A city GPS road network matching method based on hidden Markov model

Abstract

The application relates to the field of urban road information technology, in particular to a city GPS road network matching method and device based on a hidden Markov model, which comprises the following steps: by fusing the Gaussian distribution of GPS error values of GPS bus lines in different road regions and historical turning probabilities, the calculation of observation probability and state transition probability in the hidden Markov model is improved, so that the road matching accuracy of GPS data of a low sampling rate floating car is further improved, and more sufficient and accurate basis is provided for traffic planning and management of a city government.

Description

Technical Field

[0001] This invention relates to the field of urban road information technology, and more specifically, to an urban GPS road network matching method based on a hidden Markov model. Background Technology

[0002] In recent years, with the development and popularization of location-based services and mobile social networks, more and more applications, such as route finding, traffic monitoring, urban road planning, and geographic social networks, have begun to use data from GPS to achieve better service quality. However, due to the influence of external environmental conditions and the inherent accuracy limitations of GPS sensors, there are positioning and sampling errors. GPS data is difficult to pinpoint the actual location of a vehicle with perfect accuracy, and the deviation between GPS positioning data and the actual trajectory is unavoidable. Therefore, when processing and applying GPS data, it is necessary to first use appropriate estimation methods to perform road matching on the discrete GPS point set data with offsets, and then reconstruct the actual vehicle trajectory.

[0003] Traditional GPS road network matching methods include nearest path matching, shortest path matching, Euclidean distance matching, and dynamic time curvature matching. However, traditional GPS road network matching methods have large errors when processing low sampling rate data, resulting in inaccurate matching results. This is because the distance between GPS trajectory points in low sampling rate data is large, which can easily cause trajectory jumps, making the matching results inaccurate.

[0004] Traditional GPS road network matching methods exhibit significant errors when processing low-sampling-rate data, leading to inaccurate matching results. For low-sampling-rate data, the mainstream approach is currently based on Hidden Markov Models (HMMs) for GPS road network matching. However, all existing HMM-based GPS road network matching methods suffer from the following drawbacks:

[0005] 1. Method B for calculating observation probability and the candidate road selection radius ignore the differences in GPS error values ​​across different regions. In existing methods, observation probability is typically calculated based on the Gaussian distribution probability of the error distance of GPS points, with the standard deviation of the Gaussian distribution set to a fixed value. However, in urban environments with numerous high-rise buildings, complex urban landscapes such as tall, dense buildings, viaducts, and overpasses can lead to multipath effects. Furthermore, GPS signals are easily blocked when floating cars pass through tunnels or are surrounded by tall buildings, resulting in poor signal reception. Therefore, due to factors such as terrain and buildings, GPS accuracy varies across different regions, and a Gaussian distribution with a fixed standard deviation cannot accurately reflect the observation probability of different regions. This leads to inaccurate calculations of observation probability, thus affecting the accuracy of the matching results. Furthermore, existing methods select candidate matching roads for a GPS location point by considering all roads within a fixed radius centered on the location point. However, if this radius is set too large, some irrelevant roads in the road network will be included, increasing the computational burden on the algorithm. If the radius is set too small, the actual correct matching roads may be excluded. Therefore, a fixed-size candidate road selection radius cannot meet the needs of differentiated GPS data.

[0006] 2. The calculation method of state transition probability ignores the probability of turning from one road to another in actual statistical data. Existing methods typically calculate state transition probability based on the distance or direction between GPS track points, neglecting the probability of turning from the current road to another road in actual historical statistical data, given a predetermined path. Furthermore, in reality, the road network data held by researchers may not fully indicate whether each road is a one-way or two-way street, and may be assumed to be two-way. This leads to a situation where a state transition probability is calculated from road r1 to r2, but in reality, r1 cannot turn to r2. These issues all contribute to inaccurate calculations of state transition probability, thus affecting the accuracy of the matching results. Summary of the Invention

[0007] This invention provides a method and apparatus for matching urban GPS road networks based on a hidden Markov model, which at least solves the problem of low road matching accuracy in existing urban GPS road network matching methods.

[0008] According to an embodiment of the present invention, a method for matching urban GPS road networks based on a hidden Markov model is provided, comprising the following steps:

[0009] The road network data of urban bus routes and the GPS trajectory data of buses collected by the on-board GPS positioning devices of the buses corresponding to the urban bus routes are obtained. The GPS road network matching method of hidden Markov model is used to match the road of the corresponding bus route to the GPS trajectory data of the buses.

[0010] The city traffic road network data is acquired and divided into several road areas. The positioning error values ​​of bus GPS data in several road areas are statistically analyzed, and a Gaussian distribution of the GPS positioning error value of each road area is obtained based on the positioning error values ​​of bus GPS data.

[0011] Obtain the floating car GPS trajectory dataset collected from the floating car's onboard GPS positioning device, filter out the floating car GPS trajectory data with a high sampling rate from the floating car GPS trajectory dataset, and perform road network matching on the filtered floating car GPS trajectory data using the GPS road network matching method of Hidden Markov Model to obtain the historical turning probability of each intersection turning to another road when the previous road is determined.

[0012] The Gaussian distribution of GPS positioning error values ​​in each region and the historical turning probability of each intersection are incorporated into the selection of candidate roads, the calculation of observation probability, and the calculation of state transition probability in the GPS road network matching method of the Hidden Markov Model, so as to obtain the improved Hidden Markov Model after training.

[0013] The unknown floating car GPS trajectory data is matched with the trained Hidden Markov Model to obtain the road sequence corresponding to the floating car GPS trajectory data.

[0014] In one implementation, the historical turning probability is calculated as follows:

[0015] After matching each segment of floating car GPS trajectory data to the urban traffic network, the driving path is obtained. Based on each driving path, the probability p of turning to road r2 at the intersection after having traveled through road r1 is calculated, so as to obtain the historical turning probability of turning to another road at each intersection when the previous road is determined.

[0016] In one embodiment, the selection of candidate roads, calculation of observation probabilities, and calculation of state transition probabilities in the GPS road network matching method using Hidden Markov Models include:

[0017] When selecting candidate roads for GPS positioning points, the selection radius is set to the error distance that makes the cumulative distribution function value of the GPS Gaussian distribution in the road area where the GPS positioning point is located reach a preset value; where the GPS positioning point is the latitude and longitude coordinates of the current vehicle location recorded by the vehicle-mounted GPS device through satellite positioning at time t; the candidate roads are a series of roads around the GPS positioning point that need to be selected as candidates before matching the GPS positioning point to the urban traffic network roads, and this series of roads is called candidate roads;

[0018] When calculating the observation probability, the Euclidean distance between the candidate point and the GPS positioning point is first calculated, and the observation probability of the candidate point is set as the probability value corresponding to the Euclidean distance in the Gaussian distribution of the GPS positioning error value in the corresponding area; where the candidate point is the projection point of the GPS positioning point vertically projected onto the candidate road.

[0019] When calculating the state transition probability from one candidate road to another, historical turning probabilities are incorporated to improve the GPS road network matching method based on the Hidden Markov Model, thereby obtaining an improved Hidden Markov Model.

[0020] In one implementation, the Hidden Markov Model is specifically:

[0021] Obtain a candidate point set for each GPS location point. The candidate point set includes several candidate points.

[0022] Calculate the observation probability for each candidate point;

[0023] Calculate the state transition probabilities between candidate points and construct a hidden Markov chain;

[0024] After the Hidden Markov Chain is constructed, the Viterbi algorithm is used to find the optimal path in the candidate point set.

[0025] In one embodiment, urban traffic road network data is acquired, and the urban traffic road network data is divided into several road areas. The positioning error values ​​of bus GPS data in several road areas are statistically analyzed. The Gaussian distribution of the GPS positioning error value of each road area based on the positioning error values ​​of bus GPS data specifically includes:

[0026] Based on the acquired urban traffic network data, the urban traffic network data is divided into several road areas.

[0027] For each bus's historical GPS data, it is sorted by time and segmented by time interval. Then, the basic Hidden Markov Model road network matching method is used to match each segment of trajectory data with the corresponding bus route network.

[0028] Fit the probability density function of the Gaussian distribution of the GPS positioning error values ​​for each road area; the expression of the probability density function is:

[0029]

[0030] The corresponding cumulative distribution function is denoted as Where, d sz Let represent the Euclidean distance between candidate point s and GPS location point z, and gi represent the road area number. and Let be the mean and standard deviation of a Gaussian distribution.

[0031] In one embodiment, the Gaussian distribution of GPS positioning error values ​​for each region and the historical turning probability of each intersection are incorporated into the candidate road selection, observation probability calculation, and state transition probability calculation of the GPS road network matching method of the Hidden Markov Model, to obtain an improved trained Hidden Markov Model, including:

[0032] When obtaining the candidate point set for each GPS location point, the cumulative distribution function corresponding to the probability density function of the road area where the GPS location point is located will be... The error distance d is 0.96. sz As the radius of selection r;

[0033] Calculate each candidate point Observation probability When using the Gaussian distribution probability function of the GPS positioning error values ​​in the road area where the candidate point is located, To calculate the probability, d sz That is, to select candidate points To the location point z t Euclidean distance;

[0034] Calculate the state transition probability between candidate points When adding candidate points Road to candidate point Road Turning probability The improved expression for the state transition probability is:

[0035]

[0036] Where, d t→t+1 Represents the candidate point at time t Candidate points at time t+1 Euclidean distance, w t→t+1 Indicates from candidate points to candidate point The length of roads traversed in the urban transportation network.

[0037] A hidden Markov model-based urban GPS road network matching device includes:

[0038] Data matching module: used to acquire road network data of urban bus routes and bus GPS trajectory data collected by the on-board GPS positioning equipment of buses corresponding to urban bus routes, and to perform road matching of bus GPS trajectory data to corresponding bus routes using the GPS road network matching method of hidden Markov model.

[0039] Error statistics module: used to acquire urban traffic road network data, divide the urban traffic road network data into several road areas, count the positioning error values ​​of bus GPS data in several road areas, and obtain the Gaussian distribution of GPS positioning error values ​​for each road area based on the positioning error values ​​of bus GPS data.

[0040] Turning probability acquisition module: used to acquire floating car GPS trajectory dataset collected from the floating car's onboard GPS positioning device, filter floating car GPS trajectory data with high sampling rate from the floating car GPS trajectory dataset, and perform road network matching on the filtered floating car GPS trajectory data through the GPS road network matching method of Hidden Markov Model to obtain the historical turning probability of turning to another road at each intersection when the previous road is determined.

[0041] Model training module: This module incorporates the Gaussian distribution of GPS positioning error values ​​for each region and the historical turning probability of each intersection into the selection of candidate roads, the calculation of observation probabilities, and the calculation of state transition probabilities in the GPS road network matching method of the Hidden Markov Model, so as to obtain an improved trained Hidden Markov Model.

[0042] Road matching module: Used to match unknown floating car GPS trajectory data with the trained Hidden Markov Model to obtain the road sequence corresponding to the floating car GPS trajectory data.

[0043] In one embodiment, the model training module includes:

[0044] The candidate road selection unit is used to select candidate roads for GPS positioning points by setting the selection radius to the error distance that makes the cumulative distribution function value of the GPS Gaussian distribution in the road area where the GPS positioning point is located reach a preset value. Here, the GPS positioning point is the latitude and longitude coordinates of the current vehicle location recorded by the vehicle-mounted GPS device through satellite positioning at time t. The candidate roads are a series of roads around the GPS positioning point that need to be selected as candidates before matching the GPS positioning point to the urban traffic network.

[0045] The observation probability calculation unit is used to first calculate the Euclidean distance between the candidate point and the GPS positioning point when calculating the observation probability, and set the observation probability of the candidate point as the probability value corresponding to the Euclidean distance in the Gaussian distribution of the GPS positioning error value in the corresponding area. The candidate point is the projection point of the GPS positioning point vertically projected onto the candidate road.

[0046] The state transition probability calculation unit is used to improve the GPS road network matching method based on the hidden Markov model by incorporating historical turning probabilities when calculating the state transition probability from one candidate road to another, thereby obtaining an improved trained hidden Markov model.

[0047] A computer-readable medium, a computer-readable storage medium storing one or more programs, which can be executed by one or more processors to implement the steps in the hidden Markov model-based urban GPS road network matching method as described in any of the preceding claims.

[0048] A terminal device includes: a processor, a memory, and a communication bus; the memory stores a computer-readable program that can be executed by the processor;

[0049] The communication bus enables communication between the processor and memory;

[0050] When the processor executes a computer-readable program, it implements the steps in the hidden Markov model-based urban GPS road network matching method, as described above.

[0051] The urban GPS road network matching method based on the Hidden Markov Model in this embodiment of the invention improves the calculation of observation probability and state transition probability in the Hidden Markov Model by incorporating the Gaussian distribution of GPS error values ​​of GPS bus routes in different road areas and historical turning probabilities. This further improves the road matching accuracy of floating car driving GPS data with low sampling rate, and provides a more sufficient and accurate basis for urban government traffic planning and management. Attached Figure Description

[0052] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings:

[0053] Figure 1 This is a flowchart of the urban GPS road network matching method based on hidden Markov models according to the present invention;

[0054] Figure 2 This is a schematic diagram of the hidden Markov chain of the present invention;

[0055] Figure 3 This is a schematic diagram of the HMM road network matching model based on historical information-assisted decision-making according to the present invention.

[0056] Figure 4 This is a schematic diagram illustrating the principle of urban GPS road network matching based on the Hidden Markov Model of this invention.

[0057] Figure 5 This is a diagram of the terminal device of the present invention. Detailed Implementation

[0058] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0059] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0060] First, let's introduce the data terms and probabilities that will be used or may be used in this application:

[0061] Urban traffic network data: a set of traffic roads in the city for floating vehicles that can travel. Each road data includes: road ID, road shape and location geographic information data.

[0062] Bus route data: Road network data for each bus route in the city. The data for each road is in the same format as the road network data, and the road ID is the same as the corresponding road ID in the road network.

[0063] Bus GPS data: GPS data generated when each bus travels on a fixed route and is uploaded to the backend daily by the GPS device of each bus in the city. Each GPS positioning record includes six fields: vehicle ID, bus route ID, positioning longitude, positioning latitude, positioning time, up / down (1 represents up, 2 represents down).

[0064] Pre-matched urban floating car GPS data: The final hidden Markov model will use urban floating car GPS data for road network matching. Each GPS positioning record includes four fields: vehicle ID, positioning longitude, positioning latitude, and positioning time.

[0065] Historical GPS data for urban floating cars: This data is used to statistically analyze the historical turning probabilities of these floating cars on urban roads. The floating car type must be the same as or similar to the floating car type in the pre-matched data to obtain the same travel pattern distribution. Each GPS location record includes four fields: vehicle ID, location longitude, location latitude, and location time.

[0066] (2) Definition of basic concepts

[0067] Latitude and longitude coordinates: The latitude and longitude coordinates of a geographical location on the Earth's surface, which can be based on geographic coordinate systems such as WGS84 or GCJ-02. Represented as p = (lon, la).

[0068] GPS location point: The latitude and longitude coordinates of the vehicle's current location recorded by the vehicle's GPS device at time t via satellite positioning, denoted as z. t = (lon, la, t). Due to the influence of external objective conditions and the limitations of the accuracy of the equipment itself, the positioning point usually drifts.

[0069] A driving positioning trajectory: an ordered sequence of GPS positioning points generated by a GPS device when a vehicle is continuously driving over a period of time, represented as Z = (z1, z2, ..., z...). T ).

[0070] Candidate roads: Before performing GPS matching, a series of roads around the location point need to be selected as candidates. These roads are called candidate roads.

[0071] GPS candidate point: Location point z t The i-th candidate matching point is denoted as The latitude and longitude coordinates of a GPS positioning point projected vertically onto a candidate road are called candidate points. If the vertically projected point is not on the road, the nearest endpoint of the road is taken as the candidate point. GPS positioning point z t The candidate point set is represented as Candidate points The road sign is

[0072] GPS matching point: anchor point z t The candidate points finally matched by the matching algorithm are called matching points, denoted as m. tThe matching trajectory of a driving positioning trajectory is represented as M = (m1, m2, ..., m...). T ).

[0073] GPS error: The Euclidean distance between the positioning point and the actual location of the vehicle is called the GPS error.

[0074] Example 1

[0075] According to an embodiment of the present invention, a method for matching urban GPS road networks based on a hidden Markov model is provided, see [link to relevant documentation]. Figure 1 This includes the following steps:

[0076] S100: Obtain road network data of urban bus routes and bus GPS trajectory data collected by the on-board GPS positioning equipment of buses corresponding to urban bus routes, and perform road matching of bus GPS trajectory data to corresponding bus routes using the GPS road network matching method of hidden Markov model.

[0077] S200: Acquire urban traffic road network data, divide the urban traffic road network data into several road areas, statistically analyze the positioning error values ​​of bus route GPS data in several road areas, and obtain the Gaussian distribution of GPS positioning error values ​​for each road area based on the positioning error values ​​of bus GPS data.

[0078] S300: Obtain the floating car GPS trajectory dataset collected from the floating car's onboard GPS positioning device, filter out the floating car driving GPS trajectory data with a high sampling rate from the floating car GPS trajectory dataset, and perform road network matching on the filtered floating car driving GPS trajectory data using the GPS road network matching method of the Hidden Markov Model to obtain the historical turning probability of each intersection turning to another road when the previous road is determined.

[0079] S400: The Gaussian distribution of GPS positioning error values ​​in each region and the historical turning probability of each intersection are incorporated into the selection of candidate roads, the calculation of observation probability and the calculation of state transition probability in the GPS road network matching method of the Hidden Markov Model, so as to obtain the improved trained Hidden Markov Model.

[0080] S500: Match the unknown floating car GPS trajectory data with the trained Hidden Markov Model to obtain the road sequence corresponding to the floating car GPS trajectory data.

[0081] The urban GPS road network matching method based on the Hidden Markov Model in this embodiment of the invention improves the calculation of observation probability and state transition probability in the Hidden Markov Model by incorporating the Gaussian distribution of GPS error values ​​of GPS bus routes in different road areas and historical turning probabilities. This further improves the road matching accuracy of floating car driving GPS data with low sampling rate, and provides a more sufficient and accurate basis for urban government traffic planning and management.

[0082] This application provides urban traffic network data, urban bus route network data, urban bus GPS data, and urban floating car GPS data. The urban bus GPS data and urban floating car GPS data are historical data. The application pre-matches the GPS positioning trajectory Z = (z1, z2, ..., z...) for each segment of floating car travel. T Accurately match the vehicle to the city's traffic roads and obtain the matched vehicle trajectory M = (m1, m2, ..., m T This allows it to recreate the actual driving path.

[0083] Step S100 is as follows:

[0084] Prepare the road network data for each bus route in the city and the corresponding bus GPS data. Use a basic GPS road network matching method based on HMM (Hidden Markov Model) to match the data to the road. Since there is no interference from other roads, the bus GPS data can be matched to the corresponding road with relatively high accuracy. Therefore, the error distance of the bus GPS data on each bus route can be calculated.

[0085] Step S200 is as follows:

[0086] Prepare the city's road network data and divide it into multiple areas using a specific method. Calculate the GPS error distance of bus routes within each area and fit a Gaussian distribution of the GPS error values ​​of bus routes in each road area.

[0087] Obtaining the Gaussian distribution of GPS error values ​​for bus routes in various urban road areas based on historical GPS data includes the following steps:

[0088] Step 1: Divide the city into regions: Based on the acquired urban traffic network data, divide the urban traffic network data into several road regions.

[0089] Specifically, the maximum and minimum latitude and longitude coordinates of the urban road network are obtained, and then the urban road network is divided into multiple 100m×100m grid areas. Each grid is represented by the latitude and longitude coordinates of its four vertices. The area of ​​the urban transportation network is represented as G = (g1, g1, ..., g n).

[0090] Step 2, Bus GPS Road Network Matching: For the historical GPS data of each bus, sort it by time and segment it by time interval. Then, use the basic Hidden Markov Model road network matching method to match each segment of trajectory data with the corresponding bus route road network.

[0091] Specifically, the GPS data for each bus is sorted by time and segmented by time interval, ensuring that each positioning trajectory segment contains at least three positioning points with a time interval of at least 20 seconds between any two consecutive points. Then, a basic Hidden Markov Model (HMM) road network matching method is used to match each trajectory data segment against the corresponding bus route network. Since there is no interference from other roads, the data can be accurately matched to the corresponding road. The coordinates of each matching point and its error distance from the positioning point are recorded.

[0092] Step 3: Fit the GPS error distribution for each grid area: Fit the probability density function of the Gaussian distribution of the GPS positioning error values ​​for each road area.

[0093] Specifically, determine the city grid area g where the coordinates of each matching point are located. i The matching error distance is included in the error distribution statistics within the region. Finally, the probability density function of the Gaussian distribution of GPS error within each grid region is fitted based on the data. The expression of the probability density function is as follows:

[0094]

[0095] The corresponding cumulative distribution function is denoted as Where, d sz Let represent the Euclidean distance between candidate point s and GPS location point z, and gi represent the road area number. and Let be the mean and standard deviation of a Gaussian distribution.

[0096] Step S300 specifically includes:

[0097] Prepare a floating car GPS dataset of the city, and select data with high sampling rates. Use the Hidden Markov Model (HMM) GPS road network matching method to perform road network matching on the selected floating car GPS data. However, when matching GPS data from different areas, different Gaussian distribution probabilities calculated in the previous step are used. Furthermore, when selecting candidate roads for GPS positioning points, the error distance that maximizes the cumulative distribution function of the Gaussian distribution is used as the selection radius. After matching each segment of floating car GPS trajectory data onto the city's traffic road network, the driving road path is obtained. Based on each driving path, the road turning probability p(r2|r1) of turning to road r2 at the intersection after traveling on road r1 is calculated. This yields the historical turning probability of turning to other roads at each intersection given a previously determined path.

[0098] Based on historical GPS data from floating cars, the probability p of road turning at each intersection in the urban road network specifically includes:

[0099] Step 1: Floating car historical GPS road network matching:

[0100] For the GPS data of each floating car, it is sorted by time and segmented by time interval, ensuring that each positioning trajectory segment contains at least 3 positioning points (GPS positioning points) with a time interval of at least 10 seconds between any two consecutive points. Then, a basic Hidden Markov Model (HMM) road network matching method is used to perform road matching on the urban traffic road network data for each trajectory segment. However, when obtaining the candidate point set for positioning points, the grid area where the positioning point is located will be considered. The error distance d is 0.96. sz The radius r is used as the selection radius, and the grid region where the candidate point is located is used when calculating the observation probability of the candidate point. Probability calculations are performed. This yields the sequence of matching points and the corresponding driving path sequence r1r2...r for each segment of the positioning trajectory. n

[0101] Step 2: Calculate the probability of road turning:

[0102] Based on the road sequence of each travel path obtained above, the data from each intersection in the urban traffic road network data is statistically analyzed, starting from road r. i The probability of turning onto other roads, p(r) j |r i The total probability of a road turning into another road at an intersection is 1.

[0103] Step S400 is as follows:

[0104] Through the above steps S100-S300, the Gaussian distribution of the GPS error values ​​of bus routes in each road area and the historical turning probability of each intersection are initially obtained. The Gaussian distribution of the GPS positioning error values ​​of the buses and the historical turning probability are then incorporated into the candidate road selection, observation probability calculation and state transition probability calculation of the GPS road network matching method of HMM to obtain the improved trained Hidden Markov Model.

[0105] When selecting candidate roads for a GPS location point, the selection radius is set to the error distance that makes the cumulative distribution function value of the GPS Gaussian distribution in the road area where the GPS location point is located reach a preset value, which can be set to 96%. Here, the GPS location point is the latitude and longitude coordinates of the current vehicle location recorded by the vehicle-mounted GPS device at time t via satellite positioning; the candidate roads are a series of roads selected around the GPS location point as candidates before matching the GPS location point to the urban road network.

[0106] When calculating the observation probability, the Euclidean distance between the candidate point and the GPS positioning point is first calculated, and the observation probability of the candidate point is set as the probability value corresponding to the Euclidean distance in the Gaussian distribution of the GPS positioning error value in the corresponding area; where the candidate point is the projection point of the GPS positioning point vertically projected onto the candidate road.

[0107] When calculating the state transition probability from one candidate road to the next, it is necessary to incorporate the road's historical turning probability. By incorporating the historical turning probability, the GPS road network matching method based on HMM can be improved, thereby obtaining an improved trained Hidden Markov Model.

[0108] After completing the analysis and statistics of the historical data (including historical turning probabilities), historical information can be added to the basic HMM road network matching method to assist decision-making and construct an HMM road network matching model based on historical information-assisted decision-making. The HMM road network matching model based on historical data-assisted decision-making will now be explained in detail; see [link to relevant documentation]. Figure 3 :

[0109] Step 1: When obtaining the candidate point set for each positioning point, the Gaussian distribution of the GPS error value of the bus route in the road area where the positioning point is located will correspond to the cumulative distribution function. The error distance d is 0.96. sz Use the radius r as the selection radius.

[0110] Step 2: Calculate each candidate point Observation probability When using the GPS error Gaussian distribution probability function of the grid region where the candidate point is located, To calculate the probability, d szThat is, to select candidate points To the location point z t Euclidean distance.

[0111] Step 3: Calculate the state transition probabilities between candidate points. When adding the turning probability between the roads to which the candidate point belongs. The improved expression for the state transition probability is:

[0112]

[0113] Where, d t→t+1 Represents the candidate point at time t Candidate points at time t+1 Euclidean distance, w t→t+1 Indicates from candidate points to candidate point The length of roads traversed in the urban transportation network.

[0114] After the Hidden Markov Model is constructed, step S300 is used to calculate the road turning probability p of each intersection in the urban road network using the floating car historical GPS data. The new Hidden Markov Model is then used to perform road network matching of the urban floating car historical GPS data, thereby iterating to obtain a more accurate road turning probability. Finally, the iterated Hidden Markov Model is used to perform pre-matching of the urban floating car GPS data for road network matching.

[0115] Step S500 specifically includes:

[0116] The improved Hidden Markov Model (HMM) road network matching method is used to iterate step S400 again to obtain more accurate historical turning probabilities at intersections, resulting in the final HMM road network matching model.

[0117] GPS road network matching: Based on the trained HMM model, the unknown GPS trajectory data is matched to obtain the road sequence corresponding to the trajectory.

[0118] This invention can be applied to improve any HMM-based road network matching method. The following explanation uses the basic HMM road network matching method as an example. For a driving positioning trajectory Z = (z1, z2, ..., z...) T The specific steps for road matching using the basic HMM matching method are as follows: Step 1: Obtain the candidate point set for each GPS location point. The candidate point set includes several candidate points.

[0119] Specifically, for each positioning point z tThe algorithm retrieves all possible candidate roads from the urban road network, centered on the given location and within a radius of r (e.g., 100m). Candidate points are then calculated by vertically projecting the location point onto each candidate road. If the projected point is not on the road, the nearest endpoint of the road is selected as the candidate point. If a candidate point exceeds the radius r, it is filtered out. This process constructs a set of candidate points.

[0120] Step 2: Calculate the observation probability for each candidate point.

[0121] Specifically, for each candidate point Calculate its relationship with the positioning point z t Euclidean distance d sz The observation probability is calculated using the probability density function of the following Gaussian distribution.

[0122]

[0123] Where, d sz Indicate candidate points With GPS positioning point z t The Euclidean distance. μ and σ are the mean and standard deviation of the Gaussian distribution, both set to fixed values ​​(e.g., mean μ = 0, standard deviation σ = 20 meters).

[0124] Step 3: Calculate the state transition probabilities between candidate points and construct a hidden Markov chain.

[0125] Specifically, for each location point, calculate the state transition probability of each candidate point to the next candidate point of the location point at the next time step, and construct a hidden Markov chain by combining the calculated observation probabilities of the candidate points. See [link to documentation]. Figure 2 .

[0126] The state transition probability between candidate points is calculated using the following formula:

[0127]

[0128] This indicates how close the Euclidean distance between two location points is to the route distance between two candidate points (i.e., the route length from one point to another). The closer they are, the greater the probability of state transition.

[0129] Step 4: Constructing the Hidden Markov Chain. After the chain is constructed, the Viterbi algorithm is used to find the optimal path in the candidate point set.

[0130] Specifically, once the HMM chain is constructed, the optimal path can be found using the general Viterbi algorithm. The Viterbi algorithm decodes the HMM chain in the following four steps:

[0131] Initialization: For the first state S0 in the HMM chain, find the maximum probability of reaching each candidate point. Initialize it to its observation probability Right now

[0132] Recursion: For each subsequent time step t in the HMM chain, the state S... t To reach each of the candidate points maximum probability Transferred from the previous candidate point Multiply by the maximum probability The probability of observation itself, i.e. And remember the factor that makes max reach its maximum value. As The optimal state in the previous step The maximum probability of each candidate point in each step is calculated recursively.

[0133] Termination: For the last time step T in the HMM chain, choose S. T The candidate point with the highest arrival probability As the final state.

[0134] Backtracking: from the final state Begin by starting with the optimal state from the previous step. By tracing back to the initial state, the optimal path can be obtained.

[0135] The method of this invention improves the GPS road network matching method based on HMM to enhance matching accuracy.

[0136] First, the method of the present invention uses GPS data of buses with fixed routes that are spread throughout the city to calculate the error distance between the bus GPS data and the corresponding bus road, and statistically analyzes the Gaussian distribution of GPS positioning error values ​​in different areas of the city. Different Gaussian distribution variances are set for calculating the observation probability of candidate points in different road areas, so as to more accurately reflect the accuracy differences of GPS points in different areas, thereby improving the accuracy of the observation probability.

[0137] The error distance that makes the cumulative distribution function corresponding to the Gaussian distribution reach a larger value is used as the selection radius when selecting candidate roads in the region using GPS data, so as to select the candidate road set more accurately.

[0138] Secondly, the method of this invention analyzes historical GPS data to calculate the probability p(ri|r1r2...rn) of turning to other roads at intersections given a predetermined path. This turning probability is incorporated into the calculation of state transition probabilities, thus improving the calculation method. This allows for a more accurate reflection of the probability of turning from one road to another at intersections. Therefore, the method of this invention, by combining information from observed probabilities and state transition probabilities, considers the influence of GPS error values ​​and road turning probabilities in different areas, thereby improving the accuracy of GPS road network matching.

[0139] The key point of this invention is:

[0140] 1) A method is proposed to improve the selection of candidate roads and the calculation of candidate point observation probability in GPS road network matching based on HMM by fitting the Gaussian distribution of GPS error in different areas of the city using historical GPS data of buses with fixed routes throughout the city.

[0141] 2) A method is proposed to improve the calculation of state transition probability in HMM-based GPS road network matching by analyzing existing floating car historical GPS data in the city, calculating the probability of turning to other roads at each intersection under the previous driving path, and adding this turning probability to the calculation of road state transition probability.

[0142] This invention proposes an improved HMM-based urban floating car GPS road network matching method. This method uses historical GPS data from buses to fit a Gaussian distribution of GPS errors in different urban areas, which helps calculate the observation probability of GPS candidate points in different areas. The peak value of the Gaussian distribution is used as the selection radius for candidate roads for GPS positioning points within the area. The turning probability of other roads at each intersection is calculated using historical GPS data from floating cars, given the previous driving path. The road turning probability is added to the state transition probability. Finally, an improved HMM-based urban floating car GPS road network matching method is derived.

[0143] The beneficial effects of this application are as follows: Compared with existing HMM-based GPS road network matching methods, this invention first considers the differences in GPS positioning accuracy in different regions due to factors such as terrain or buildings. By statistically analyzing the GPS error distribution in different areas of the city, it improves the selection of candidate matching roads and the calculation of candidate point observation probabilities, reduces the complexity of the algorithm, and improves the matching accuracy in a more granular way. Secondly, this invention further improves the matching accuracy of HMM-based GPS road network matching methods by considering the historical turning probability of turning from one road to another in actual situations and incorporating it into the state transition probability calculation of the HMM model.

[0144] The present invention conducted experiments on the improved Hidden Markov Model (HMM) on a dataset generated from public transportation and taxi travel in Shenzhen. The experimental results show that the improved HMM proposed in this invention can effectively improve the matching accuracy of the basic HMM road network matching model.

[0145] Example 2

[0146] According to another embodiment of the present invention, an urban GPS road network matching device based on a hidden Markov model is provided, see [link to relevant documentation]. Figure 4 ,include:

[0147] Data matching module 100: used to acquire road network data of urban bus routes and bus GPS trajectory data collected by the on-board GPS positioning equipment of buses corresponding to urban bus routes, and to perform road matching of bus GPS trajectory data to corresponding bus routes using the GPS road network matching method of hidden Markov model.

[0148] Error statistics module 200: used to acquire urban traffic road network data, divide the urban traffic road network data into several road areas, count the positioning error values ​​of bus GPS data in several road areas, and obtain the Gaussian distribution of the bus route GPS positioning error values ​​for each road area based on the bus GPS data error values.

[0149] Turning probability acquisition module 300: used to acquire floating car GPS trajectory dataset collected from the floating car's onboard GPS positioning device, filter floating car GPS trajectory data with high sampling rate from the floating car GPS trajectory dataset, and perform road network matching on the filtered floating car GPS trajectory data through the GPS road network matching method of Hidden Markov Model to obtain the historical turning probability of turning to another road at each intersection when the previous road is determined.

[0150] Model training module 400: This module incorporates the Gaussian distribution of GPS error values ​​for bus routes in each road area and the historical turning probability of each intersection into the selection of candidate roads, the calculation of observation probability, and the calculation of state transition probability in the GPS road network matching method of the Hidden Markov Model, so as to obtain an improved Hidden Markov Model.

[0151] Road matching module 500: Used to match unknown floating car GPS trajectory data with the trained Hidden Markov Model to obtain the road sequence corresponding to the floating car GPS trajectory data.

[0152] The urban GPS road network matching method based on the Hidden Markov Model in this embodiment of the invention improves the calculation of observation probability and state transition probability in the Hidden Markov Model by incorporating the Gaussian distribution of GPS error values ​​of GPS bus routes in different road areas and historical turning probabilities. This further improves the road matching accuracy of floating car driving GPS data with low sampling rate, and provides a more sufficient and accurate basis for urban government traffic planning and management.

[0153] This application provides urban traffic network data, urban bus route network data, urban bus GPS data, and urban floating car GPS data. The urban bus GPS data and urban floating car GPS data are historical data. The application pre-matches the GPS positioning trajectory Z = (z1, z2, ..., z...) for each segment of floating car travel. T Accurately match the vehicle to the city's traffic roads and obtain the matched vehicle trajectory M = (m1, m2, ..., m T This allows it to recreate the actual driving path.

[0154] In a preferred technical solution, the model training module includes:

[0155] The candidate road selection unit is used to select candidate roads for GPS positioning points by setting the selection radius to the error distance that makes the cumulative distribution function value of the GPS Gaussian distribution in the road area where the GPS positioning point is located reach a preset value. Here, the GPS positioning point is the latitude and longitude coordinates of the current vehicle location recorded by the vehicle-mounted GPS device through satellite positioning at time t. The candidate roads are a series of roads around the GPS positioning point that need to be selected as candidates before matching the GPS positioning point to the urban traffic network.

[0156] The observation probability calculation unit is used to first calculate the Euclidean distance between the candidate point and the GPS positioning point when calculating the observation probability, and set the observation probability of the candidate point as the probability value corresponding to the Euclidean distance in the Gaussian distribution of the GPS positioning error value in the corresponding area. The candidate point is the projection point of the GPS positioning point vertically projected onto the candidate road.

[0157] The state transition probability calculation unit is used to improve the GPS road network matching method based on the hidden Markov model by incorporating historical turning probabilities when calculating the state transition probability from one candidate road to another, thereby obtaining an improved trained hidden Markov model.

[0158] After completing the analysis and statistics of the above historical data information (including historical turning probabilities), historical information can be added to the basic HMM road network matching method to assist decision-making and construct an HMM road network matching model based on historical information-assisted decision-making. For a detailed explanation of the HMM road network matching model based on historical data-assisted decision-making, please refer to step S400 above.

[0159] Example 3

[0160] Based on the above-described urban GPS road network matching method based on Hidden Markov Model, this embodiment provides a computer-readable storage medium storing one or more programs that can be executed by one or more processors to implement the steps in the urban GPS road network matching method based on Hidden Markov Model as described in the above embodiment.

[0161] Example 4

[0162] A terminal device includes: a processor, a memory, and a communication bus; the memory stores a computer-readable program that can be executed by the processor; the communication bus enables communication between the processor and the memory; when the processor executes the computer-readable program, it implements the steps in the above-mentioned urban GPS road network matching method based on a hidden Markov model.

[0163] Based on the aforementioned urban GPS road network matching method based on Hidden Markov Models, this application provides a terminal device, such as... Figure 5 As shown, it includes at least one processor 20; a display screen 21; and a memory 22, and may also include a communications interface 23 and a bus 24. The processor 20, display screen 21, memory 22, and communications interface 23 can communicate with each other via the bus 24. The display screen 21 is configured to display a preset user guide interface in the initial setup mode. The communications interface 23 can transmit information. The processor 20 can invoke logical instructions in the memory 22 to execute the methods described in the above embodiments.

[0164] Furthermore, the logical instructions in the aforementioned memory 22 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium.

[0165] The memory 22, as a computer-readable storage medium, can be configured to store software programs, computer-executable programs, such as program instructions or modules corresponding to the methods in the embodiments of this disclosure. The processor 20 executes functional applications and data processing by running the software programs, instructions, or modules stored in the memory 22, thereby implementing the methods in the above embodiments.

[0166] The memory 22 may include a program storage area and a data storage area. The program storage area may store the operating system and application programs required for at least one function; the data storage area may store data created based on the use of the terminal device. Furthermore, the memory 22 may include high-speed random access memory (RAM) and non-volatile memory. Examples include various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks, as well as transient storage media.

[0167] Furthermore, the specific process of loading and executing multiple instruction processors in the aforementioned storage medium and terminal device has been described in detail in the above method, and will not be repeated here.

[0168] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for matching urban GPS road networks based on hidden Markov models, characterized in that, Includes the following steps: S100: Obtain the road network data of the city bus routes and the bus GPS trajectory data collected by the on-board GPS positioning device of the bus corresponding to the city bus routes, and perform road matching of the bus GPS trajectory data to the corresponding bus routes using the GPS road network matching method of Hidden Markov Model. S200: Acquire urban traffic road network data, divide the urban traffic road network data into several road areas, count the positioning error values ​​of bus GPS data in several road areas, and obtain a Gaussian distribution of the GPS positioning error value of each road area based on the positioning error values ​​of the bus GPS data. S300: Obtain the floating car GPS trajectory dataset collected from the floating car's onboard GPS positioning device, filter out the floating car GPS trajectory data with a high sampling rate from the floating car GPS trajectory dataset, and perform road network matching on the filtered floating car GPS trajectory data using the GPS road network matching method of the hidden Markov model to obtain the historical turning probability of turning to another road at each intersection when the previous road is determined. S400: The Gaussian distribution of the GPS positioning error value of each region and the historical turning probability of each intersection are incorporated into the selection of candidate roads, the calculation of observation probability and the calculation of state transition probability in the GPS road network matching method of the Hidden Markov Model, so as to obtain the Hidden Markov Model after improved training. S500: Match the unknown floating car GPS trajectory data with the trained Hidden Markov Model to obtain the road sequence corresponding to the floating car GPS trajectory data.

2. The urban GPS road network matching method based on a hidden Markov model according to claim 1, characterized in that, The calculation method for the historical turning probability is as follows: After matching each segment of the floating car's GPS trajectory data to the urban traffic network, a driving path is obtained. Based on each segment of the driving path, the probability p of turning to road r2 at the intersection after having traveled through road r1 is calculated, so as to obtain the historical turning probability of turning to another road at each intersection when the previous road is determined.

3. The urban GPS road network matching method based on a hidden Markov model according to claim 1, characterized in that, The selection of candidate roads, calculation of observation probabilities, and calculation of state transition probabilities in the GPS road network matching method based on the Hidden Markov Model include: When selecting candidate roads for a GPS positioning point, the selection radius is set to the error distance that makes the cumulative distribution function value of the GPS Gaussian distribution in the road area where the GPS positioning point is located reach a preset value; wherein, the GPS positioning point is the latitude and longitude coordinates of the current vehicle location recorded by the vehicle-mounted GPS device through satellite positioning at time t; the candidate roads are a series of roads around the GPS positioning point that need to be selected as candidates before matching the GPS positioning point to the urban traffic network roads, and this series of roads is called candidate roads; When calculating the observation probability, the Euclidean distance between the candidate point and the GPS positioning point is first calculated, and the observation probability of the candidate point is set as the probability value corresponding to the Euclidean distance in the Gaussian distribution of the GPS positioning error value in the corresponding area; wherein, the candidate point is the projection point of the GPS positioning point vertically projected onto the candidate road; When calculating the state transition probability of moving from one candidate road to another, the historical turning probability is incorporated to improve the GPS road network matching method based on the Hidden Markov Model, thereby obtaining the improved Hidden Markov Model.

4. The urban GPS road network matching method based on a hidden Markov model according to claim 3, characterized in that, The Hidden Markov Model is specifically as follows: Obtain a candidate point set for each GPS positioning point, wherein the candidate point set includes several candidate points; Calculate the observation probability for each candidate point; Calculate the state transition probabilities between the candidate points and construct a hidden Markov chain; After the Hidden Markov Chain is constructed, the Viterbi algorithm is used to solve for the optimal path in the candidate point set.

5. The urban GPS road network matching method based on a hidden Markov model according to claim 3, characterized in that, The process of acquiring urban traffic road network data, dividing the urban traffic road network data into several road areas, statistically analyzing the positioning error values ​​of bus GPS data within several road areas, and obtaining a Gaussian distribution of the GPS positioning error value for each road area based on the positioning error values ​​of the bus GPS data specifically includes: Based on the acquired urban traffic network data, the urban traffic network data is divided into several road areas; For each bus's historical GPS data, it is sorted by time and segmented by time interval. Then, the basic Hidden Markov Model road network matching method is used to match each segment of trajectory data with the corresponding bus route network. The probability density function is fitted to the Gaussian distribution of the GPS positioning error values ​​for each road area; the expression for the probability density function is: The corresponding cumulative distribution function is denoted as Where, d sz Let represent the Euclidean distance between candidate point s and the GPS positioning point z, and let gi represent the number of the road region. and Let be the mean and standard deviation of a Gaussian distribution.

6. The urban GPS road network matching method based on a hidden Markov model according to claim 5, characterized in that, The step of incorporating the Gaussian distribution of the GPS positioning error values ​​of each region and the historical turning probability of each intersection into the candidate road selection, observation probability calculation, and state transition probability calculation of the GPS road network matching method of the Hidden Markov Model to obtain the improved trained Hidden Markov Model includes: When obtaining the candidate point set for each GPS positioning point, the cumulative distribution function corresponding to the probability density function of the road area where the GPS positioning point is located will be... The error distance d is 0.

96. sz As the radius of selection r; Calculate each candidate point Observation probability When using the Gaussian distribution probability function of the GPS positioning error values ​​in the road area where the candidate point is located, To calculate the probability, d sz That is, to select candidate points To the location point z t Euclidean distance; Calculate the state transition probability between candidate points When adding candidate points Road to candidate point Road Turning probability The improved expression for the state transition probability is: Where, d t→t+1 Represents the candidate point at time t Candidate points at time t+1 Euclidean distance, w t→t+1 Indicates from candidate points to candidate point The length of roads traversed in the urban transportation network.

7. A city GPS road network matching device based on a hidden Markov model, characterized in that, include: Data matching module: used to acquire road network data of urban bus routes and bus GPS trajectory data collected by the on-board GPS positioning equipment of the buses corresponding to the urban bus routes, and to perform road matching of the bus GPS trajectory data to the corresponding bus routes using the GPS road network matching method of Hidden Markov Model. Error statistics module: used to acquire urban traffic road network data, divide the urban traffic road network data into several road areas, count the positioning error values ​​of bus GPS data in several road areas, and obtain a Gaussian distribution of the GPS positioning error value of each road area based on the positioning error values ​​of the bus GPS data. Turning probability acquisition module: used to acquire floating car GPS trajectory dataset collected from the floating car's onboard GPS positioning device, filter floating car GPS trajectory data with high sampling rate from the floating car GPS trajectory dataset, and perform road network matching on the filtered floating car GPS trajectory data through the GPS road network matching method of the hidden Markov model to obtain the historical turning probability of turning to another road at each intersection when the previous road is determined. Model training module: used to incorporate the Gaussian distribution of the GPS positioning error value of each region and the historical turning probability of each intersection into the selection of candidate roads, the calculation of observation probability and the calculation of state transition probability of the GPS road network matching method of the hidden Markov model, so as to obtain the improved trained hidden Markov model. Road matching module: used to match unknown floating car GPS trajectory data with the trained Hidden Markov Model to obtain the road sequence corresponding to the floating car GPS trajectory data.

8. The urban GPS road network matching device based on a hidden Markov model according to claim 7, characterized in that, The model training module includes: The candidate road selection unit is used to set the selection radius to an error distance such that the cumulative distribution function value of the GPS Gaussian distribution in the road area where the GPS positioning point is located reaches a preset value when selecting candidate roads for the GPS positioning point; wherein, the GPS positioning point is the latitude and longitude coordinates of the current vehicle location recorded by the vehicle-mounted GPS device through satellite positioning at time t; the candidate roads are a series of roads around the GPS positioning point that need to be selected as candidates before matching the GPS positioning point to the urban traffic network roads, and this series of roads is called candidate roads; The observation probability calculation unit is used to calculate the Euclidean distance between the candidate point and the GPS positioning point when calculating the observation probability, and set the observation probability of the candidate point as the probability value corresponding to the Euclidean distance in the Gaussian distribution of the GPS positioning error value in the corresponding area, wherein the candidate point is the projection point of the GPS positioning point vertically projected onto the candidate road. The state transition probability calculation unit is used to improve the GPS road network matching method based on the hidden Markov model by incorporating the historical turning probability when calculating the state transition probability from one candidate road to another, thereby obtaining the improved trained hidden Markov model.

9. A computer-readable medium, characterized in that, The computer-readable storage medium stores one or more programs, which can be executed by one or more processors to implement the steps in the urban GPS road network matching method based on a hidden Markov model as described in any one of claims 1-6.

10. A terminal device, characterized in that, include: Processor, memory, and communication bus; The memory stores a computer-readable program that can be executed by the processor; The communication bus enables communication between the processor and the memory; When the processor executes the computer-readable program, it implements the steps of the urban GPS road network matching method based on a hidden Markov model as described in any one of claims 1-6.

Images