CN113205113A

CN113205113A - Map matching method, and method and device for determining map matching model

Info

Publication number: CN113205113A
Application number: CN202110395970.6A
Authority: CN
Inventors: 狄烨; 李亚旭; 李洋
Original assignee: Beijing Didi Infinity Technology and Development Co Ltd
Current assignee: Beijing Didi Infinity Technology and Development Co Ltd
Priority date: 2021-04-13
Filing date: 2021-04-13
Publication date: 2021-08-03

Abstract

The embodiment of the invention discloses a map matching method, a map matching model determining method and a map matching model determining device, wherein candidate road sections are determined according to position points corresponding to target tasks, position data of the position points are input into the map matching model to be processed, selection parameters of the candidate road sections are obtained, and target road sections are determined according to the selection parameters of the candidate road sections, wherein the map matching model comprises a first sub-model and a second sub-model, the second sub-model is a student model obtained by training according to a knowledge distillation method, training sample data of a teacher model corresponding to the student model is marked and determined according to a data processing result of the first sub-model, so that a plurality of features in the position data can be subjected to fusion processing to improve the map matching accuracy, and the size of the map matching model can be reduced, the map matching efficiency is improved.

Description

Map matching method, and method and device for determining map matching model

Technical Field

The invention relates to the technical field of computers, in particular to a map matching method, a map matching model determining method and a map matching model determining device.

Background

Map Matching (Map Matching) technology uses an electronic Map and positioning information to determine the exact position of a vehicle on a road, the basic idea being to relate the vehicle's positioning track obtained by a positioning device to road information in an electronic Map database and thereby determine the vehicle's position relative to the Map. Currently, HMM (Hidden Markov Model) is generally used for map matching. When the existing HMM model is used for map matching, multiple features (such as distance, direction, speed, and the like) that affect the transition probability are independent from each other, and the influence that may be brought by the relevance of each feature is ignored, resulting in a low map matching accuracy.

Disclosure of Invention

In view of this, embodiments of the present invention provide a map matching method, a map matching model determining method, and a map matching model determining device, so as to perform fusion processing on multiple features in position data, so as to improve map matching accuracy, reduce the size of the map matching model, and improve map matching efficiency.

In a first aspect, an embodiment of the present invention provides a map matching method, where the method includes:

determining at least one candidate road section according to the position point corresponding to the target task;

inputting the position data of the position point into a first sub-model in a map matching model for processing, and acquiring first parameters corresponding to each candidate road section, wherein the position data comprises the association information of the position point and each candidate road section;

inputting the position data into a second sub-model of the map matching model for fitting processing to obtain second parameters corresponding to each candidate road section;

determining a selection parameter of each candidate road section according to the first parameter and the second parameter of each candidate road section;

determining a target road section according to the selection parameters of the candidate road sections;

and the second sub-model is a student model obtained by training according to a knowledge distillation method, and the training sample data of the teacher model corresponding to the student model is labeled and determined according to the data processing result of the first sub-model.

Optionally, the location data includes one or more of location point data, a distance between the location point and each candidate road segment, a speed of the location point, speed limit information of each candidate road segment, and included angle information between a motion direction of the location point and a direction of each candidate road segment.

Optionally, the teacher model is obtained by training through the following steps:

inputting the position data of each sample into the first sub-model, and determining the selection parameters of each candidate road section corresponding to each sample, wherein the position data of the sample at least comprises the correlation information of the position point of the sample and the corresponding candidate road section;

marking the candidate road section information with the selection parameter larger than a first threshold value as positive sample data of the teacher model;

marking the candidate road section information of which the selection parameter is smaller than a second threshold value as negative sample data of the teacher model;

and performing classification task training on the teacher model according to the positive sample data and the negative sample data to obtain a trained teacher model.

Optionally, the second sub-model is obtained by training through the following steps:

inputting the position data of each sample into the trained teacher model, and acquiring a second parameter of each candidate road section corresponding to each sample;

determining a training sample of the second submodel according to the position data of each sample and the second parameter of each candidate road section corresponding to each sample;

and performing regression task training on the second submodel according to the training sample of the second submodel to obtain the trained second submodel.

Optionally, determining the selection parameter of each candidate road segment according to the first parameter and the second parameter of each candidate road segment includes:

and calculating the product of the first parameter and the second parameter of the candidate road section and the selection parameter of the target road section corresponding to the previous position point, and determining the selection parameter of the candidate road section.

Optionally, the first sub-model is a hidden markov model, and the second sub-model is an XGBoost model; the first parameter is used for representing the transmission probability from the position point to the corresponding candidate road section, and the second parameter is used for representing the transition probability from the target road section corresponding to the previous position point to the corresponding candidate road section.

In a second aspect, an embodiment of the present invention provides a method for determining a map matching model, where the map matching model includes a first sub-model and a second sub-model, and the method includes:

determining training sample data of a teacher model according to the position data of each sample and the selection parameters of each candidate road section corresponding to each sample;

performing classification task training on the teacher model according to training sample data of the teacher model to obtain a trained teacher model;

inputting the position data of each sample into the trained teacher model, and acquiring second parameters of each candidate road section corresponding to each sample;

determining training sample data of the second submodel according to the position data of each sample and the second parameters of each candidate road section corresponding to each sample;

and performing regression task training on the second submodel according to the training sample data of the second submodel to obtain the trained second submodel so as to determine the map matching model.

Optionally, the training sample data of the teacher model includes positive sample data and negative sample data;

determining training sample data of a teacher model according to the position data of each sample and the selection parameters of each candidate road section corresponding to each sample comprises the following steps:

and marking the candidate road section information with the selection parameter smaller than a second threshold value as the negative sample data of the teacher model.

Optionally, the first sub-model is a hidden markov model, and the second sub-model is an XGBoost model.

In a third aspect, an embodiment of the present invention provides a map matching apparatus, where the apparatus includes:

the candidate road section determining unit is configured to determine at least one candidate road section according to the position point corresponding to the target task;

a first parameter obtaining unit, configured to input position data of the position point into a first sub-model in a map matching model for processing, and obtain a first parameter corresponding to each candidate road segment, where the position data includes association information of the position point and each candidate road segment;

the second parameter acquisition unit is configured to input the position data into a second sub-model of the map matching model for fitting processing, and acquire second parameters corresponding to the candidate road sections;

a selection parameter determination unit configured to determine a selection parameter of each of the candidate links according to the first parameter and the second parameter of each of the candidate links;

a target link determination unit configured to determine a target link according to the selection parameter of each of the candidate links;

In a fourth aspect, an embodiment of the present invention provides an apparatus for determining a map matching model, where the map matching model includes a first sub-model and a second sub-model, and the apparatus includes:

the first processing unit is configured to input position data of each sample to the first sub-model, and determine selection parameters of each candidate road section corresponding to each sample, wherein each sample comprises a plurality of position points in a plurality of historical tasks;

the first sample acquisition unit is configured to determine training sample data of a teacher model according to the position data of each sample and the selection parameters of each candidate road section corresponding to each sample;

the first training unit is configured to perform classification task training on the teacher model according to training sample data of the teacher model and acquire a trained teacher model;

the second processing unit is configured to input the position data of each sample into the trained teacher model, and obtain a second parameter of each candidate road section corresponding to each sample;

a second sample acquisition unit configured to determine training sample data of the second sub-model according to the position data of each sample and a second parameter of each candidate road section corresponding to each sample;

and the second training unit is configured to perform regression task training on the second submodel according to the training sample data of the second submodel, and obtain the trained second submodel to determine the map matching model.

In a fifth aspect, the present invention provides an electronic device, which includes a memory and a processor, wherein the memory is configured to store one or more computer program instructions, and wherein the one or more computer program instructions are executed by the processor to implement the method according to the first aspect of the present invention and/or the method according to the second aspect of the present invention.

In a sixth aspect, embodiments of the present invention provide a computer-readable storage medium on which computer program instructions are stored, which when executed by a processor, implement a method according to the first aspect of embodiments of the present invention and/or a method according to the second aspect of embodiments of the present invention.

In a seventh aspect, embodiments of the present invention provide a computer program product, which when run on a computer, causes the computer to perform the method according to the first aspect of embodiments of the present invention and/or the method according to the second aspect of embodiments of the present invention.

According to the embodiment of the invention, the candidate road sections are determined according to the position points corresponding to the target task, the position data of the position points are input into the map matching model for processing, the selection parameters of each candidate road section are obtained, the target road section is determined according to the selection parameters of each candidate road section, wherein the map matching model comprises a first sub-model and a second sub-model, the second sub-model is a student model obtained by training according to a knowledge distillation method, and the training sample data of the teacher model corresponding to the student model is marked and determined according to the data processing result of the first sub-model.

Drawings

The above and other objects, features and advantages of the present invention will become more apparent from the following description of the embodiments of the present invention with reference to the accompanying drawings, in which:

FIG. 1 is a flow chart of a method of determining a map matching model of an embodiment of the present invention;

FIG. 2 is a flow chart of a map matching method of an embodiment of the present invention;

FIG. 3 is a process diagram of a map matching method of an embodiment of the invention;

FIG. 4 is a schematic diagram of a map matching model determination apparatus according to an embodiment of the present invention;

FIG. 5 is a schematic diagram of a map matching apparatus according to an embodiment of the present invention;

fig. 6 is a schematic diagram of an electronic device of an embodiment of the invention.

Detailed Description

The present invention will be described below based on examples, but the present invention is not limited to only these examples. In the following detailed description of the present invention, certain specific details are set forth. It will be apparent to one skilled in the art that the present invention may be practiced without these specific details. Well-known methods, procedures, components and circuits have not been described in detail so as not to obscure the present invention.

Further, those of ordinary skill in the art will appreciate that the drawings provided herein are for illustrative purposes and are not necessarily drawn to scale.

Unless the context clearly requires otherwise, throughout the description, the words "comprise", "comprising", and the like are to be construed in an inclusive sense as opposed to an exclusive or exhaustive sense; that is, what is meant is "including, but not limited to".

In the description of the present invention, it is to be understood that the terms "first," "second," and the like are used for descriptive purposes only and are not to be construed as indicating or implying relative importance. In addition, in the description of the present invention, "a plurality" means two or more unless otherwise specified.

In a map matching application using an HMM model, a trajectory formed by GPS points can be used as an observation sequence in the HMM model, and a sequence of the acquired road segments can be used as a hidden sequence in the HMM model. The core of the HMM model is the emission probability and the transition probability, and then the Viterbi probability used for determining the target road section is obtained.

The emission probability may be multiplied by several impact factors: the distance from the GPS point to the link (distance influence factor), the difference in the direction of the GPS point traveling direction and the link (direction difference influence factor), the difference in the speed of the GPS point and the link speed limit value, and the like (speed difference influence factor). The influencing factors are as follows: (1) when the distance from the GPS point to the center of the road section is far, the distance influence factor of the emission probability is lower, otherwise, the distance influence factor is higher. (2) When the difference between the GPS point and the road section direction is large, the direction difference influence factor of the emission probability is low, otherwise, the direction difference influence factor is high. (3) When the speed difference between the GPS point speed and the highest speed limit of the road section speed is large, the speed difference influence factor of the emission probability is low, otherwise, the speed difference influence factor is high.

The transition probability refers to the probability of occurrence from a road segment corresponding to the previous point to a road segment corresponding to the previous point in the process of going from the previous point to the previous point. The factor factors for the transition probability are as follows: (1) the greater the distance between two GPS points and the distance difference between two projected points on the corresponding road segment (distance difference influence factor), the lower the transition probability, and otherwise, the higher the transition probability. (2) The larger the direction difference (direction difference influence factor) of the projection points corresponding to the two GPS points, the lower the transition probability, and otherwise, the higher the transition probability. (3) The greater the speed difference (speed difference influence factor) between two GPS points, the lower the transition probability, and vice versa.

When map matching is performed in an existing HMM model, transition probabilities corresponding to a plurality of features that affect the transition probability are usually multiplied, that is, several features are assumed to be independent from each other, which does not affect the calculation of the transition probabilities of other features. Therefore, the conventional map matching method cannot comprehensively exert the effect of each feature, and the accuracy of map matching is low. In addition, in order to further improve the feature processing effect, the size of the network model for the conventional multi-feature fitting processing is often large, so that the data processing efficiency is low, and the network model is not suitable for processing online real-time tasks. Therefore, the embodiment of the invention provides a method for determining a map matching model and a method for performing map matching based on the map matching model, so that the accuracy of map matching is improved by fusing a plurality of features in position data, and meanwhile, the size of the map matching model is reduced, and the map matching efficiency is improved.

Fig. 1 is a flowchart of a method of determining a map matching model according to an embodiment of the present invention. In this embodiment, the map matching model includes a first sub-model and a second sub-model, the first sub-model is configured to determine a first parameter according to the returned location data of the location point, the second sub-model is configured to determine a second parameter according to the returned location data of the location point, and the map matching module of the map matching model determines the target road segment corresponding to the location point according to the first parameter and the second parameter. As shown in fig. 1, the method for determining a map matching model of the present embodiment includes the following steps:

step S110, inputting the position data of each sample into the first sub-model, and determining the selection parameters of each candidate road segment corresponding to each sample. The position data of the sample at least comprises the association information of the position point of the sample and the corresponding candidate road section. The selection parameter is used to characterize the probability of the candidate road segment being selected. Optionally, the position data of the sample includes a position point (for example, a GPS point sequence) of the historical navigation task, a distance between the position point and the corresponding candidate road segment, speed information of the position point, speed limit information of the candidate road segment, direction included angle information between a motion direction of the position point and the corresponding candidate road segment, and the like.

In an alternative implementation, the first sub-model may be an existing HMM model for map matching, with parameters chosen to characterize the viterbi probabilities. Step S110 includes: and inputting the position data of the sample into the HMM model for processing, and determining the Viterbi probability, namely the selection parameter, of each candidate road section corresponding to the sample. In the HMM model, the emission probability and the transition probability of each candidate road section corresponding to a sample are determined according to the position data of the sample, and the Viterbi probability of each candidate road section is determined according to the emission probability, the transition probability and the Viterbi probability of a target road section corresponding to the previous position point of the sample. Optionally, in the map matching process, generally, the GNSS point sequence acquired by the positioning device is acquired, preprocessing such as deleting an abnormal point or supplementing a missing point is performed on the returned GNSS point sequence, and then map matching processing is performed according to the preprocessed GNSS point sequence. The target road sections corresponding to the adjacent GNSS points are definitely related, for example, the target road sections belong to the same road section or a road section which can be directly transferred, and the HMM model can perform map matching according to the adjacent GNSS points in the GNSS point sequence, so that the map matching efficiency is improved.

And step S120, determining training sample data of the teacher model according to the position data of each sample and the selection parameters of each candidate road section corresponding to each sample. In an alternative implementation, the training sample data of the teacher model includes positive sample data and negative sample data. Step S120 includes: and marking the candidate road section information with the selection parameter larger than the first threshold value as the positive sample data of the teacher model, and marking the candidate road section information with the selection parameter smaller than the second threshold value as the negative sample data of the teacher model. In other optional implementation manners, the information of the candidate road segment which has the selection parameter larger than the first parameter threshold and is adopted in the corresponding historical navigation task is determined as positive sample data, and the information of the candidate road segment which has the selection parameter smaller than the second parameter threshold and is not adopted is determined as negative sample data. The present embodiment does not limit the specific determination manner of the positive and negative samples. Optionally, the positive sample data includes information such as position data and transition probability of a position point corresponding to the candidate link, and the negative sample data includes information such as position data and transition probability of a position point corresponding to the candidate link. The position data of the position point comprises the distance between the position point and the corresponding candidate road section, the speed information of the position point, the speed limit information of the candidate road section, the direction included angle information between the motion direction of the position point and the corresponding candidate road section and the like.

And step S130, performing classification task training on the teacher model according to the training sample data of the teacher model, and acquiring the trained teacher model.

In an alternative implementation, the teacher model is an XGBoost model, which is a lifting tree model and integrates many tree models together to form a strong classifier. In the XGboost model, adding a tree each time is equivalent to learning a new function to fit the residual error predicted last time. In the embodiment, the XGBoost model is adopted, and the transition probability is determined by determining the influence of the plurality of features on the transition probability, so that the plurality of features determine the corresponding transition probability by fitting, rather than independently determining and then multiplying the determined plurality of features to obtain the corresponding transition probability, thereby effectively utilizing the plurality of features and improving the efficiency of map matching.

Optionally, the XGBoost model of this embodiment is trained through a training set composed of positive and negative sample data based on a predetermined loss function, so that the probability of the positive sample output by the XGBoost model approaches the transition probability corresponding to the positive sample, and the probability of the negative sample output approaches the transition probability corresponding to the negative sample. Therefore, in the embodiment, the probability output by the trained XGBoost model may be used as the transition probability of the corresponding candidate segment. Optionally, the present embodiment adopts cross entropy for classification as a loss function.

Optionally, the objective function of the XGBoost model is:

wherein,

characterizing the loss function, y_iThe transition probabilities in the training sample data are characterized,

characterization of prediction probability of XGboost model output, n characterization of sample number, Ω (f)_k) Characterizing the regularization term to prevent overfitting, f_kAnd characterizing the kth regression tree, and characterizing the number of regression trees in the XGboost model by K.

It should be understood that the HMM model and the XGBoost model are used as examples in the present embodiment, and other existing map matching models and classification models (e.g., GBDT model, SVM, etc.) may also be used as the first sub-model and the teacher model in the present embodiment, and the present embodiment is not limited thereto.

Step S140, inputting the position data of each sample to the trained teacher model, and obtaining the second parameter of each candidate road section corresponding to each sample. Optionally, the position data of each sample is input into the trained teacher model, and the transition probability of each candidate road section corresponding to each sample is obtained.

Step S150, determining training sample data of a second sub-model according to the position data of each sample and the second parameter of each candidate road section corresponding to each sample.

And step S160, performing regression task training on the second submodel according to the training sample data of the second submodel, and acquiring the trained second submodel to determine the map matching model. Optionally, the position data of each sample is input into the second submodel for processing, and the number of the regression trees and the parameters thereof in the second submodel are determined based on a predetermined loss function, so that for the same sample, the second parameters of each candidate road section output by the second submodel correspond to the second parameters of each candidate road section output by the teacher model. And the model size of the second sub-model is smaller than that of the teacher model. In an optional implementation manner, the teacher model is a complex XGBoost model with a large amount of parameter computation, and the second sub-model is a simple XGBoost model with a small amount of parameter computation as the student model.

A teacher-student network belongs to transfer learning and is a mode of model compression. Transfer learning is the migration of the performance of one model to another. In the teacher-student network, the teacher network is often a more complex network with very good performance and generalization capability, and the teacher network is used for guiding another simpler student network, so that a simpler student model with less parameter calculation amount can also have performance similar to that of the teacher network.

In this embodiment, position data of each sample is input to the first sub-model, selection parameters of each candidate road section corresponding to each sample are determined, training sample data of a teacher model is determined according to the position data of each sample and the selection parameters of each candidate road section corresponding to each sample, the teacher model is subjected to classification task training according to the training sample data of the teacher model to obtain a trained teacher model, the position data of each sample is input to the trained teacher model to obtain second parameters of each candidate road section corresponding to each sample, training sample data of a second sub-model is determined according to the position data of each sample and the second parameters of each candidate road section corresponding to each sample, the training sample data of the second sub-model is subjected to regression task training according to the training sample data of the second sub-model to obtain a trained second sub-model, to determine the map matching model. That is, in this embodiment, training sample data of the teacher model is determined by the first sub-model, and then the trained teacher model is obtained according to the training sample data, and the second sub-model, which is simple in training and has less parameter computation amount, is guided according to the trained teacher model to determine the map matching model. Therefore, the map matching model of the embodiment can determine the transition probability by determining the influence of the plurality of features on the transition probability, so that the plurality of features determine the corresponding transition probability by fitting, rather than independently determining and then multiplying the determined plurality of features to obtain the corresponding transition probability, therefore, the plurality of features can be effectively utilized, the map matching efficiency is improved, and meanwhile, the parameter operation amount of the map matching model can be reduced, the map matching efficiency is improved, so that the map matching model can be applied to online map matching.

Fig. 2 is a flowchart of a map matching method according to an embodiment of the present invention. As shown in fig. 2, the map matching method according to the embodiment of the present invention includes the steps of:

step S210, determining at least one candidate road section according to the position point corresponding to the target task. Optionally, the location point is a GNSS point returned by the device in real time.

In an optional implementation manner, the R-tree mapping table is retrieved according to the position information of the position point to determine the corresponding one or more candidate road segments. The R-tree is a spatial index data structure, and in this embodiment, an R-tree mapping table is first established for road segments in a road network, and surrounding road segments are captured according to the distance between a position point and a road segment area. It should be understood that the method of acquiring the candidate link is not limited by the present embodiment.

Step S220, inputting the position data of the position point into a first sub-model in the map matching model for processing, and obtaining a first parameter corresponding to each candidate road segment. The position data comprises the association information of the position point and each candidate road section. Optionally, the location data includes a distance between the location point and the corresponding candidate road segment, speed information of the location point, speed limit information of the candidate road segment, information of a direction included angle between the location point and the corresponding candidate road segment, and the like. In an alternative implementation manner, the first sub-model is an HMM model, and the first parameter is used to characterize the emission probability of the location point to the corresponding candidate road segment. The method comprises the steps of inputting position point information (such as longitude and latitude of a position point), the distance between the position point and a candidate road section, the speed of the position point, speed limit information of the candidate road section, included angle information of the motion direction of the position point and the direction of the candidate road section and the like into an HMM model, and determining corresponding emission probability.

Step S230, inputting the position data of the position point into a second sub-model of the map matching model for fitting, and obtaining a second parameter corresponding to each candidate road segment. In an optional implementation manner, the second sub-model is an XGBoost model, and the second parameter is used to characterize a transition probability from a target road segment corresponding to a previous location point to a corresponding candidate road segment. The information of the position points (such as the longitude and latitude of the position points and the like), the distance between the position points and the candidate road sections, the speed of the position points, the speed limit information of the candidate road sections, the included angle information between the motion direction of the position points and the direction of the candidate road sections and the like are input into the XGboost model to be subjected to fitting processing, and the corresponding transition probability is determined.

In step S240, a selection parameter of each candidate road segment is determined according to the first parameter and the second parameter of each candidate road segment. In an alternative implementation manner, the selection parameter of the candidate road segment is determined by calculating a product of the first parameter and the second parameter of the candidate road segment and the selection parameter of the target road segment corresponding to the previous location point. In other alternative implementations, the selection parameter of each candidate road segment may also be determined by calculating a product, or an addition, or a weighted sum, or an average of the first parameter and the second parameter, which is not limited in this embodiment.

And step S250, determining a target road section according to the selection parameters of the candidate road sections. In an optional implementation manner, the candidate road segments are sorted according to the size of the selection parameter, and the candidate road segment with the largest selection parameter is determined as the target road segment.

In this embodiment, the second sub-model in the map matching model is a student model obtained by training according to a knowledge distillation method, and the training sample data of the corresponding teacher model is determined by labeling according to the data processing result of the first sub-model.

In an alternative implementation, the teacher model of the second sub-model is obtained by training: inputting the position data of each sample into the first sub-model, determining the selection parameters of each candidate road section corresponding to each sample, marking the candidate road section information of which the selection parameters are greater than a first threshold as the positive sample data of the teacher model, marking the candidate road section information of which the selection parameters are less than a second threshold as the negative sample data of the teacher model, and performing classification task training on the teacher model according to the positive sample data and the negative sample data to obtain the trained teacher model. Optionally, the positive sample data includes information such as position data and transition probability of a position point corresponding to the candidate link, and the negative sample data includes information such as position data and transition probability of a position point corresponding to the candidate link. The position data of the position point comprises the distance between the position point and the corresponding candidate road section, the speed information of the position point, the speed limit information of the candidate road section, the direction included angle information between the motion direction of the position point and the corresponding candidate road section and the like.

In an alternative implementation, the second submodel is obtained by training: inputting the position data of each sample into the trained teacher model, obtaining second parameters of each candidate road section corresponding to each sample, determining a training sample of a second sub-model according to the position data of each sample and the second parameters of each candidate road section corresponding to each sample, and performing regression task training on the second sub-model according to the training sample of the second sub-model to obtain the trained second sub-model. Optionally, the method for determining the teacher model and the second sub-model in this embodiment is similar to the embodiment shown in fig. 1, and is not described herein again.

According to the method, the candidate road sections are determined according to the position points corresponding to the target tasks, the position data of the position points are input into the map matching model to be processed, the selection parameters of the candidate road sections are obtained, the target road sections are determined according to the selection parameters of the candidate road sections, the map matching model comprises a first sub model and a second sub model, the second sub model is a student model obtained through training according to a knowledge distillation method, and the training sample data of a teacher model corresponding to the student model is marked and determined according to the data processing result of the first sub model.

Fig. 3 is a process diagram of a map matching method according to an embodiment of the present invention. As shown in fig. 3, after receiving the returned position point P, the candidate link 31, the candidate link 32, and the candidate link 33 corresponding to the position point P are obtained according to the position information of the position point P. In the present embodiment, the distances from the candidate link 31, the candidate link 32, and the candidate link 33 are determined from the position information of the position point P, respectively, the moving direction and the moving speed of the position point P are determined from the position point P and the position point returned before it, and the angle information between the moving direction of the position point P and the link directions of the candidate links 31 to 33, respectively, and the speed limit information of the candidate links 31 to 33 are determined. Position data such as position information of the position point P, movement speed and movement direction information, distance information between the position point P and the candidate road sections 31-33, included angle information between the movement direction of the position point P and the road section directions of the candidate road sections 31-33, speed limit information of the candidate road sections 31-33 and the like are input into the map matching model 34.

In the present embodiment, the map matching model 34 includes a first sub-model 341, a second sub-model 342, and a link determination module 343. Optionally, the first sub-model 341 is an HMM model, and the second sub-model 342 is an XGBoost model. And the second sub-model is a student model obtained by training according to a knowledge distillation method, and the training sample data of the corresponding teacher model is labeled and determined according to the data processing result of the first sub-model. Therefore, the map matching model 34 of the present embodiment has a small parameter computation amount, can improve the map matching efficiency, and can be applied to online map matching.

The first sub-model 341 processes the input position data and outputs the emission probabilities p1-p3 corresponding to the candidate segments 31-33. The second sub-model 342 performs fitting processing on the input position data, and outputs transition probabilities p1'-p3' corresponding to the candidate links 31-33. The link determining module 343 calculates the product (or sum, or weighted sum, or average) of the emission probability p1 and the transition probability p1' of the candidate link 31, obtains the selection parameter s31 of the candidate link 31, calculates the product (or sum, or weighted sum, or average) of the emission probability p2 and the transition probability p2' of the candidate link 32, obtains the selection parameter s32 of the candidate link 32, calculates the product (or sum, or weighted sum, or average) of the emission probability p3 and the transition probability p3' of the candidate link 33, obtains the selection parameter s33 of the candidate link 33, compares the selection parameters 31-33, and determines the maximum selection parameter as the selection parameter 32. Thus, the map matching model 34 outputs the target link corresponding to the position point P as the candidate link 32.

Fig. 4 is a schematic diagram of a map matching model determination apparatus according to an embodiment of the present invention. As shown in fig. 4, the determination apparatus 4 of the map matching model of the present embodiment includes a first processing unit 41, a first sample acquisition unit 42, a first training unit 43, a second processing unit 44, a second sample acquisition unit 45, and a second training unit 46.

The first processing unit 41 is configured to input the position data of each sample to the first sub-model, and determine the selection parameter of each candidate road segment corresponding to each sample, where each sample includes a plurality of position points in a plurality of historical tasks. The first sample acquiring unit 42 is configured to determine training sample data of the teacher model according to the position data of each sample and the selection parameter of each candidate link corresponding to each sample. The first training unit 43 is configured to perform classification task training on the teacher model according to the training sample data of the teacher model, and obtain a trained teacher model.

In an optional implementation, the training sample data of the teacher model includes positive sample data and negative sample data. The first sample acquiring unit 42 comprises a positive sample acquiring sub-unit 421 and a negative sample acquiring sub-unit 422. The positive sample acquiring subunit 421 is configured to label the candidate road segment information of which the selection parameter is greater than the first threshold as the positive sample data of the teacher model. The negative sample acquiring subunit 422 is configured to label the candidate road segment information of which the selection parameter is smaller than the second threshold as the negative sample data of the teacher model.

The second processing unit 44 is configured to input the position data of each sample into the trained teacher model, and obtain a second parameter of each candidate segment corresponding to each sample. The second sample obtaining unit 45 is configured to determine training sample data of the second sub-model according to the position data of each sample and the second parameter of each candidate segment corresponding to each sample. The second training unit 46 is configured to perform regression task training on the second submodel according to the training sample data of the second submodel, and obtain a trained second submodel to determine the map matching model.

In an optional implementation manner, the first sub-model is a hidden markov model, and the second sub-model is an XGBoost model.

In the embodiment, training sample data of the teacher model is determined through the first sub-model, a trained teacher model is further obtained according to the training sample data, and a second sub-model which is simple in training and small in parameter calculation amount is guided according to the trained teacher model so as to determine the map matching model. Therefore, the map matching model of the embodiment can determine the transition probability by determining the influence of the plurality of features on the transition probability, so that the plurality of features determine the corresponding transition probability by fitting, rather than independently determining and then multiplying the determined plurality of features to obtain the corresponding transition probability, therefore, the plurality of features can be effectively utilized, the map matching efficiency is improved, and meanwhile, the parameter operation amount of the map matching model can be reduced, the map matching efficiency is improved, so that the map matching model can be applied to online map matching.

Fig. 5 is a schematic diagram of a map matching apparatus according to an embodiment of the present invention. As shown in fig. 5, the map matching device 5 of the present embodiment includes a candidate link determining unit 51, a first parameter acquiring unit 52, a second parameter acquiring unit 53, a selection parameter determining unit 54, and a target link determining unit 55.

The candidate road segment determining unit 51 is configured to determine at least one candidate road segment according to the position point corresponding to the target task. The first parameter obtaining unit 52 is configured to input the position data of the position point into a first sub-model in the map matching model for processing, and obtain a first parameter corresponding to each candidate link, where the position data includes association information of the position point and each candidate link. Optionally, the location data includes one or more of location point data, a distance between the location point and each candidate road segment, a speed of the location point, speed limit information of each candidate road segment, and included angle information between a motion direction of the location point and a direction of each candidate road segment.

The second parameter obtaining unit 53 is configured to input the position data into a second sub-model of the map matching model for fitting processing, and obtain a second parameter corresponding to each candidate link. The selection parameter determination unit 54 is configured to determine a selection parameter for each of the candidate road segments based on the first and second parameters of each of the candidate road segments. Optionally, the selection parameter determining unit 54 is further configured to calculate a product of the first parameter and the second parameter of the candidate road segment and the selection parameter of the target road segment corresponding to the previous location point, and determine the selection parameter of the candidate road segment. The target link determination unit 55 is configured to determine a target link according to the selection parameter of each of the candidate links.

And the second sub-model is a student model obtained by training according to a knowledge distillation method, and the training sample data of the teacher model corresponding to the student model is labeled and determined according to the data processing result of the first sub-model. Optionally, the first sub-model is a hidden markov model, and the second sub-model is an XGBoost model; the first parameter is used for representing the transmission probability from the position point to the corresponding candidate road section, and the second parameter is used for representing the transition probability from the target road section corresponding to the previous position point to the corresponding candidate road section.

In an alternative implementation, the map matching device 5 further includes a teacher model training unit 56 and a student model training unit 57.

The teacher model training unit 56 includes a first processing subunit 561, a positive sample determination subunit 562, a positive sample determination subunit 563, and a teacher model training subunit 564. The first processing subunit 561 is configured to input the position data of each sample to the first submodel, determine the selection parameter of each candidate road segment corresponding to each sample, and the position data of the sample at least includes the association information of the position point of the sample and the corresponding candidate road segment. The positive sample determination sub-unit 562 is configured to label candidate section information of which the selection parameter is greater than a first threshold as positive sample data of the teacher model. The positive sample determination subunit 563 is configured to label, as negative sample data of the teacher model, candidate link information whose selection parameter is smaller than a second threshold. The teacher model training subunit 564 is configured to perform classification task training on the teacher model according to the positive sample data and the negative sample data, and obtain a trained teacher model.

The student model training unit 57 includes a second processing subunit 571, a training sample determination subunit 572, and a student model training subunit 573. The second processing subunit 571 is configured to input the position data of each sample to the trained teacher model, and obtain a second parameter of each candidate road segment corresponding to each sample. The training sample determination subunit 572 is configured to determine a training sample of the second sub-model according to the position data of each of the samples and the second parameter of each of the candidate segments corresponding to each of the samples. The student model training sub-unit 573 is configured to perform regression task training on the second sub-model according to the training samples of the second sub-model, and obtain a trained second sub-model.

Fig. 6 is a schematic diagram of an electronic device of an embodiment of the invention. As shown in fig. 6, the electronic device 6 is a general-purpose data processing apparatus comprising a general-purpose computer hardware structure including at least a processor 61 and a memory 62. The processor 61 and the memory 62 are connected by a bus 63. The memory 62 is adapted to store instructions or programs executable by the processor 61. The processor 61 may be a stand-alone microprocessor or a collection of one or more microprocessors. Thus, the processor 61 implements the processing of data and the control of other devices by executing instructions stored by the memory 62 to perform the method flows of embodiments of the present invention as described above. The bus 63 connects the above components together, and also connects the above components to a display controller 64 and a display device and an input/output (I/O) device 65. Input/output (I/O) devices 65 may be a mouse, keyboard, modem, network interface, touch input device, motion sensing input device, printer, and other devices known in the art. Typically, the input/output device 65 is connected to the system through an input/output (I/O) controller 66.

As will be appreciated by one skilled in the art, embodiments of the present application may be provided as a method, apparatus (device) or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may employ a computer program product embodied on one or more computer-readable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.

The present application is described with reference to flowchart illustrations of methods, apparatus (devices) and computer program products according to embodiments of the application. It will be understood that each flow in the flow diagrams can be implemented by computer program instructions.

These computer program instructions may be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows.

These computer program instructions may also be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows.

Another embodiment of the invention is directed to a non-transitory storage medium storing a computer-readable program for causing a computer to perform some or all of the above-described method embodiments.

That is, as can be understood by those skilled in the art, all or part of the steps in the method for implementing the embodiments described above may be accomplished by specifying the relevant hardware through a program, where the program is stored in a storage medium and includes several instructions to enable a device (which may be a single chip, a chip, or the like) or a processor (processor) to execute all or part of the steps of the method described in the embodiments of the present application. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, and other various media capable of storing program codes.

The embodiment of the invention discloses a TS1 and a map matching method, which comprises the following steps:

TS2, the method according to TS1, where the location data includes one or more items of location point data, a distance between the location point and each of the candidate links, a speed of the location point, speed limit information of each of the candidate links, and angle information between a moving direction of the location point and a direction of each of the candidate links.

TS3, according to the method of TS1, the teacher model is trained by:

TS4, the method according to TS3, the second submodel being obtained by training:

TS5, the method of TS1, wherein determining the selection parameter for each of the candidate road segments according to the first and second parameters for each of the candidate road segments comprises:

TS6, according to any one of TS1-TS5, the first sub-model is a hidden Markov model, and the second sub-model is an XGboost model; the first parameter is used for representing the transmission probability from the position point to the corresponding candidate road section, and the second parameter is used for representing the transition probability from the target road section corresponding to the previous position point to the corresponding candidate road section.

The embodiment of the invention discloses a TS7 and a method for determining a map matching model, wherein the map matching model comprises a first submodel and a second submodel, and the method comprises the following steps:

TS8, according to the method of TS7, training sample data of the teacher model includes positive sample data and negative sample data;

TS9, the method according to TS7 or TS8, the first sub-model being a hidden Markov model and the second sub-model being an XGboost model.

The embodiment of the invention discloses TS10 and a map matching device, wherein the device comprises:

TS11, the device according to TS10, the position data includes one or more items of position point data, distance between the position point and each candidate road section, speed of the position point, speed limit information of each candidate road section, and included angle information between the motion direction of the position point and the direction of each candidate road section.

TS12, the apparatus of TS10, the apparatus comprising a teacher model training unit, the teacher model training unit comprising:

the first processing sub-unit is configured to input position data of each sample into the first sub-model, and determine selection parameters of each candidate road section corresponding to each sample, wherein the position data of the sample at least comprises association information of the position point of the sample and the corresponding candidate road section;

a positive sample determination subunit configured to label candidate road segment information of which the selection parameter is greater than a first threshold as positive sample data of the teacher model;

a positive sample determination subunit configured to label candidate road segment information of which the selection parameter is smaller than a second threshold as negative sample data of the teacher model;

and the teacher model training subunit is configured to perform classification task training on the teacher model according to the positive sample data and the negative sample data, and acquire a trained teacher model.

TS13, the apparatus of TS12, the apparatus comprising a student model training unit comprising:

the second processing subunit is configured to input the position data of each sample to the trained teacher model, and obtain a second parameter of each candidate road section corresponding to each sample;

a training sample determination subunit configured to determine a training sample of the second sub-model according to the position data of each sample and a second parameter of each candidate segment corresponding to each sample;

and the student model training subunit is configured to perform regression task training on the second submodel according to the training sample of the second submodel to obtain a trained second submodel.

TS14, the device according to TS10, the selection parameter determining unit further configured to calculate a product of the first parameter, the second parameter of the candidate road segment and the selection parameter of the target road segment corresponding to the previous location point, and determine the selection parameter of the candidate road segment.

TS15, according to any one of TS10-TS14, the first sub-model is a hidden Markov model, and the second sub-model is an XGboost model; the first parameter is used for representing the transmission probability from the position point to the corresponding candidate road section, and the second parameter is used for representing the transition probability from the target road section corresponding to the previous position point to the corresponding candidate road section.

The embodiment of the invention discloses a TS16 and a device for determining a map matching model, wherein the map matching model comprises a first sub-model and a second sub-model, and the device comprises:

TS17, the apparatus of TS16, the training sample data of the teacher model comprising positive and negative sample data;

the first sample acquiring unit includes:

a positive sample acquiring subunit configured to label candidate road segment information of which the selection parameter is greater than a first threshold as positive sample data of the teacher model;

a negative sample acquiring subunit configured to label the candidate road segment information of which the selection parameter is smaller than a second threshold as negative sample data of the teacher model.

TS18, the device according to TS16 or TS17, the first sub-model is a hidden Markov model, and the second sub-model is an XGboost model.

An embodiment of the invention discloses a TS19, an electronic device comprising a memory and a processor, wherein the memory is configured to store one or more computer program instructions, wherein the one or more computer program instructions are executed by the processor to implement a method as described in any one of TS1-TS 9.

An embodiment of the invention discloses a TS20, a computer readable storage medium having stored thereon computer program instructions which, when executed by a processor, implement a method as described in any of TS1-TS 9.

The embodiment of the invention discloses a TS21 and a computer program product, which when run on a computer causes the computer to execute the method as set forth in any one of TS1-TS 9.

The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention, and various modifications and changes may be made by those skilled in the art. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A map matching method, the method comprising:

2. The method of claim 1, wherein the location data comprises one or more of location point data, a distance between the location point and each of the candidate links, a speed of the location point, speed limit information of each of the candidate links, and angle information between a moving direction of the location point and a direction of each of the candidate links.

3. The method of claim 1, wherein the teacher model is trained by:

4. The method of claim 3, wherein the second submodel is obtained by training by:

5. The method of claim 1, wherein determining the selection parameter for each of the candidate road segments based on the first parameter and the second parameter for each of the candidate road segments comprises:

6. The method according to any of claims 1-5, wherein the first sub-model is a hidden Markov model and the second sub-model is an XGBoost model; the first parameter is used for representing the transmission probability from the position point to the corresponding candidate road section, and the second parameter is used for representing the transition probability from the target road section corresponding to the previous position point to the corresponding candidate road section.

7. A method of determining a map matching model, the map matching model comprising a first sub-model and a second sub-model, the method comprising:

8. An electronic device, comprising a memory and a processor, wherein the memory is configured to store one or more computer program instructions, wherein the one or more computer program instructions are executed by the processor to implement the method of any one of claims 1-7.

9. A computer-readable storage medium on which computer program instructions are stored, which computer program instructions, when executed by a processor, are to implement a method according to any one of claims 1-7.

10. A computer program product, characterized in that, when the computer program product is run on a computer, it causes the computer to perform the method according to any of claims 1-7.