08-21 | 2019-12-03 | 成都电科慧安科技有限公司 | A kind of indoor orientation method based on depth adaptive network |
| CN110536257B (en) * | 2019-08-21 | 2022-02-08 | 成都电科慧安科技有限公司 | Indoor positioning method based on depth adaptive network |
| CN110536245A (en) * | 2019-08-28 | 2019-12-03 | 杭州电子科技大学 | A kind of indoor wireless positioning method and system based on deep learning |
| CN110536245B (en) * | 2019-08-28 | 2020-10-30 | 杭州电子科技大学 | Deep learning-based indoor wireless positioning method and system |
| CN110830939A (en) * | 2019-08-29 | 2020-02-21 | 天津大学 | Positioning method based on improved CPN-WLAN fingerprint positioning database |
| CN110830939B (en) * | 2019-08-29 | 2021-04-27 | 天津大学 | Positioning method based on improved CPN-WLAN fingerprint positioning database |
| CN110933596A (en) * | 2019-12-04 | 2020-03-27 | 哈尔滨工业大学 | Fingerprint positioning method based on metric learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107037399A (en) | A kind of Wi Fi indoor orientation methods based on deep learning | |
| CN105101408B (en) | Indoor orientation method based on distributed AP selection strategy | |
| CN103889051B (en) | Indoor WLAN fingerprint positioning methods based on AP ID filterings and Kalman filtering | |
| 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 | |
| CN108540929B (en) | Indoor fingerprint positioning method based on RSSI signal strength sequencing | |
| CN114067160A (en) | Small sample remote sensing image scene classification method based on embedded smooth graph neural network | |
| CN109614973A (en) | Rice seedling and Weeds at seedling image, semantic dividing method, system, equipment and medium | |
| CN114092832B (en) | High-resolution remote sensing image classification method based on parallel hybrid convolutional network | |
| CN108627798B (en) | WLAN indoor positioning algorithm based on linear discriminant analysis and gradient lifting tree | |
| CN109299707A (en) | A kind of unsupervised pedestrian recognition methods again based on fuzzy depth cluster | |
| CN112765381A (en) | Image retrieval method, electronic equipment and related product | |
| CN112653991A (en) | WLAN indoor positioning method of TebNet neural network model based on deep learning | |
| CN107817466B (en) | Indoor positioning method based on stacking limited Boltzmann machine and random forest algorithm | |
| CN113761259A (en) | Image processing method and device and computer equipment | |
| CN106792559A (en) | The automatic update method of fingerprint base in a kind of WiFi indoor locating systems | |
| CN107180245A (en) | A kind of indoor and outdoor scene recognition method and device | |
| CN106970379A (en) | Based on distance-measuring and positioning method of the Taylor series expansion to indoor objects | |
| CN104091181A (en) | Injurious insect image automatic recognition method and system based on deep restricted Boltzmann machine | |
| CN114359702A (en) | Method and system for identifying building violation of remote sensing image of homestead based on Transformer | |
| CN113269182A (en) | Target fruit detection method and system based on small-area sensitivity of variant transform | |
| CN113915840B (en) | Refrigerator and food material positioning method thereof | |
| CN108985378B (en) | Domain self-adaption method based on hybrid cross-depth network | |
| CN115567871A (en) | WiFi fingerprint indoor floor identification and position estimation method | |
| CN111639712A (en) | Positioning method and system based on density peak clustering and gradient lifting algorithm | |
| CN110826691A (en) | Intelligent seismic velocity spectrum pickup method based on YOLO and LSTM |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170811 |