CN114818869B

CN114818869B - A MR indoor and outdoor distinction method

Info

Publication number: CN114818869B
Application number: CN202210322133.5A
Authority: CN
Inventors: 袁大森; 柴悦; 梁虎
Original assignee: Bixing Iot Technology Shenzhen Co ltd; Inner Mongolia Autonomous Region Public Security Bureau
Current assignee: Bixing Iot Technology Shenzhen Co ltd; Inner Mongolia Autonomous Region Public Security Bureau
Priority date: 2022-03-29
Filing date: 2022-03-29
Publication date: 2025-03-21
Anticipated expiration: 2042-03-29
Also published as: CN114818869A

Abstract

The present invention discloses a method for distinguishing indoor and outdoor MR data, comprising: setting coding rules and forming a classifier in an offline stage; distinguishing indoor and outdoor MR data in an online stage. The present invention has the following beneficial effects: compared with supervised learning, the algorithm only uses one label, and the other label is unknown, which has lower use cost; compared with unsupervised learning, the algorithm uses a label known in the original data, which has higher accuracy.

Description

MR indoor and outdoor division method

Technical Field

The invention belongs to the field of mobile communication big data, and relates to indoor or outdoor distinction of 4G and 5G whole network wireless MR (measurement report) data, in particular to an MR indoor and outdoor distinction method.

Background

Indoor and outdoor positioning based on MR data can help operators to perform network optimization and data rendering on one hand, and can be used in the vertical industry to play a great value in fields such as public safety, epidemic prevention and control and the like on the other hand. The current main current MR indoor and outdoor positioning algorithms in the industry all adopt a two-stage architecture, namely firstly, the MR is distinguished indoor and outdoor, and secondly, the indoor MR and the outdoor MR are respectively subjected to position estimation by adopting different algorithms. The reason for this is mainly because the outdoor has rich GNSS data, which is written into the MR and reported to the base station together to be collected, which can help the positioning algorithm learn the mapping relationship between the outdoor signal characteristics and the outdoor location. However, due to the severe attenuation of satellite signals in indoor environment, GNSS positioning is difficult to be performed indoors, so that indoor MR lacks sufficient GNSS data to learn the mapping relationship between indoor signal characteristics and positions, and further, other algorithms are required to be used for optimizing indoor positioning results, so that positioning algorithms different from outdoor are differentiated to perform targeted optimization on indoor positioning results.

In summary, before the positioning algorithm, an indoor and outdoor segmentation module is needed to divide the MR to be positioned into indoor and outdoor areas, and instruct the subsequent procedure to use the corresponding indoor or outdoor positioning algorithm to finish the high-precision positioning of MR data. In addition, the indoor and outdoor segmentation algorithms need to have extremely high precision to ensure that corresponding MR is not misplaced into the corresponding positioning algorithms of other scenes.

The indoor and outdoor distinction scenario mainly involves two entities, a base station and a terminal. Terminals may be distributed indoors or outdoors, as may base stations. Base stations located indoors are generally called indoor substations, and these stations are mainly used for solving the coverage problem of indoor signals, and the coverage area is small (generally about 20 m), so that most of terminals connected to the indoor substations are located indoors, only a very small number of terminals located outdoors will be connected to the indoor substations, and because the coverage area of the indoor substations is small, these outdoor terminals are also very close to a building, so that the terminals connected to the indoor substations by default are located indoors.

The base stations located outdoors are called outdoor macro stations, which can solve the coverage problem of indoor and outdoor signals at the same time, have a large coverage (typically 400 m), and have a high ratio in the total MR reported by the whole network. Terminals accessing these base stations may be located indoors or outdoors and need to be distinguished by means of an indoor/outdoor differentiation algorithm.

Traditional indoor and outdoor segmentation algorithms can be divided into two types, supervised and unsupervised. The supervised indoor and outdoor distinguishing algorithm collects indoor and outdoor MR data marked by manpower in a crowdsourcing test mode, and the indoor and outdoor MR data are input into the supervised AI algorithm as training samples to obtain a classification model for indoor and outdoor distinction of online stage whole network data. This approach has high accuracy of the separation of indoor and outdoor areas in some scenarios, but the disadvantages are also apparent. On the one hand, the cost of manually carrying out indoor and outdoor MR marking is very high, because the wireless network environment is changed at a moment, data marking is required to be carried out regularly, otherwise, the model accuracy obtained by means of previous data training is drastically reduced, which brings great cost, on the other hand, the manual sampling is difficult to cover all scenes, the sampled data is influenced by factors such as terminal type, mobile phone pose and the like, and only one sampling of the whole network MR data is obtained, so that the trained model is not stable enough, has high accuracy in certain scenes and can not be used in certain scenes.

The non-supervision indoor and outdoor segmentation algorithm automatically learns the distribution characteristics of indoor and outdoor signals under the characteristics based on the service characteristics, and completes indoor and outdoor estimation based on the distribution characteristics. Compared with a supervised algorithm, the algorithm does not need to manually mark the MR, so that the commercial cost is effectively reduced. However, the accuracy of the unsupervised algorithm is limited, and the service features used by the unsupervised algorithm belong to expert experiences, which cannot cover all scenes in consideration of the variability of the wireless environment, so that the accuracy is poor in some scenes.

Disclosure of Invention

The invention provides an indoor and outdoor MR (magnetic resonance) method, which solves the cost problem of traditional supervised learning and the precision problem of unsupervised learning.

To solve the above problems, in one aspect, the present invention provides a method for differentiating an MR room from an MR room, comprising:

Setting coding rules and forming a classifier through an offline stage;

the MR data are distinguished from each other indoors and outdoors by an online phase.

The step of setting the coding rules and forming the classifier through the offline stage comprises the following steps:

Automatically labeling outdoor MR data;

Forming coding rules for use in the online stage;

The classifier is trained according to a method of incomplete supervised learning.

The automatic labeling of MR data outside the room includes:

acquiring full MR data over a period of time;

Extracting MR data carrying GNSS data from the full MR data;

MR data carrying GNSS data is marked outdoors, and MR data not carrying GNSS data is marked indoors.

The forming encoding rules for use in an online phase includes:

Grouping the full MR data according to the service cells;

Extracting service cell and neighbor cell lists from all MR data in each group, and performing de-duplication processing to obtain a cell list set serving as a feature column;

For a serving cell and a neighbor cell in each piece of MR data, if the serving cell and the neighbor cell appear in a cell set, filling in the RSRP value of the cell at the position of the corresponding cell in the set to generate a sparse vector, wherein each column of the vector represents one cell, and the value of the column represents the RSRP value of the current MR in the column;

Generating a matrix according to all the MR data in the same group to be used as the encoding result of the group MR data until each group obtains an encoded matrix respectively, thereby completing the encoding of all the data.

The method for training the classifier according to the incomplete supervised learning comprises the following steps:

Setting all data in the selected service cell as a set A, wherein the number of the data is M, using a set { X ₁,X₂,...,X_M } to represent the data, the ith MR data is represented as X _i, and the label is defined as y _i;

The number of cells after de-duplication is set as N, and is represented by a set { c ₁,c₂,...,c_N };

Forming a matrix of M (N+1), wherein the (N+1) th column represents the corresponding indoor or outdoor label, and according to the Bayesian theorem, the method can obtain:

In combination with laplace smoothing, when the indoor and outdoor tag y _j is known, the probability P (c _t|y_j) that the cell c _t corresponds to the RSRP value is:

Wherein N (c _t,X_i) represents the number of times that the RSRP of the t-th cell in the ith MR in the full set a is a valid value, if the RSRP thereof is not 0, the value of N (c _t,X_i) is 1, otherwise, the value is 0;

Setting that the values of the N cells in each MR are independently distributed in the same way, when X _i is known, the probability P (y _i＝1|X_i) that the tag is outdoor is:

Obtainable P (y _i＝0|X_i)＝1-P(y_i＝1|X_i)

And constructing a classifier.

The construction classifier includes:

extracting outdoor data of the tag in the set A to form a set O, and forming a set M by other indoor data;

Initializing model parameters P (y _i＝1|X_i):

setting P (y _i＝1|X_i) after the t-th iteration to be defined as P _t(y_i＝1|X_i), and performing loop iteration;

When the cycle is terminated, the values of P (c _k|y_i＝1)、P(c_k|y_i＝0)、P(y_i =1) and P (y _i =0) of the last round are saved, where k=1, 2..n, i=1, 2..m, which constitute the classifier of the cell.

The performing loop iteration includes:

If |P _t+1(y_i＝1|X_i)-P_t(y_i＝1|X_i) | < T, a loop is performed:

For each data X _i in set M:

The present round P (y _i)、P(c_t|y_i) is updated according to the previous round P (y _i-1)、P(c_t|y_i-1)

Calculating the current round P (y _i＝1|X_i) according to the updating result;

wherein, for the data of X _i e M, P (y _i＝1|X_i) =1 does not change with the iterative process, always remains constant.

The indoor and outdoor distinction of MR data by the online phase comprises:

finding out a corresponding coding rule of the service cell according to the selected service cell, and converting the MR data into a corresponding vector form according to the coding rule;

Finding corresponding values of P _i(c_k|y_i＝1)、P_i(c_k|y_i＝0)、P_i(y_i =1) and P _i(y_i =0 according to the service cell, and calculating to obtain the probability P (y _i＝1|X_i) that the current MR data is judged to be outdoor according to the calculation formula of P (y _i＝1|X_i);

If P (y _i＝1|X_i) >0.5, the MR data is determined to be outdoor, otherwise it is determined to be indoor.

The method for searching the coding rule corresponding to the service cell according to the selected service cell, converting the MR data into a corresponding vector form according to the coding rule, and further comprises the following steps:

if the selected cell in the MR data is not in the encoding rule, the cell is not encoded.

In one aspect, a computer readable storage medium having stored therein a plurality of instructions adapted to be loaded by a processor to perform a MR indoor and outdoor differentiation method as described above is provided.

The beneficial effects of the invention are as follows:

compared with supervised learning, the algorithm uses only one label, the other label is unknown, the use cost is lower, and compared with unsupervised learning, the algorithm uses a label known in the original data, and the algorithm has higher precision.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings that are needed in the description of the embodiments will be briefly described below, it being obvious that the drawings in the following description are only some embodiments of the present invention, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.

FIG. 1 is a flow chart of a method for differentiating MR images in and out of an MR chamber according to an embodiment of the invention;

Fig. 2 is a schematic diagram of MR data encoding according to an embodiment of the present invention.

Detailed Description

The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to fall within the scope of the invention.

In the description of the present invention, it should be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc. indicate orientations or positional relationships based on the drawings are merely for convenience in describing the present invention and simplifying the description, and do not indicate or imply that the apparatus or elements referred to must have a specific orientation, be configured and operated in a specific orientation, and thus should not be construed as limiting the present invention. Furthermore, the terms "first," "second," and the like, are used for descriptive purposes only and are not to be construed as indicating or implying a relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defining "a first" or "a second" may explicitly or implicitly include one or more features. In the description of the present invention, the meaning of "a plurality" is two or more, unless explicitly defined otherwise.

In the present invention, the term "exemplary" is used to mean "serving as an example, instance, or illustration. Any embodiment described as "exemplary" in this disclosure is not necessarily to be construed as preferred or advantageous over other embodiments. The following description is presented to enable any person skilled in the art to make and use the invention. In the following description, details are set forth for purposes of explanation. It will be apparent to one of ordinary skill in the art that the present invention may be practiced without these specific details. In other instances, well-known structures and processes have not been described in detail so as not to obscure the description of the invention with unnecessary detail. Thus, the present invention is not intended to be limited to the embodiments shown, but is to be accorded the widest scope consistent with the principles and features disclosed herein.

The scheme is based on an indoor and outdoor segmentation algorithm of the incompletely supervised learning, has lower cost than the supervised learning and higher precision than the unsupervised learning. The algorithm uses outdoor GNSS (global navigation satellite system ) data as outdoor MR tags, which can be automatically acquired from the whole-network MR data without manual labeling. For indoor MR labels, automatic marking is carried out through the outdoor labels and an incomplete supervision learning algorithm, and meanwhile, training of indoor and outdoor regional models can be completed and the indoor and outdoor MR labels are used for indoor and outdoor MR distinction in an online stage. Because the whole process is not manually participated, the cost can be effectively reduced. In addition, the method is based on the model which is effective for indoor and outdoor areas and automatically learns by the whole-network part outdoor tag, so that the scene coverage is more comprehensive, the phenomenon of low precision in certain scenes caused by experience of a specialist in an unsupervised algorithm can be effectively avoided, and the method has higher precision than unsupervised learning.

Prior to the description of the algorithm, it is necessary to solve the MR data. The present specification MR data is processed into a format as shown in the following table. Where (ci: val) represents the level value (also referred to as RSRP value) of a cell, where ci represents the cell number, val represents the signal strength in dBm received by the MR by the cell. The first data in each MR represents the serving cell, the other data represents the neighbor, and the subsequent indoor distinction is in MR distinction granularity.

Table 1MR data schematic

Sequence number	Format of the form
		MR1	c1:-90,c2:-95,c4:-100,c5:-103
MR2	c6:-75,c1:-80.c2:-99,c10:-103
		......	......
MRN	c3:-80,c2:-90,c8:-105

Referring to fig. 1, fig. 1 is a schematic flow chart of an MR indoor and outdoor division method according to an embodiment of the invention, which is generally divided into an offline stage and an online stage. The off-line stage mainly completes the generation of the coding rules and the classifier, and the on-line stage completes the indoor and outdoor distinction of the MR data by the products of the off-line stage. The method for dividing the MR into the indoor region and the outdoor region comprises the following steps of S1-S2:

s1, setting coding rules and forming a classifier through an offline stage, wherein the step S1 comprises the steps of S11-S13:

S11, automatically labeling outdoor MR data, wherein the total MR data in a certain time period is acquired, the MR data carrying GNSS data in the total MR data is extracted, the MR data carrying the GNSS data is labeled outdoor, and the MR data not carrying the GNSS data is labeled indoor.

In this embodiment, the outdoor MR data is automatically labeled, and only part of the data of the outdoor tag is generated, and the indoor and outdoor tags of other data are unknown. That is, the total MR data in a certain period of time is extracted, the MR with GNSS data carried therein is extracted, these MR data are marked outdoor, the other data are temporarily marked indoor, and for simplicity, the outdoor and indoor tags are abbreviated as 1 and 0, respectively. The following examples are shown:

TABLE 2 schematic indoor and outdoor MR labeling

Data	Whether or not it is a GNSS	Labeling results
			MR1	Is that	1
MR2	Is that	1
			MR3	Whether or not	0
......	......	......
			MRN	Whether or not	0

S12, forming a coding rule for use in an online stage, wherein the step S12 comprises the steps S121-S124:

S121, grouping the full MR data according to a service cell;

S122, extracting service cell and neighbor cell lists from all MR data in each group, and performing de-duplication processing to obtain a cell list set serving as a feature column;

In this embodiment, for all MR data in each packet, the serving cell and the neighbor cell list are extracted, and deduplication processing is performed to obtain a cell list set as a feature column.

S123, for a serving cell and a neighbor cell in each piece of MR data, if the serving cell and the neighbor cell exist in a cell set, filling in an RSRP value of the cell at the position of the corresponding cell in the set to generate a sparse vector, wherein each column of the vector represents one cell, and the value of the column represents the RSRP value of the current MR in the column;

In this embodiment, for the serving cell and the neighbor cell in each piece of MR data, if they appear in the cell set, the RSRP value of the cell is filled in the location of the corresponding cell in the set, otherwise, 0 is filled in. A sparse vector is finally generated, each column of the vector representing a cell, the value of the column representing the RSRP value of the current MR in that column.

S124, generating a matrix according to all the MR data in the same group to be used as the encoding result of the MR data of the group until each group obtains an encoded matrix respectively, thereby completing the encoding of all the data.

In this embodiment, step S123 is performed on all MR data in the packet, and a matrix is generated as the encoding result of the MR data of the packet. Assuming that the number of the packet cells after de-duplication is M and the number of the MRs is N, a sparse matrix of n×m will be generated finally.

Finally, step S121-step S124 are executed for each cell, and each packet will obtain a coded matrix, thus completing the coding of all data. Illustratively, the results of encoding the MR inputs in table 1 using the method described above are shown in fig. 2, where it is assumed that m=10.

The mapping relation from the cell to the column position in each group is stored as the coding rule of the group to form the coding rule for use in the on-line stage.

S13, training the classifier according to an incomplete supervised learning method. The method comprises the steps of setting all data in a selected service cell as a set A, setting the number as M, using a set { X ₁,X₂,...,X_M } to represent the data, setting the ith MR data as X _i, defining a label as y _i, setting the number of cells after de-duplication as N, using a set { c ₁,c₂,...,c_N } to represent the data, and forming an M (N+1) matrix, wherein the (N+1) th column represents a corresponding indoor or outdoor label.

In this embodiment, all data under a certain serving cell is defined as a set a, and the number of the data is assumed to be M, and the set { X ₁,X₂,...,X_M } is used, the ith MR is denoted as X _i, and the label is defined as y _i. The number of cells after deduplication is N, represented by the set { c ₁,c₂,...,c_N }. After passing through the labeling module and the data encoding module, an M (n+1) matrix is formed, wherein the n+1 th column represents the corresponding indoor and outdoor labels.

According to the bayesian theorem, it is possible to:

Obtainable P (y _i＝0|X_i)＝1-P(y_i＝1|X_i)

And constructing a classifier.

The construction classifier includes:

Initializing model parameters P (y _i＝1|X_i):

Wherein the performing loop iteration includes:

If |P _t+1(y_i＝1|X_i)-P_t(y_i＝1|X_i) | < T, a loop is performed:

For each data X _i in set M:

In this embodiment, for each data X _i in the set M, the present round P is updated (y _i)、P(c_t|y_i) according to the above formula and the last round result, and the present round P is calculated (y _i＝1|X_i) according to the updated result. In the iterative process, for the data of X _i e M, P (y _i＝1|X_i) =1 does not change with the iterative process, and remains constant all the time.

After all the data grouped by serving cells are subjected to the above procedure, each serving cell generates a set of corresponding P (c _k|y_i＝1)、P(c_k|y_i＝0)、P(y_i =1) and P (y _i =0), and stores them in table 3 to form a classifier.

Table 3 off-line stage classifier structure schematic diagram

S2, carrying out indoor and outdoor distinction on MR data through an online stage.

In this embodiment, the online stage performs indoor or outdoor identification on each MR which is not subjected to indoor or outdoor distinction, and the MR formats to be distinguished are shown in { c1 } -88, c2 } -93, c6: -102, c15 } -103}, and the serving cell is c1.

Step S2 includes steps S21-S23:

S21, finding out the corresponding coding rule of the service cell according to the selected service cell, converting the MR data into a corresponding vector form according to the coding rule, and if the selected cell in the MR data is not in the coding rule, not coding the cell.

In this embodiment, according to the serving cell, the coding rule corresponding to the serving cell is found out from the coding rule table generated in step S12, and the MR is converted into a corresponding vector form according to the coding rule. If a cell in the MR is not in the coding rule, the cell is not coded.

S22, find the corresponding P _i(c_k|y_i＝1)、P_i(c_k|y_i＝0)、P_i(y_i =1) and P _i(y_i =0) values according to the serving cell, and calculate the probability P (y _i＝1|X_i) that the current MR data is determined to be outdoor according to the calculation formula of P (y _i＝1|X_i);

In this embodiment, according to the serving cell, corresponding values of P _i(c_k|y_i＝1)、P_i(c_k|y_i＝0)、P_i(y_i =1) and P _i(y_i =0) are found from the classifier table generated in step S13. And according to the calculation formula of P (y _i＝1|X_i) in step S12, the probability P (y _i＝1|X_i) that the current MR is determined to be outdoor is calculated.

S23, if P (y _i＝1|X_i) >0.5, judging the MR data to be outdoor, otherwise judging the MR data to be indoor.

In summary, the invention provides an indoor and outdoor segmentation algorithm based on incomplete supervised learning, which uses only one label, has unknown other labels, has lower use cost compared with supervised learning, uses a known label in original data, and has higher precision theoretically compared with unsupervised learning. An off-line stage and an on-line stage which are extended by the algorithm, and a label marking method, a feature engineering and model training use and the like which are contained in the two stages.

Those of ordinary skill in the art will appreciate that all or a portion of the steps of the various methods of the above embodiments may be performed by instructions, or by instructions controlling associated hardware, which may be stored in a computer-readable storage medium and loaded and executed by a processor. To this end, an embodiment of the present invention provides a storage medium having stored therein a plurality of instructions capable of being loaded by a processor to perform the steps of any of the MR indoor and outdoor differentiation methods provided by the embodiment of the present invention.

The storage medium may include a Read Only Memory (ROM), a random access Memory (RAM, random Access Memory), a magnetic disk, an optical disk, or the like.

Because the instructions stored in the storage medium can execute the steps in any of the methods for dividing the interior and the exterior of the MR provided by the embodiments of the present invention, the beneficial effects that any of the methods for dividing the interior and the exterior of the MR provided by the embodiments of the present invention can be achieved, and detailed descriptions of the previous embodiments are omitted.

The foregoing description of the preferred embodiments of the invention is not intended to be limiting, but rather is intended to cover all modifications, equivalents, and alternatives falling within the spirit and principles of the invention.

Claims

1. A method for differentiating between MR cells, comprising:

Setting coding rules and forming a classifier through an offline stage;

the MR data are distinguished indoors and outdoors through an online stage;

Automatically labeling outdoor MR data;

Forming coding rules for use in the online stage;

training a classifier according to an incomplete supervised learning method;

the automatic labeling of MR data outside the room includes:

acquiring full MR data over a period of time;

Extracting MR data carrying GNSS data from the full MR data;

Labeling MR data carrying GNSS data as outdoor, and labeling MR data not carrying GNSS data as indoor;

The forming encoding rules for use in an online phase includes:

Grouping the full MR data according to the service cells;

Generating matrixes according to all MR data in the same group to be used as the coding result of the group MR data until each group respectively obtains a coded matrix, thereby finishing the coding of all data;

Forming a matrix of M x (n+1), wherein the n+1 th column represents the corresponding indoor or outdoor label;

according to the bayesian theorem, it is possible to:

Obtainable P (y _i＝0|X_i)＝1-P(y_i＝1|X_i)

And constructing a classifier.

2. The method of MR indoor and outdoor differentiation according to claim 1, wherein the constructing a classifier comprises:

Initializing model parameters P (y _i＝1|X_i):

3. The method of MR indoor and outdoor segmentation according to claim 2, wherein the performing loop iteration comprises:

If |P _t+1(y_i＝1|X_i)-P_t(y_i＝1|X_i) | < T, a loop is performed:

For each data X _i in set M:

4. The method of MR indoor and outdoor differentiation according to claim 3, wherein the performing indoor and outdoor differentiation of MR data by the online phase comprises:

5. The method according to claim 4, wherein the searching for the coding rule corresponding to the serving cell according to the selected serving cell, and converting the MR data into the corresponding vector form according to the coding rule, further comprises:

6. A computer readable storage medium, characterized in that the storage medium has stored therein a plurality of instructions adapted to be loaded by a processor to perform an MR indoor and outdoor differentiation method according to any one of claims 1 to 5.