CN113723240B - General fingerprint positioning method and system based on Boosting and sample difference - Google Patents
General fingerprint positioning method and system based on Boosting and sample difference Download PDFInfo
- Publication number
- CN113723240B CN113723240B CN202110952933.0A CN202110952933A CN113723240B CN 113723240 B CN113723240 B CN 113723240B CN 202110952933 A CN202110952933 A CN 202110952933A CN 113723240 B CN113723240 B CN 113723240B
- Authority
- CN
- China
- Prior art keywords
- positioning
- fingerprint
- sample
- boosting
- features
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D30/00—Reducing energy consumption in communication networks
- Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Software Systems (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Medical Informatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Collating Specific Patterns (AREA)
- Position Fixing By Use Of Radio Waves (AREA)
Abstract
The invention provides a general fingerprint positioning method and a system based on Boosting and sample difference, which are used for performing positioning model training and fingerprint database data acquisition under a positioning scene in an off-line stage, wherein the positioning model training comprises the steps of respectively performing multiple RSS information acquisitions at a plurality of known points of an unlimited scene, and pairing all samples to obtain a positive sample pair and a negative sample pair; computing a relative feature to represent the difference between the two samples in each sample pair; taking the extracted relative features as input of a classifier, and training the classifier in a Boosting mode to obtain a classification model; in the positioning stage, when positioning is needed at a certain moment, an observation list of a group of wireless data is obtained through scanning, then corresponding fingerprint information is screened by using an AP, relative characteristics are calculated according to screening results and the observation list, and the obtained characteristic vector is input into a positioning model trained in the off-line stage; the positioning model outputs the corresponding probability of each feature vector, and the point with the highest probability is the final positioning result.
Description
Technical Field
The invention relates to the technical field of fingerprint positioning, in particular to a general fingerprint positioning method and system based on Boosting and sample difference.
Background
The following describes the main positioning schemes and drawbacks of the prior art.
Geometric intersection positioning scheme: scheme of geometric intersection. The position of the point to be measured is solved by finding the intersection of a plurality of circles (ranging, TOA/TOF/RSS, etc.) or hyperbolas (ranging dispersion mode, TDOA).
Fingerprint positioning scheme: according to the non-machine learning scheme, similarity between fingerprint points and signal information of points to be detected is calculated by using some similarity calculation methods (Euclidean distance, manhattan distance, pearson coefficient, cosine similarity and the like), and positions of the points to be detected are calculated by using one or a plurality of fingerprint points with highest similarity (one is directly output, and a plurality of positions can be calculated by means of averaging or weighting averaging).
Geometric intersection positioning scheme defect: the position of the base station is required to be known, the requirement on the layout network shape of the base station is high, and in addition, the network shape of the base station is required to be guaranteed well in order to ensure the precision during online positioning.
Fingerprint positioning scheme defect: the non-machine learning scheme, namely various schemes for solving the signal similarity, cannot better reflect the nonlinear relation between the signal information of the fingerprint points and the points to be tested by a direct differential mode; in the machine learning scheme, various multi-classification schemes face the prominent problem of small samples, in addition, the models do not have migration capability, the models trained in one scene cannot be used in another scene, one model in one scene does not have practical application value, and if a large number of models are deployed on line, effective updating and management (fingerprint positioning requires periodical updating of data and models) cannot be performed.
The related names are explained as follows:
GBDT (gradient boost decision tree) decision promoting tree
KNN (k nearest neighbor) k nearest neighbor point
AP (access point) access point
TOA (time of arrival) time of arrival of signal
TDOA (time difference of arrival) time difference of arrival of signals
RSS (received signal strength) signal reception intensity
TOF (time of flight) time of flight
FP (Fingerprint Point) fingerprint points
MAC (Media Access Control Address) the MAC address, also called physical address
Disclosure of Invention
In order to solve the defects in the prior art, the invention provides a general fingerprint positioning method based on Boosting and sample difference.
In order to achieve the above purpose, the technical proposal provided by the invention is a general fingerprint positioning method based on Boosting and sample difference, in the off-line stage, positioning model training and fingerprint database data acquisition under a positioning scene are carried out,
the implementation mode of the positioning model training is that multiple times of RSS information acquisition are respectively carried out at a plurality of known points of an unlimited scene, and data acquired each time is used as a sample; after the RSS information is collected, all samples are paired to obtain a positive sample pair and a negative sample pair; calculating relative features to represent differences between two samples in each sample pair, the relative features including coincidence features, ordering features, similarity features, and shift features; taking the extracted relative features as the input of a classifier, training the classifier in a Boosting mode to obtain a two-classification model, wherein the output of the two-classification model is the probability that two sample data are derived from the same point or adjacent points;
the implementation mode of fingerprint database data acquisition in the positioning scene is that wireless signal data are acquired on a plurality of fingerprint points with known coordinates in the positioning scene, the wireless signal data acquired on each fingerprint point are called fingerprint information, and the wireless signal data acquired on each fingerprint point comprise scanned AP names, MAC and corresponding signal intensities;
in the positioning stage, when positioning is needed at a certain moment, an observation list of a group of wireless data is obtained through scanning, then corresponding fingerprint information is screened by using an AP, relative characteristics are calculated according to screening results and the observation list, and the obtained characteristic vector is input into a positioning model trained in the off-line stage; the positioning model outputs the corresponding probability of each feature vector, and the point with the highest probability is the final positioning result.
Moreover, the similarity features include Euclidean distance, cosine similarity, chebyshev distance, pearson coefficient, manhattan distance, dot product ratio, normalized dot product ratio, and morphological similarity distance.
Moreover, the shift features include bit variance, bit mean difference, swap difference, and swap distance difference.
And, the RSS information acquisition result is normalized.
The invention provides a general fingerprint positioning system based on Boosting and sample difference, which is used for realizing the general fingerprint positioning method based on Boosting and sample difference.
Furthermore, the device comprises the following modules,
a first module for performing positioning model training in an off-line stage,
the implementation mode of the positioning model training is that multiple times of RSS information acquisition are respectively carried out at a plurality of known points of an unlimited scene, and data acquired each time is used as a sample; after the RSS information is collected, all samples are paired to obtain a positive sample pair and a negative sample pair; calculating relative features to represent differences between two samples in each sample pair, the relative features including coincidence features, ordering features, similarity features, and shift features; taking the extracted relative features as the input of a classifier, training the classifier in a Boosting mode to obtain a two-classification model, wherein the output of the two-classification model is the probability that two sample data are derived from the same point or adjacent points;
the second module is used for collecting fingerprint database data in a positioning scene in an off-line stage;
the implementation mode of fingerprint database data acquisition in the positioning scene is that wireless signal data are acquired on a plurality of fingerprint points with known coordinates in the positioning scene, the wireless signal data acquired on each fingerprint point are called fingerprint information, and the wireless signal data acquired on each fingerprint point comprise scanned AP names, MAC and corresponding signal intensities;
the third module is used for obtaining an observation list of a group of wireless data through scanning when the positioning is needed at a certain moment in the positioning stage, then screening corresponding fingerprint information by using an AP, calculating relative characteristics according to a screening result and the observation list, and inputting the obtained characteristic vector into a positioning model trained in the off-line stage; the positioning model outputs the corresponding probability of each feature vector, and the point with the highest probability is the final positioning result.
Alternatively, the fingerprint positioning device comprises a processor and a memory, wherein the memory is used for storing program instructions, and the processor is used for calling the stored instructions in the memory to execute a general fingerprint positioning method based on Boosting and sample difference.
Or, a readable storage medium having stored thereon a computer program which, when executed, implements a general fingerprint positioning method based on Boosting and sample differences as described above.
The invention proposes the following improvements:
1) The training model has universality and can be migrated to each scene for use;
2) The absolute features originally used in fingerprint positioning are converted into various relative features;
3) Converting the original multi-classification problem into a classification problem;
therefore, the invention has the following advantages:
advantage 1: the machine learning classification scheme can more sufficiently reflect the difference and similarity between signal vectors than a scheme using similarity calculation.
Advantage 2: compared with a multi-classification scheme, the two-classification scheme provided by the patent can greatly expand the number of samples through the cross combination among the samples, the multi-classification scheme faces small sample learning, the model is easy to fit, and meanwhile, the adaptability of the model is extremely weak; in addition, the scheme using the relative features enables the trained model to have migration capability, the model trained by data of a certain scene is applicable to another scene, the multi-classification scheme must ensure that the model of each scene is trained, the model cannot be migrated to other scenes, the multi-classification scheme is not applicable to practical application, when the model is deployed on the practical line, the problem that a large number of models cannot be effectively managed is still solved, the data and the models in fingerprint positioning are periodically updated, and the scheme of one model of each scene is obviously not applicable.
The scheme of the invention is simple and convenient to implement, has strong practicability, solves the problems of low practicability and inconvenient practical application existing in the related technology, can improve user experience, and has important market value.
Drawings
FIG. 1 is a general block diagram of a system according to an embodiment of the present invention;
FIG. 2 is a diagram of positive and negative sample selection indication according to an embodiment of the present invention;
FIG. 3 is a schematic diagram of a feature calculation mode according to an embodiment of the present invention;
FIG. 4 is a schematic diagram of an optional feature list according to an embodiment of the invention.
Detailed Description
The technical scheme of the invention is specifically described below with reference to the accompanying drawings and examples.
The general fingerprint positioning model based on Boosting and sample difference is suitable for various wireless signal fingerprint positioning algorithms such as WiFi, bluetooth, UWB and the like, and the overall structure of the system is shown in figure 1.
Fig. 1 shows the general structure of the proposed method, including two main stages, offline and online.
1. The offline stage mainly comprises positioning model training and fingerprint database data acquisition under the positioning scene.
1) The training data used for training the positioning model can be wireless signal data collected at some known points in the positioning scene, can be wireless signal data collected at some known points in other scenes, and can be wireless signal data mixture collected at some known points in a plurality of scenes. Such as: the data used for model training may be wireless signal data collected at some known points within building 1, or data collected within building 2. The method is equally applicable to different floor data, such as floor 1 data trained models are equally applicable to floor 5.
While the conventional method requires that fingerprint data in a positioning scene must be used for training, the method of the invention does not require which scene the training data comes from, and therefore has higher applicability.
In an embodiment, the training data originates from acquiring wireless signal data at a number of known points of an unlimited scene, preferably at each known point a plurality of times, respectively, in order to improve the quality of the training samplesRSS acquisition, wherein each acquired data is taken as a sample (the offline acquisition part in FIG. 1 is acquired multiple times at each known point, and each acquisition corresponds to a group of scan lists, for example, the list of each scan in the known point A is FP A1 、FP A2 、FP A3 ....wherein FP A1 { "AP_01": -43dBm; "AP_02": 55dBm; "AP_03": -67dBm; ... Once again, it is known that in a similar way in point B, the list of each scan is FP B1 、FP B2 、FP B3 ......). After the RSS information is collected, all samples are paired. If two samples in a pair of samples are from the same known point or adjacent known points, then these pairs of samples are considered positive pairs of samples for subsequent positive sample feature computation. If they come from different known points while the distance between them is large enough, they are negative sample pairs for subsequent negative sample feature calculations. Then, some new relative features are calculated to represent the difference between the two samples in each sample pair, including coincidence, ordering, similarity, and shift features. The new feature extracted is the input to the classifier, rather than training using the original MAC-RSS pair as the input to the classifier as in the conventional approach. The patent adopts Boosting method (lifting method) to train the classifier: training with some basic classifiers, the output of the former classifier will be the input of the next classifier, and samples with more concern for misclassification will be taken at the next classification (offline training part in fig. 1, transmitting training residual error to the next classifier after each classifier training is finished, and ensuring the minimum residual error of each classifier, namely ensuring the optimal classifier parameters 1 、ω 2 、ω 3 ......ω n The weights may be equal weights or may be determined based on the inverse of the training residual). And finally, outputting a classification model, wherein the output of the model is the probability that two sample data are from the same point or adjacent points.
2) The fingerprint database data acquisition mainly acquires wireless signal data at a plurality of points with known coordinates in a positioning scene, each known point is called a fingerprint point, the wireless signal data acquired at each point is called fingerprint information, the wireless signal data acquired at each fingerprint point is usually a wireless signal data list, and the list comprises scanned MAC and corresponding signal intensity. The fingerprint data used, if located within the building 1, must be the fingerprint data collected within the building 1.
In the implementation, a fingerprint library is established, which can be that a plurality of known points (namely fingerprint points) are arranged in advance in the positioning scene, wireless signal data (namely fingerprint information) are collected at each known point, the information is stored in a database (namely the fingerprint library), the information is used in real-time positioning, namely, the fingerprint information on which fingerprint point is found out in the fingerprint library is most similar to the signal intensity of the wireless signal data scanned in real-time positioning, and the fingerprint point with the most similar signal intensity is output as a positioning result.
2. In the positioning stage, a certain moment needs to be positioned, a group of wireless data observation list is scanned at the moment, the scanned APs in the observation list are scanned on some fingerprint points, and the fingerprint points are screened out for further calculation. Then, the characteristics are calculated by using the filtered fingerprint information and the observation list (the method is the same as that of calculating the relative characteristics by using two sample data when training a model). And calculating the fingerprint information on each fingerprint point and the observation list to obtain a feature vector, and inputting the calculated feature vectors into the positioning model trained in the off-line stage one by one. The localization model outputs a probability of one of each feature vector (being a probability of a neighboring point or the same point) and an attribute value (whether it is a neighboring point or the same point). Finally, the point with the highest probability is the final positioning result.
For ease of implementation reference, the implementation of the two phases of the embodiments is described in detail below:
part A off-line acquisition and model training
Step 1: signal fingerprint information is acquired at known points (multiple acquisitions at each fingerprint point are recommended).
Step 2: and carrying out normalization processing on each piece of acquired information on each fingerprint point.
Wherein, the liquid crystal display device comprises a liquid crystal display device,rss std mean and variance of RSS list are represented, respectively, +.>Indicating the signal strength of the ith AP in the scan list.
Step 3: pairing the collected fingerprint information pairwise, and if two pieces of information come from the same fingerprint point or the distance between the fingerprint points from which the two pieces of information come is smaller than a certain threshold value, using the pair of information for generating positive samples subsequently. If two pieces of information come from different fingerprint points and the distance between the two fingerprint points is greater than a certain threshold value, such a stack of information is used for subsequent generation of negative samples. (specific pairing modes see FIG. 2)
In fig. 2, different shapes represent different fingerprint points, and a plurality of points of the same shape represent multiple acquisitions of the fingerprint points. Here, to show how the fingerprint points are paired, two sets of scan lists from the same fingerprint point form positive sample pairs for positive sample feature computation, and two sets of scan lists from different fingerprint points form negative sample pairs for negative sample feature computation.
Step 4: the new relative characteristics are extracted by using the information of the one-to-one pair arranged in step2, wherein the new relative characteristics comprise similarity characteristics, ordering characteristics, shifting characteristics and coincidence number characteristics, namely, a new characteristic vector is obtained by comparing and calculating two pieces of information in the one pair, and the characteristic vector can be a positive sample or a negative sample, and specific division is seen in step2. (see FIG. 3 for specific selected features).
Fig. 3 lists the relative features that need to be calculated, i.e. the relative feature list includes the number of coincidences feature, the ordering feature, the similarity feature and the shift feature. Ordering characteristics include bit 1, bit 2, bit 3 … (two scan lists, one of which is taken as a reference list, here looking at the ordering of RSS first strong APs in the other list, the ordering of second strong APs in the reference list, and so on.) similarity characteristics include euclidean distance, cosine similarity, chebyshev distance, pearson coefficient, manhattan distance, dot product ratio, normalized dot product ratio, and morphological similarity distance. The shift features include bit variance, bit mean difference, swap difference, and swap distance difference.
Step 5: the classification model is trained with positive and negative samples sorted in step3, while setting the probability of the model output as positive samples.
Part B online positioning
Step 1: when a certain moment needs to be positioned, a group of observation lists of wireless data are obtained through scanning, firstly, AP screening is carried out, signal data of a mobile AP are screened out, firstly, compared with signals scanned earlier, the signals have larger abnormal movement distance, but the RSS of one or more APs has little change and is deleted; and secondly, comparing a keyword list (which is maintained by the scanning information system, namely ' mobile ', ' smart ', ' HUAWE ', ' OPPO ', ' and the like) }, and deleting the field in the keyword list if the AP name contains.
Step 2: normalization processing comprises the Step1 screening of RSS in the observation list.
Step 3: and extracting key fingerprint points, and screening out fingerprint points containing the AP which is positioned and scanned in real time from a fingerprint library.
Step 4: and comparing the results of the real-time scanning information processed by Step1 and Step2 with the fingerprint point information screened by Step3 to make the same feature vector, namely the relative feature, as that of the offline training.
Fig. 4 shows how the relative features are calculated using the filtered fingerprint points and the observation list, and fig. 4 shows two scan lists, one being a list of APs scanned at the time of positioning (where the MAC of the AP is AP 1 ，AP 2 ，AP 3 ,. the signal strengths corresponding thereto are RSS, respectively 1’ ，RSS 2’ ，RSS 3’ ,.. Once more), one is a list of APs stored at a certain fingerprint point (whereinThe MAC of the AP is AP 1 ，AP 2 ，AP 3 ,. the signal strengths corresponding thereto are RSS, respectively 1 ，RSS 2 ，RSS 3 ，......)。
Step 5: and inputting the sorted characteristics into an offline training model, and outputting the probability that each screened fingerprint point is a neighboring point.
Step 6: and sequencing the probabilities output in step5, and outputting the fingerprint point position with the highest probability as a final positioning result.
In order to provide similarity features and shift features among the relative features, a specific explanation is made below.
The following are various similarity distance calculation formulas:
a. euclidean distance:
wherein the method comprises the steps ofand/>Respectively represent ap in list A i Ap in RSS and list B of (E) i Is the number of mac overlapping in the two lists
b. Cosine similarity, cosine Similarity:
c. chebyshev distance, chebyshev Distance:
where Max (·) represents the maximum calculation function.
d. Pearson coefficients, pearson's Coefficient:
wherein the method comprises the steps ofAnd->Average RSS of list a and B are shown, respectively.
e. Manhattan distance, manhattan Distance:
f. dot product ratio, dot Product Ratio (DPR):
g. normalized dot product ratio, normalization Dot Product Ratio (NDPR):
h. morphology similarity distance, morphological Similarity Distance:
the following are various shift feature calculation formulas:
a. bit variance Location Square Deviation (LSD)
Wherein N represents the length of the AP list, and the A and B are respectivelyRespectively represent different lists ap i [A]Representing ap i Ordering in A, ap i [B]Representing ap i Ordering in B.
b. Bit mean difference, location Mean Deviation (LMD):
c. exchange difference, swap Device (SD):
D sd ＝Min(W(A→B))＝Min(∑ s∈A→B 1) (11)
where A→B represents a shift operation from A to B, W (-) represents a weight calculation function, and Min (-) represents a minimum calculation function. In specific implementation, the weights may be 1 (equal weight), or alternatively: 1/shift number, 1+δ is taken when the shift number is 0, and δ represents a minimum value, such as 0.001.
d. Exchange distance difference, swap Distance Deviation (SDD):
D sdd ＝Min(W(A→B))＝Min(∑ s∈A→B |i-j|) (12)
where i, j represents the ordering of an AP in the AP lists a and B, respectively.
In particular, the method according to the technical solution of the present invention may be implemented by those skilled in the art using computer software technology to implement an automatic operation flow, and a system apparatus for implementing the method, such as a computer readable storage medium storing a corresponding computer program according to the technical solution of the present invention, and a computer device including the operation of the corresponding computer program, should also fall within the protection scope of the present invention.
In some possible embodiments, a universal fingerprint positioning system based on Boosting and sample diversity is provided, comprising the following modules,
a first module for performing positioning model training in an off-line stage,
the implementation mode of the positioning model training is that multiple times of RSS information acquisition are respectively carried out at a plurality of known points of an unlimited scene, and data acquired each time is used as a sample; after the RSS information is collected, all samples are paired to obtain a positive sample pair and a negative sample pair; calculating relative features to represent differences between two samples in each sample pair, the relative features including coincidence features, ordering features, similarity features, and shift features; taking the extracted relative features as the input of a classifier, training the classifier in a Boosting mode to obtain a two-classification model, wherein the output of the two-classification model is the probability that two sample data are derived from the same point or adjacent points;
the second module is used for collecting fingerprint database data in a positioning scene in an off-line stage;
the implementation mode of fingerprint database data acquisition in the positioning scene is that wireless signal data are acquired on a plurality of fingerprint points with known coordinates in the positioning scene, the wireless signal data acquired on each fingerprint point are called fingerprint information, and the wireless signal data acquired on each fingerprint point comprise scanned AP names, MAC and corresponding signal intensities;
the third module is used for obtaining an observation list of a group of wireless data through scanning when the positioning is needed at a certain moment in the positioning stage, then screening corresponding fingerprint information by using an AP, calculating relative characteristics according to a screening result and the observation list, and inputting the obtained characteristic vector into a positioning model trained in the off-line stage; the positioning model outputs the corresponding probability of each feature vector, and the point with the highest probability is the final positioning result.
In some possible embodiments, a general fingerprint positioning system based on Boosting and sample difference is provided, including a processor and a memory, the memory is used for storing program instructions, and the processor is used for calling the stored instructions in the memory to execute a general fingerprint positioning method based on Boosting and sample difference as described above.
In some possible embodiments, a general fingerprint positioning system based on Boosting and sample difference is provided, which comprises a readable storage medium, wherein a computer program is stored on the readable storage medium, and the computer program is executed to realize a general fingerprint positioning method based on Boosting and sample difference.
The specific embodiments described herein are offered by way of example only to illustrate the spirit of the invention. Those skilled in the art may make various modifications or additions to the described embodiments or substitutions thereof without departing from the spirit of the invention or exceeding the scope of the invention as defined in the accompanying claims.
Claims (8)
1. A general fingerprint positioning method based on Boosting and sample difference is characterized in that: in the off-line stage, the positioning model training and the fingerprint database data acquisition under the positioning scene are carried out,
the implementation mode of the positioning model training is that multiple times of RSS information acquisition are respectively carried out at a plurality of known points of an unlimited scene, and data acquired each time is used as a sample; after the RSS information is collected, all samples are paired to obtain a positive sample pair and a negative sample pair; calculating relative features to represent differences between two samples in each sample pair, the relative features including coincidence features, ordering features, similarity features, and shift features; taking the extracted relative features as the input of a classifier, training the classifier in a Boosting mode to obtain a two-classification model, wherein the output of the two-classification model is the probability that two sample data are derived from the same point or adjacent points; the RSS represents signal reception strength;
the implementation mode of fingerprint database data acquisition in the positioning scene is that wireless signal data are acquired on a plurality of fingerprint points with known coordinates in the positioning scene, the wireless signal data acquired on each fingerprint point are called fingerprint information, and the wireless signal data acquired on each fingerprint point comprise scanned AP names and corresponding signal intensities; the AP represents an access point;
in the positioning stage, when positioning is needed at a certain moment, an observation list of a group of wireless data is obtained through scanning, then corresponding fingerprint information is screened by using an AP, relative characteristics are calculated according to screening results and the observation list, and the obtained characteristic vector is input into a positioning model trained in the off-line stage; the positioning model outputs the corresponding probability of each feature vector, and the point with the highest probability is the final positioning result; the relative characteristics are calculated according to the screening result and the observation list, and the implementation mode is as follows,
and comparing the results of the real-time scanning information after AP screening and RSS normalization processing with the fingerprint point information of the AP screened out from the fingerprint library and containing the real-time positioning scanning, and calculating to make the same feature vector as that in offline training to obtain the relative features.
2. The general fingerprint positioning method based on Boosting and sample difference according to claim 1, wherein: the similarity features include Euclidean distance, cosine similarity, chebyshev distance, pearson coefficient, manhattan distance, dot product ratio, normalized dot product ratio, and morphological similarity distance.
3. The general fingerprint positioning method based on Boosting and sample difference according to claim 1, wherein: the shift features include bit variance, bit mean difference, swap difference, and swap distance difference.
4. The general fingerprint positioning method based on Boosting and sample difference according to claim 1, wherein: and normalizing the RSS information acquisition result.
5. A general fingerprint positioning system based on Boosting and sample difference is characterized in that: a method for implementing a universal fingerprint positioning method based on Boosting and sample differentiation according to any one of claims 1-4.
6. The universal fingerprint positioning system based on Boosting and sample differentiation as recited in claim 5, wherein: comprising the following modules, wherein the modules are arranged in a row,
a first module for performing positioning model training in an off-line stage,
the implementation mode of the positioning model training is that multiple times of RSS information acquisition are respectively carried out at a plurality of known points of an unlimited scene, and data acquired each time is used as a sample; after the RSS information is collected, all samples are paired to obtain a positive sample pair and a negative sample pair; calculating relative features to represent differences between two samples in each sample pair, the relative features including coincidence features, ordering features, similarity features, and shift features; taking the extracted relative features as the input of a classifier, training the classifier in a Boosting mode to obtain a two-classification model, wherein the output of the two-classification model is the probability that two sample data are derived from the same point or adjacent points;
the second module is used for collecting fingerprint database data in a positioning scene in an off-line stage;
the implementation mode of fingerprint database data acquisition in the positioning scene is that wireless signal data are acquired on a plurality of fingerprint points with known coordinates in the positioning scene, the wireless signal data acquired on each fingerprint point are called fingerprint information, and the wireless signal data acquired on each fingerprint point comprise scanned AP names and corresponding signal intensities;
the third module is used for obtaining an observation list of a group of wireless data through scanning when the positioning is needed at a certain moment in the positioning stage, then screening corresponding fingerprint information by using an AP, calculating relative characteristics according to a screening result and the observation list, and inputting the obtained characteristic vector into a positioning model trained in the off-line stage; the positioning model outputs the corresponding probability of each feature vector, and the point with the highest probability is the final positioning result.
7. The universal fingerprint positioning system based on Boosting and sample differentiation as recited in claim 6, wherein: comprising a processor and a memory for storing program instructions, the processor being adapted to invoke the stored instructions in the memory to perform a general fingerprint positioning method based on Boosting and sample differences as claimed in any of claims 1-4.
8. The universal fingerprint positioning system based on Boosting and sample differentiation as recited in claim 6, wherein: comprising a readable storage medium having stored thereon a computer program which, when executed, implements a general fingerprint positioning method based on Boosting and sample differences as claimed in any one of claims 1-4.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110952933.0A CN113723240B (en) | 2021-08-19 | 2021-08-19 | General fingerprint positioning method and system based on Boosting and sample difference |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110952933.0A CN113723240B (en) | 2021-08-19 | 2021-08-19 | General fingerprint positioning method and system based on Boosting and sample difference |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113723240A CN113723240A (en) | 2021-11-30 |
| CN113723240B true CN113723240B (en) | 2023-08-22 |
Family
ID=78676849
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110952933.0A Active CN113723240B (en) | 2021-08-19 | 2021-08-19 | General fingerprint positioning method and system based on Boosting and sample difference |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113723240B (en) |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103476115A (en) * | 2013-09-22 | 2013-12-25 | 中国地质大学（武汉） | Method for WiFi fingerprint positioning based on AP set similarity |
| CN107484118A (en) * | 2016-06-07 | 2017-12-15 | 滴滴（中国）科技有限公司 | A kind of indoor scene localization method and system based on building WiFi fingerprints |
| CN108540929A (en) * | 2018-03-29 | 2018-09-14 | 马梓翔 | Indoor fingerprint location system based on the sequence of RSSI signal strengths |
| CN111182453A (en) * | 2020-02-12 | 2020-05-19 | 腾讯科技（深圳）有限公司 | Positioning method, positioning device, electronic equipment and storage medium |
| CN112004203A (en) * | 2020-09-14 | 2020-11-27 | 西安科技大学 | Indoor positioning method and device based on position prediction and error compensation |
| CN112887902A (en) * | 2021-01-22 | 2021-06-01 | 湖南大学 | Indoor positioning method of WiFi fingerprint based on Gaussian clustering and hybrid measurement |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9408040B2 (en) * | 2013-02-14 | 2016-08-02 | Fuji Xerox Co., Ltd. | Systems and methods for room-level location using WiFi |
-
2021
- 2021-08-19 CN CN202110952933.0A patent/CN113723240B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103476115A (en) * | 2013-09-22 | 2013-12-25 | 中国地质大学（武汉） | Method for WiFi fingerprint positioning based on AP set similarity |
| CN107484118A (en) * | 2016-06-07 | 2017-12-15 | 滴滴（中国）科技有限公司 | A kind of indoor scene localization method and system based on building WiFi fingerprints |
| CN108540929A (en) * | 2018-03-29 | 2018-09-14 | 马梓翔 | Indoor fingerprint location system based on the sequence of RSSI signal strengths |
| CN111182453A (en) * | 2020-02-12 | 2020-05-19 | 腾讯科技（深圳）有限公司 | Positioning method, positioning device, electronic equipment and storage medium |
| CN112004203A (en) * | 2020-09-14 | 2020-11-27 | 西安科技大学 | Indoor positioning method and device based on position prediction and error compensation |
| CN112887902A (en) * | 2021-01-22 | 2021-06-01 | 湖南大学 | Indoor positioning method of WiFi fingerprint based on Gaussian clustering and hybrid measurement |
Non-Patent Citations (1)
| Title |
|---|
| 基于RSS 空间线性相关的WLAN 位置指纹定位算法;徐小良 等;《电信科学》;第14-21页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113723240A (en) | 2021-11-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106570505B (en) | Method and system for analyzing histopathological images | |
| CN106355577B (en) | Rapid image matching method and system based on significant condition and global coherency | |
| WO2019015246A1 (en) | Image feature acquisition | |
| CN108107444B (en) | Transformer substation foreign matter identification method based on laser data | |
| CN111652293B (en) | Vehicle weight recognition method for multi-task joint discrimination learning | |
| US20060120589A1 (en) | Image matching system using 3-dimensional object model, image matching method, and image matching program | |
| CN110188225B (en) | Image retrieval method based on sequencing learning and multivariate loss | |
| CN111524132B (en) | Method, device and storage medium for identifying abnormal cells in sample to be detected | |
| CN112633382A (en) | Mutual-neighbor-based few-sample image classification method and system | |
| CN110175615B (en) | Model training method, domain-adaptive visual position identification method and device | |
| JP5906071B2 (en) | Information processing method, information processing apparatus, and storage medium | |
| CN110619280B (en) | Vehicle re-identification method and device based on deep joint discrimination learning | |
| CN110163130B (en) | Feature pre-alignment random forest classification system and method for gesture recognition | |
| CN111144462A (en) | Unknown individual identification method and device for radar signals | |
| CN111985507B (en) | Rock mass three-dimensional point cloud joint trace extraction method | |
| CN113723240B (en) | General fingerprint positioning method and system based on Boosting and sample difference | |
| CN113128518A (en) | Sift mismatch detection method based on twin convolution network and feature mixing | |
| Gong et al. | Robust stepwise correspondence refinement for low-altitude remote sensing image registration | |
| CN116704171A (en) | Cross-view and cross-mode geographic positioning method based on multi-scale local descriptor aggregation network | |
| CN113326892A (en) | Relation network-based few-sample image classification method | |
| CN116310466A (en) | Small sample image classification method based on local irrelevant area screening graph neural network | |
| CN116798087A (en) | Employee on-duty state detection method and system | |
| CN113269678A (en) | Fault point positioning method for contact network transmission line | |
| CN116662588B (en) | Intelligent searching method and system for mass data | |
| CN113723468B (en) | Object detection method of three-dimensional point cloud |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |