Vehicle travel path identification method in urban dense road network environment

Abstract

The present invention discloses a method for identifying vehicle travel paths in a dense urban road network environment, comprising the following steps: step one, collecting mobile phone signaling data, constructing a target urban road network, and unifying the coordinate system; step two, pre-processing the signaling data collected in step one, eliminating the signaling data generated by communication failure, data duplication, field missing, ping-pong switching, and signal drift; step three, smoothing the signaling data based on spatiotemporal clustering and Gaussian filtering, and constructing a complete individual travel signaling trajectory; step four, fitting the travel path based on a non-uniform rational B-spline curve; step five, matching the travel path; step six, repairing the travel path, and completing the travel path identification. The present invention ultimately achieves a technical breakthrough in the identification of travel paths in a dense urban road network environment based on mobile phone signaling data, which can help with tracing the source of congestion, tracking epidemic patients and virus transmission paths, and also provides a basis for the construction of new roads and road reconstruction and expansion in cities.

Description

Technical Field

[0001] The present invention belongs to the field of signaling data information recognition of traffic big data, and specifically relates to a vehicle travel path recognition method in an urban dense road network environment based on mobile phone signaling data, which is mainly used for motorized travel road recognition. Background Art

[0002] The vehicle travel path is the basic information of an individual's complete travel chain. It is of great significance for congestion tracing, traffic prediction model improvement, especially the improvement of non-aggregate prediction model, and provides help for tracking epidemic patients and virus transmission paths. By accumulating the arrival, departure and transit traffic volume of key roads, it also provides a basis for the construction of new roads and road reconstruction and expansion in the city, and ensures the scientific rationality of government decision-making and planning and construction. However, the existing survey methods have problems such as high cost, difficulty in implementation, poor dynamic update, and many subjective memories, which are difficult to meet actual needs. At the same time, existing data collection methods, such as coils, RFID, and smart videos, can only estimate travel paths through road section traffic, which is costly and susceptible to external interference; GPS data can effectively identify traffic travel paths, but it requires the cooperation of a large number of residents and is difficult to implement on a large scale and for a long time. A new method is urgently needed to make up for the shortcomings of existing survey methods.

[0003] Signaling data is anonymous and passively collected, which ensures the advantages of easy acquisition, low cost, wide coverage, and strong dynamism, overcoming the difficulties of individual privacy leakage and large-scale, long-term real-time. However, the positioning accuracy of signaling data can only reach the level of communication base stations, and is mostly used to observe population flows at the granularity of traffic communities and above. There is a large gap in the field of individual fine travel feature identification. In travel path identification, existing methods mostly use signaling data to replace travel trajectories, or identify sparse road networks, and cannot identify individual travel paths in complex environments.

[0004] Therefore, based on mobile phone signaling data, the present invention constructs a technology for extracting urban motorized travel trajectories in a dense road network environment. First, the spatiotemporal clustering algorithm, Gaussian filtering and non-uniform rational B-spline curves are used to achieve refined fitting of travel paths, and then the modified probability matching algorithm and the repair algorithm considering the direction angle and the road network topology are used to identify travel roads. The present invention can provide support for real-time detection and dynamic control of urban road traffic conditions, and is of great significance to the refined management of urban traffic at this stage. Summary of the invention

[0005] In order to overcome the above-mentioned shortcomings of the prior art, the present invention proposes a vehicle travel path identification method in an urban dense road network environment based on mobile phone signaling data, aiming to solve the problem of individual motorized travel path identification in a dense road network environment. The present invention first constructs an urban road network, collects and preprocesses individual signaling data, and then uses a spatiotemporal clustering algorithm, Gaussian filtering and non-uniform rational B-spline curves to finely construct a travel path fitting curve, and then uses an improved probability matching algorithm to match the nodes of the fitting curve to the actual road network, and finally repairs the identification result according to the direction angle and road network topology characteristics to complete the travel path identification.

[0006] The technical solution adopted by the present invention to solve the technical problem is: a method for identifying vehicle travel paths in a dense urban road network environment, comprising the following steps:

[0007] Step 1: Collect mobile phone signaling data, build the target city road network, and unify the coordinate system;

[0008] Step 2: pre-process the signaling data collected in step 1 to remove signaling data generated by communication failure, data duplication, field missing, ping-pong switching and signal drift;

[0009] Step 3: Based on the spatiotemporal clustering algorithm and Gaussian filtering to smooth the signaling data, a complete individual signaling trajectory is constructed;

[0010] Step 4: Fitting the travel path based on the non-uniform rational B-spline curve;

[0011] Step 5: Travel path matching;

[0012] Step 6: Repair the travel path and complete the travel path identification.

[0013] Compared with the prior art, the present invention has the following positive effects:

[0014] In view of the defects of current travel path identification, which directly uses signaling data as travel trajectory or can only identify travel paths of high-speed roads in sparse road network environments, the present invention proposes a travel path identification method in a dense road network environment. In the path fitting module, according to the distribution characteristics of signaling data, the spatiotemporal clustering algorithm and Gaussian filtering of signaling data smoothing are integrated, and the travel trajectory is fitted using non-uniform rational B-spline curves according to the arbitrary characteristics of the shape of the travel path; in the path identification module, the existing probability matching algorithm is corrected according to the road network distribution to achieve the corresponding road matching of the fitting trajectory nodes, and the positional relationship between the fitting nodes and the road network is considered. The direction angle and the road network topology structure are integrated to correct the travel path identification results; finally, the travel path in a dense road network environment is identified. Specific advantages include:

[0015] 1. Signaling data smoothing based on spatiotemporal clustering algorithm and Gaussian filtering

[0016] The density of base stations varies in different areas, and most base stations are located in buildings on both sides of the road, causing the signaling data to show a fluctuating distribution relative to the travel section. Especially when the vehicle turns around or stops, the mobile phone terminal will interact repeatedly with the surrounding base stations, which can easily cause misidentification.

[0017] To this end, the present invention first uses a spatiotemporal clustering algorithm to cluster the signaling data according to its spatiotemporal characteristics, then finds out the data that has not been clustered, and uses it as a basis to divide the individual signaling data into segments, and then uses Gaussian filtering for each segment of the signaling data to gradually perform weighted averaging of the signaling position information to achieve data smoothing. Finally, the smoothed and unsmoothed signaling data are recombined to construct complete individual signaling data.

[0018] 2. Travel path fitting based on non-uniform rational B-spline curve

[0019] Actual travel varies greatly with the urban road network structure, individual attributes and preferences, and travel modes, presenting travel paths of arbitrary shapes. To this end, the present invention uses a non-uniform rational B-spline curve, takes the smoothed signaling data as the control point, gradually fits the actual travel trajectory, and extracts all node coordinates (x i ,y i ), providing a basis for travel path identification. The non-uniform rational B-spline curve has outstanding advantages in local expression, and further reduces the impact of signaling data fluctuations on path fitting.

[0020] 3. Modify the probability matching algorithm for path matching

[0021] The probability matching algorithm is a common method for path identification, which achieves travel path matching by analyzing all roads within a fixed threshold range of trajectory points. However, in actual travel, the uneven distribution of base stations makes the number of roads within the fixed distance threshold vary too much, and some discrete points cannot be matched or match too many roads, resulting in large errors in the final matching results. In response to this, the present invention makes the following two-step corrections to the probability matching algorithm, and uses the fitting nodes of non-uniform rational B-splines as input for matching.

[0022] 1) Modify the fixed distance threshold to the road number threshold m to ensure that each fitting node has a stable number of matching roads.

[0023] 2) Whether there is a common matching road is used as the basis for selecting statistical objects. When there is no common matching road between the current e and e+1 fitting nodes, the first e fitting nodes are used as statistical objects to realize automatic node splitting, and then find the matching road with the highest likelihood probability for this group of fitting nodes. The denser the fitting nodes, the smaller the misidentification caused by splitting deviation.

[0024] 4. Travel path identification and correction considering direction angle characteristics and road network topology

[0025] In areas with dense and complex road networks, the road sections around the actual travel path are prone to intersect with the fitting curve, resulting in misidentification of the intersecting road sections. At the same time, since each fitting node will perform road matching, a large number of matching results will be repeated. In this regard, the present invention proposes a repair method that considers the direction angle and road network topology, which includes the following five steps:

[0026] 1) Eliminate duplicate data and mutation data. First, remove the continuous and repeated recognition results. Then, for any continuous road section with the same road number, if there is a small amount of data with other road numbers, remove it;

[0027] 2) For continuous recognition results with the same road number, the discontinuous part of the road section number is supplemented;

[0028] 3) Calculate the direction angles of all identified road sections and fitting nodes. For a road section and the corresponding fitting node, if the ratio of the angle less than the angle threshold is less than the set probability, or the number of fitting nodes is less than the discrimination number threshold, remove the road section;

[0029] 4) For continuous recognition results with the same road number, re-supplement the discontinuous part of the road section number;

[0030] 5) If the distance between two consecutive identified sections with different road numbers is greater than the distance threshold, the remaining sections under the two road numbers are supplemented in sequence until there is a minimum distance between the supplemented sections or they intersect.

[0031] The advantages of the present invention are: in terms of data collection, the signaling data is stored by the mobile operator and is relatively easy to obtain. At the same time, the mobile phone signaling data has the characteristics of wide coverage, strong continuity, passive collection, and low cost. Compared with other data types, it is more suitable for extracting travel characteristics such as individual trajectories. In terms of travel path fitting, the spatial position relationship between the signaling data and the actual travel path is deeply considered, and the spatiotemporal clustering algorithm and Gaussian filtering are combined for smoothing for the first time; and considering the differences in individual travel trajectories, non-uniform rational B-splines are selected for path fitting to further reduce the impact of signaling data fluctuations on path fitting. In terms of path identification, the distribution characteristics of the signaling data are deeply analyzed, and the existing path matching algorithm is corrected to overcome the problem that the probabilistic matching algorithm has poor matching effect when the base station distribution is uneven; and a correction method based on direction angles is proposed for the first time, which can quickly and accurately eliminate repeated matches and misidentified paths caused by road intersections. The present invention ultimately achieves a technological breakthrough in identifying travel routes in dense urban road networks based on mobile phone signaling data, which can help trace the source of congestion, track epidemic patients and virus transmission routes, and also provide a basis for the construction of new roads and road reconstruction and expansion in cities, ensuring the scientific rationality of government decision-making and planning and construction. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] The present invention will now be described by way of example with reference to the accompanying drawings, in which:

[0033] Figure 1 is an overall identification flow chart of the present invention;

[0034] Figure 2 Schematic diagram of the smoothing effect of different distributions of signaling data of the present invention, wherein: (a) the signaling data is on both sides of the road during straight driving; (b) the signaling data is on one side of the road during straight driving; (c) the signaling data is on one side of the road during U-turn driving; (d) the signaling data is on both sides of the road during U-turn driving;

[0035] Figure 3 Schematic diagram of the modified probability matching algorithm;

[0036] Figure 4 Figure 1 is the recognition result diagram, where: (a) the original signaling data distribution; (b) the signaling data distribution after high-speed filtering and smoothing; (c) the travel path fitting using non-uniform rational B-spline curve; (d) the travel path recognition result;

[0037] Figure 5 This is a diagram for analyzing the model recognition effect, including: (a) the shortest distance recognition algorithm; (b) the linear interpolation point filling algorithm; (c) not considering the result repair method; (d) the model proposed by the present invention. DETAILED DESCRIPTION

[0038] like Figure 1 As shown, the overall identification flow chart of the present invention includes:

[0039] 1. Basic Data Extraction and Analysis

[0040] First, the signaling data and road network data fields involved in the present invention are analyzed respectively. The signaling data field is provided by the mobile operator and mainly includes an identification code, a timestamp, a current base station location area code and a cell code, and the current base station longitude and latitude. The meaning of each field is shown in Table 1.

[0041] Table 1 Main fields of signaling data

[0042]

[0043] The road network data is constructed by web crawlers or manually. Each road contains multiple sections, mainly including road number, section number, section starting latitude and longitude, and section end latitude and longitude. The meaning of each field is shown in Table 2.

[0044] Table 2 Main fields of road network data

[0045]

[0046] 2. Basic Data Preprocessing

[0047] Data preprocessing is used to process signaling data caused by unstable communication systems, including small samples, communication failures, data duplication, field missing, ping-pong switching, and signal drift data, to reduce interference with the recognition process, including the following situations:

[0048] 1) Delete data with LAC and CID changes less than once to avoid being unable to analyze due to the small amount of data for a single trip.

[0049] 2) Directly delete the signaling data generated by communication failure, field missing, and data duplication and redundancy to avoid matching errors.

[0050] 3) Set a maximum speed threshold, gradually calculate the switching speed between adjacent signaling data, eliminate all signaling data with a switching speed greater than the set threshold, and iterate multiple times to ensure that the switching speed of all data is less than the threshold to avoid interference with recognition caused by ping-pong switching and signal drift.

[0051]

[0052] Formula (1) is the formula for calculating the switching speed, where (x1, y1), (x2, y2) are the coordinates of any adjacent signaling data, ΔT is the corresponding time interval, and R is the radius of the earth, which is 6371004m; is the radian conversion constant, which is π / 180.

[0053] 3. Refined Fitting of Travel Paths

[0054] The refined fitting of travel paths includes two parts: signaling data smoothing and travel path fitting. First, the fluctuation of signaling data is weakened, and then the weakened signaling data is used as a control point to realize path fitting.

[0055] 1. Signaling data smoothing

[0056] Since base stations are mostly deployed in buildings on both sides of the road, the actual signaling data presents a fluctuating distribution relative to the travel section. It is necessary to first use the spatiotemporal clustering algorithm to select the smoothing object, and then use Gaussian filtering to smooth the fluctuation.

[0057] In the spatiotemporal clustering algorithm, the longitude, latitude, and timestamp of each signaling data are used as input to calculate the amount of signaling data within the specific spatiotemporal range of each data point, find the core clustering point, and form several irregular clusters through multiple iterations. The signaling data of each cluster has highly similar spatiotemporal characteristics. Subsequently, the data that is not clustered in the clustering results is found, and all signaling data is divided into several segments based on it, which will be used as the input of Gaussian filtering smoothing.

[0058] In Gaussian filtering fluctuation smoothing, the longitude and latitude of each segment of signaling data is used as input, and the weighted average of each signaling coordinate and the coordinates of several surrounding signaling data is gradually calculated, where the weight is centered on the current signaling data and satisfies the Gaussian distribution. Figure 2 The figure is a schematic diagram of the signaling data distribution and Gaussian filter smoothing effect during different driving processes. After completing the position smoothing of each segment of signaling data, the smoothed and unsmoothed signaling data are recombined in the order of the original timestamps to construct complete individual signaling data.

[0059] 2. Travel path fitting

[0060] The present invention considers that non-uniform rational B-spline curves can fit arbitrary shapes and have outstanding local expression advantages. The latitude, longitude and timestamp of the smoothed individual complete signaling data are used as input control points. After the number of fitting nodes is set, the latitude, longitude and timestamp of each node are gradually calculated based on any three consecutive control points. The travel fitting trajectory with time characteristics is output.

[0061]

[0062]

[0063]

[0064] Formula (2) is the non-uniform rational B-spline expression with h+1 control points, P r is the r-th control point set, i.e., the r-th signaling point cluster in the spatiotemporal clustering; C(s) is the coordinate of the fitting path node s; N r,c (s) is the c-order B-spline basis function, defined by the Cox-deBoor recursive formula.

[0065] 4. Travel Path Matching and Repair

[0066] Travel path matching and repair includes two parts: travel path matching and travel path repair. After completing the path fitting, it is necessary to match the fitting nodes with the actual road network, and correct the results according to the matching effect, and finally realize the mapping from signaling data to the actual road network.

[0067] 1. Travel path matching

[0068] The present invention combines the low positioning accuracy of signaling data, the dense distribution of communication base stations and road networks in the city center and the sparse distribution in the suburbs, and makes the following corrections to the probability matching algorithm based on GPS data:

[0069] 1) Modify the fixed distance threshold to the road number threshold m to ensure that each fitting node has a stable number of matching roads.

[0070] 2) Whether there is a common matching road is used as the basis for selecting statistical objects. When there is no common matching road between the current e and e+1 fitting nodes, the first e fitting nodes are used as statistical objects to realize automatic node splitting, and then find the matching road with the highest likelihood probability for this group of fitting nodes. The denser the fitting nodes, the smaller the misidentification caused by splitting deviation.

[0071] The present invention uses the fitted nodes and the constructed road network as input to perform road matching. The modified probability matching algorithm is shown as follows: Figure 3 As shown, the fitting nodes {C1, C2, C3, C4, C5} have a common road A, and node C6 is not matched to road A. Therefore, road A with the highest probability is regarded as the matching road of the first five fitting nodes, and then matched to the nearest road segments in road A respectively (the result is {A1, A2, A2, A3, A3}). Then, the fitting node {C6, C7} is matched to segment E1 according to this method (the result is {E1, E1}).

[0072] 2. Travel route repair

[0073] There are two types of problems with the modified probability matching algorithm: (1) Each fitting node will perform road matching, resulting in a large amount of duplicate data in the matching results (such as Figure 3 The matching result is {A1, A2, A2, A3, A3, E1, E1}, and the actual travel path is {A1, A2, A3, E1}). (2) In areas with dense and complex road networks, the road sections around the actual travel path are prone to intersect with the fitting curve, resulting in misidentification of the intersecting road sections. In this regard, the present invention proposes a repair method that considers the direction angle and road network topology, including the following 5 steps:

[0074] 1) Eliminate duplicate data and mutation data. First, for continuous and repeated identification sections, retain unique values. Then, for any identification result with the same road number and continuous section number, if there are 1 to 2 other road number data, the data of other road numbers will be eliminated.

[0075] 2) Supplement the missing sections within the same road number. For a section of continuous recognition results with the same road number, if its section numbers are not continuous, they are supplemented to form a basic travel path.

[0076] 3) Eliminate misidentified data of intersecting sections based on the direction angle. Calculate the direction angle θ1 (-90°<θ1≤90°) of all identified sections and fitting nodes. For a certain identified section, when the angle θ2 (0°<θ2≤90°) between the fitting node matched to the section and the section direction angle is less than the angle threshold When the ratio is less than the probability threshold μ, or the number of fitting nodes is less than the discrimination number threshold η, the road section is regarded as an intersection section and removed.

[0077] 4) Re-supplement the missing road sections with the same road number. Re-search for all the continuous recognition results with the same road number, and when the road section numbers are not continuous, make them continuous.

[0078] 5) Supplement missing sections between different road numbers. For any two consecutive identified sections with different road numbers, when the distance between the sections is greater than the distance threshold λ, the remaining sections under the two road numbers are supplemented in sequence until there is a minimum distance between the supplemented sections or they intersect.

[0079] 5. Travel Path Identification Example

[0080] The travel path recognition example includes three parts: individual travel examples, comparison of existing methods, and comparison of recognition effects in different environments. In the individual travel example, the distribution of original signaling data and the recognition effects under the main steps are shown in detail. In the comparison of existing methods, the difference in effect between the method proposed by the present invention and the existing path matching method is shown. In the recognition effects in different environments, the recognition effects of the present invention under different travel lengths, urban environments, and road grades are shown.

[0081] 1. Individual travel examples

[0082] Figure 4 This is an example of travel path identification based on signaling data. In this example, the travel path is located in the central area of ​​the city with a dense and complex road network, with parallel sections of about 5 km, where the shortest interval between parallel sections is 265 meters and the longest interval is about 800 meters. Figure 4 (a) is the distribution of the original travel signaling data, which has large fluctuations and repetitions compared to the actual travel path. Figure 4 (b) is the result of smoothing after spatiotemporal clustering and Gaussian filtering. Even if the travel route is unknown in advance, the smoothed result still basically falls on the actual travel route, and the fitting effect becomes better as the density of signaling data increases, which improves the feasibility of travel path identification from the source. Figure 4 (c) is the non-uniform rational B-spline travel path fitting result based on the filtered data. By encrypting the number of nodes, the effective application of the modified probability matching algorithm is guaranteed, and the data fluctuation characteristics are further smoothed. Figure 4 (b) The gyration caused by repeated interactions with the base station within the circle reduces the risk of misidentification. Figure 4 (d) is the travel path recognition result. Only on the east side where the signaling data is sparse, the recognition result differs from the actual path.

[0083] 2. Comparison of existing methods

[0084] The present invention compares the shortest distance recognition algorithm, linear interpolation point filling algorithm, and three types of algorithms that do not consider the result repair in the existing road matching. The shortest distance recognition algorithm is an identification method that directly distributes the pre-processed signaling data to the nearest road section and repairs it according to the road network structure. The linear interpolation point filling algorithm is a method that inserts intermediate trajectory points in equal proportion according to the signaling data interval time and longitude and latitude, and then matches the trajectory points to the nearest road section according to the shortest distance and repairs it.

[0085] Figure 5 is the final comparison result, where the horizontal axis represents the actual travel length and the vertical axis represents the length of the identified travel path. It can be seen that as the travel distance increases, the results of the shortest distance recognition algorithm deviate significantly from the true value, and the length of the wrong travel increases slowly; the linear interpolation point filling algorithm greatly improves the recognition accuracy under different conditions, but the length of the wrong recognition also increases significantly; the method that does not consider the result repair is basically consistent with the linear interpolation point filling in the correct recognition result, and significantly reduces the length of the wrong recognition; after introducing the angular feature, the recognition result of the model proposed by the present invention is closer to the real travel, and the length of the wrong recognition decreases again, indicating that the recognition effect of the model proposed by the present invention is significant.

[0086] 3. Comparison of recognition effects in different environments

[0087] Table 3 shows the recognition results of different path lengths and path turning points. A path turning point refers to the intersection of two adjacent road sections with a turning angle greater than 40° in the driving direction. The recognition accuracy of different path lengths is maintained above 80%, and most of them are above 85%. It decreases slightly with the increase of path length, and the length index is slightly higher than the number index. In terms of recognition error rate, the error rates of the two indicators are maintained at 6%-7% on average, showing a trend of first increasing and then decreasing, and the length index is lower than the number index. From the perspective of path turning points, the recognition accuracy and error first decrease and then increase. When the number of turning points is 10-15, the accuracy rate drops to 81.60% and 77.25%, respectively, and the error rate drops to 2.92% and 2.84%.

[0088] Table 3 Recognition results under different travel lengths

[0089]

[0090] Table 4 Recognition results under different urban areas, road network densities, and traffic conditions. The recognition results of different road network densities are similar to those of urban areas, with the accuracy rate remaining at 85%-90% and the error rates being around 5% and 7%, respectively. This is related to the differences in the construction of road network densities in different urban areas. From the perspective of traffic conditions, the recognition accuracy rate increases from 87% to 93% with road congestion. This is because congestion often occurs in the central area of ​​the city, where the road network density is large and the signaling data is more intensive and frequent, resulting in a higher error rate in congested sections. The error rate for different traffic conditions is between 5% and 8%.

[0091] Table 4 Recognition results in different urban environments

[0092]

[0093] Table 5 shows the recognition results of different road grades. Road grade refers to the grade of the longest road that an individual travels through during his trip. The recognition accuracy remains above 90%, and the error rate increases significantly with the decrease of road grade, reaching 5%, 9.41% and 21.59% respectively. In addition to the low sample size and dense road network, the trip length mainly on secondary roads is short (average 3.72km). The length of the road section has a very obvious impact on the recognition results. Even a small part of the road section fitting error will cause a large error.

[0094] Table 5 Recognition results under different road levels

[0095]

Claims

1. A method for identifying vehicle travel paths in a dense urban road network environment, characterized by: The steps include: Step 1: Collect mobile phone signaling data, build the target city road network, and unify the coordinate system; Step 2: pre-process the signaling data collected in step 1 to remove the signaling data caused by communication failure, data duplication, field missing, ping-pong switching and signal drift; wherein the method for removing the signaling data caused by ping-pong switching and signal drift is: Set a maximum speed threshold, gradually calculate the switching speed between adjacent signaling data, eliminate all signaling data with a switching speed greater than the set threshold, and iterate multiple times to ensure that the switching speed of all data is less than the threshold, wherein the switching speed is calculated according to the following formula: Where (x1, y1), (x2, y2) are the coordinates of any adjacent signaling data, ΔT is the corresponding time interval, and R is the radius of the earth, which is 6371004m; is the radian conversion constant, and its value is π / 180; Step 3: Based on spatiotemporal clustering and Gaussian filtering to smooth the signaling data, a complete individual signaling trajectory is constructed; Step 4: Fitting the travel path based on the non-uniform rational B-spline curve; Step 5: Travel path matching; Step 6: Repair the travel path and complete the travel path identification: The first step is to remove duplicates from the continuous and repeated recognition results and remove the sudden recognition sections: first, for the continuous and repeated recognition sections, retain the unique value; then for any recognition result with the same road number and continuous section number, if there are 1 to 2 other road number data, remove the data of other road numbers; Step 2: For the continuous recognition results with the same road number, the parts with discontinuous road section numbers are supplemented; Step 3: Calculate the direction angles of all identified road sections and fitting nodes. For the angle between a road section and the corresponding fitting node, if the ratio of the angle less than the angle threshold is less than the set probability, or the number of fitting nodes is less than the discrimination number threshold, remove the road section. Step 4: For the continuous recognition results with the same road number, the discontinuous part of the road section number is supplemented again; Step 5: For two consecutive identified sections with different road numbers, when the distance between them is greater than the distance threshold, the remaining sections under the two road numbers are supplemented in sequence until there is a minimum distance or the supplemented sections intersect.

2. The method for identifying vehicle travel paths in a dense urban road network environment according to claim 1, characterized in that: Step three smoothes the signaling data. The method for constructing complete individual signaling data is as follows: first, use the spatiotemporal clustering algorithm to cluster the signaling data according to its spatiotemporal characteristics, then find the unclustered data, and use it as a benchmark to segment the individual signaling data. Then, for each segment of the signaling data, use Gaussian filtering to gradually perform weighted averaging of the signaling position information to achieve data smoothing. Finally, recombine the smoothed and unsmoothed signaling data in the order of the original timestamps to construct a complete individual signaling trajectory.

3. The vehicle travel path identification method in a dense urban road network environment according to claim 1 is characterized by: The method for fitting the travel path based on the non-uniform rational B-spline curve described in step 4 is: using the non-uniform rational B-spline curve, taking the smoothed signaling data as the control point, gradually fitting the actual travel trajectory, and extracting the coordinates of all nodes of the fitting trajectory.

4. The method for identifying vehicle travel paths in a dense urban road network environment according to claim 3, characterized in that: The node coordinates are calculated according to the following formula: Where h is the number of control points of the non-uniform rational B-spline curve, P r is the rth signaling point cluster in spatiotemporal clustering; C(s) is the coordinate of the fitting path node s; N r,c (s) is the c-order B-spline basis function.

5. The vehicle travel path identification method in a dense urban road network environment according to claim 1, characterized in that: The method for matching the travel path described in step 5 is: modify the probability matching algorithm based on GPS data as follows: 1) Modify the fixed distance threshold to the road number threshold to ensure that each fitting node has a stable number of matching roads; 2) Whether there is a common matching road is used as the basis for selecting the statistical object. When there is no common matching road between the current e and e+1 fitting nodes, the first e discrete points are split as statistical objects, and then the matching roads of the split nodes are found.

6. The vehicle travel path identification method in a dense urban road network environment according to claim 1, characterized in that: The mobile phone signaling data in step 1 includes: MSID, timestamp, LAC, CID, base station longitude and latitude; the road network data includes: road number, section number, section starting point longitude and latitude, and section end point longitude and latitude.

Images