A map matching method based on forward-looking trusted regions

Abstract

The present invention discloses a map matching method based on forward-looking credible regions, comprising the following steps: S1, extracting characteristic trajectory points in a trajectory T based on average direction and road width to form a compressed trajectory T′; S2, determining the region circle of each trajectory point in the compressed trajectory T′ and its credible region within the region circle, the road section within the credible region being the candidate road section of the corresponding trajectory point; S3, performing spatial analysis, temporal analysis and directional analysis on the candidate road section of each trajectory point to obtain the best matching path of the trajectory T. Detect the driving state of the trajectory and perform incremental patching on the missing trajectory. Compress the trajectory data using the average direction and road width of the trajectory point to extract the trajectory structure. In the matching stage, partition the candidates for trajectory point matching to obtain the candidate credible region of the trajectory point to reduce candidate estimation. Finally, estimate the candidates within the credible region to obtain the optimal candidate.

Description

Technical Field

[0001] The present invention belongs to the technical field of map matching, and more specifically, to a map matching method based on forward-looking credible regions. Background Art

[0002] In the field of Intelligent Transportation Systems (ITS), map matching, a technique that maps each raw trajectory obtained by a global positioning system onto a road network, has been widely adopted and is the basis for many trajectory data-based applications such as route optimization, traffic scheduling, and autonomous driving.

[0003] Many map matching methods have been proposed, among which the Hidden Markov Model (HMM) matching method for noise is favored by the majority of researchers. For example, the THMM method, the STD-matching method, the interactive matching method, and the interactive voting method based on routing selection, all filter out outliers that exceed the distance threshold when processing noisy trajectory data. The above four methods show good performance when analyzing outlier noisy trajectory data, but do not consider the impact of missing trajectory points on matching, resulting in matching errors in some sections where trajectory points are missing. Affected by factors such as electromagnetic interference and poor GPS signals, the trajectory data reported by mobile objects carrying positioning devices may be abnormally missing. When the missing trajectory point is located at an intersection consisting of multiple sections, the existing matching method may find it difficult to determine which section the trajectory point is in, because it is more difficult to extract useful information from the missing trajectory.

[0004] In addition, in dense road networks, the number of candidate road sections for each GPS point increases significantly, which hinders the effectiveness of HMM-based algorithms. In order to improve the accuracy of candidates, Chen D et al. established recognition areas at intersections. However, this method is only applied to roads at intersections and ignores the impact of direction factors on candidate decisions, making it difficult to determine the correct candidate. Summary of the invention

[0005] The present invention provides a map matching method based on forward-looking credible regions, aiming to improve the above-mentioned problem.

[0006] The present invention is implemented as follows: a map matching method based on forward-looking credible regions, the method specifically comprising the following steps:

[0007] S1, extracting characteristic trajectory points in trajectory T based on average direction and road width to form a compressed trajectory T′;

[0008] S2, determining the area circle of each trajectory point in the compressed trajectory T′ and its credible area within the area circle, and the road section within the credible area is the candidate road section of the corresponding trajectory point;

[0009] S3. Perform spatial analysis, temporal analysis, and directional analysis on the candidate sections of each trajectory point to obtain the best matching path of trajectory T.

[0010] Furthermore, the method for extracting characteristic trajectory points is as follows:

[0011] S11, taking the first trace point in the trajectory T as the initial feature trajectory point;

[0012] S12, calculate each subsequent trajectory point p in turn f The average turning angle and trajectory point p f Distance to the straight line f ;

[0013] S13, if the average turning angle of the trajectory points Greater than the angle threshold θ, or the distance l f If it is greater than the width threshold w / 2, then the trajectory point p is identified f is the characteristic trajectory point.

[0014] Furthermore, the trajectory point p f The average turning angle The calculation formula is as follows:

[0015]

[0016] Among them, HD f Represents the trajectory point p f The direction angle of the driving direction, HD i Track point p i The driving direction of the trajectory is represented by a, and a represents the number of trajectory points.

[0017] Furthermore, the trajectory point p f The method for determining the radius of the area circle is as follows:

[0018]

[0019] Among them, d i,f Represents the trajectory point p f With trajectory point p i The distance between f ,t i Represent the trajectory point p f , trajectory point p i The timestamp, w s represents the size of the window, Δt represents the trajectory point The timestamp and trajectory point p f-1 The difference in timestamps.

[0020] Furthermore, the trajectory point p fThe method for determining the trusted area in the area circle is as follows:

[0021] Get the trajectory point p in the area circle f-3 , trajectory point p f-2 and trajectory point p f-1 The angle range Δθ from the minimum direction angle to the maximum direction angle is expanded to both sides by an angle θ′, and the area formed is the trajectory point p f Trusted region within the region circle.

[0022] Furthermore, before step S1, the method further includes:

[0023] S4. Interpolate the original trajectory with missing trajectories to form trajectory T.

[0024] Furthermore, the detection method of missing tracks is as follows:

[0025] Calculate two adjacent trajectory points p i+1 、p i If the time interval and distance interval satisfy the following conditions, then the two trajectory points p are determined. i+1 、p i There are missing trajectories between them;

[0026] p i+1 (t)-p i (t)≥T max (1)

[0027] p i+1 (d)-p i (d)≥D max (2)

[0028] Among them, p i+1 (t), p i (t) represents the timestamp of the i+1th and ith trajectory points respectively, p i+1 (d) p i (d) represent the positions of the i+1th and ith trajectory points, respectively, T max represents the threshold of the time interval, D max Indicates the threshold value of the distance interval.

[0029] Furthermore, the trajectory interpolation method between two trajectory points is as follows:

[0030] S41, inserting a new trajectory point between two adjacent trajectory points with missing trajectories, the position of the new trajectory point is located at the midpoint of the line connecting the adjacent trajectory points, and the time is the median of the timestamps of the adjacent trajectory points;

[0031] S42, re-detecting whether there is a missing track between the new track point and two adjacent track points. If the detection result is yes, executing step S41 until there is no missing track between all adjacent track points.

[0032] Furthermore, before detecting the missing track, the following steps are also included:

[0033] Delete the trajectory points with abnormal speed and position in the original trajectory.

[0034] The map matching method provided by the present invention mainly includes three parts: incremental trajectory reconstruction, trajectory structure extraction and map matching. First, the driving state of the trajectory is detected and the missing trajectory is incrementally repaired. The trajectory data is compressed using the average direction and road width of the trajectory point to extract the trajectory structure. In the matching stage, the candidates for trajectory point matching are partitioned to obtain the candidate credible area of ​​the trajectory point to reduce the candidate estimation. Finally, the optimal candidate is obtained by estimating the candidates within the credible area. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 A schematic diagram of the direction of a moving object is provided for an embodiment of the present invention;

[0036] Figure 2 Provide a schematic diagram of missing movement tracks for an embodiment of the present invention;

[0037] Figure 3 A schematic diagram of a trajectory interpolation process is provided for an embodiment of the present invention;

[0038] Figure 4 Schematic diagram of a traditional trajectory compression algorithm provided by an embodiment of the present invention, wherein (a) is the original trajectory, (b) is the compressed trajectory after being compressed by the OpenWindow algorithm, and (c) is the compressed trajectory after being compressed by the TopDown algorithm;

[0039] Figure 5 A schematic diagram of an abnormal trajectory processing process provided by an embodiment of the present invention, wherein (a) is trajectory data including abnormal trajectory points, and (b) is trajectory data processed by using the absolute value of the direction difference of the trajectory points as the rotation angle;

[0040] Figure 6 A schematic diagram of average rotation angle calculation provided by an embodiment of the present invention;

[0041] Figure 7 Schematic diagram of trajectory extraction based on angle threshold provided in an embodiment of the present invention, wherein (a) is the original trajectory, (b) the angle threshold is 45°, (c) the angle threshold is 30°, and (d) the angle threshold is 10°;

[0042] Figure 8A schematic diagram of a straight-line travel line of a moving object provided by an embodiment of the present invention;

[0043] Fig. 9 The average direction and road width strategy provided by the embodiment of the present invention is Figure 7 (a) The compressed trajectory after data trajectory compression;

[0044] Fig.10 A schematic diagram of candidate road segment selection based on a regional circle provided in an embodiment of the present invention, wherein (a) is candidate road segment determination based on an existing method, and (b) is candidate road segment determination based on a forward-looking credible area;

[0045] Fig.11 The trajectory data with a "Z"-shaped error in the input position provided by the embodiment of the present invention;

[0046] Fig.12 A schematic diagram of dividing the trusted area and the untrusted area within the area circle provided in an embodiment of the present invention;

[0047] Fig.13 A schematic diagram of the moving direction of the trajectory points of a sliding trajectory with a short sliding window of 3 provided in an embodiment of the present invention;

[0048] Fig.14 The projection planes of the elevated road and the ground road provided in the embodiment of the present invention are overlapping schematic diagrams;

[0049] Fig.15 A schematic diagram of a scenario where sampling points and candidate points coincide with each other provided by an embodiment of the present invention;

[0050] Fig.16 A schematic diagram of a scene of multiple similar candidate road sections provided by an embodiment of the present invention. ; DETAILED DESCRIPTION

[0051] The specific implementation modes of the present invention are further explained in detail below by describing the embodiments with reference to the accompanying drawings, so as to help those skilled in the art to have a more complete, accurate and in-depth understanding of the inventive concept and technical solution of the present invention.

[0052] The definitions and symbols related to the present invention are shown in Table 1.

[0053] Table 1 Character table

[0054]

[0055]

[0056] Definition 1: Origin Trajectory: The trajectory of a moving object is a series of space-time points, where each trajectory point P consists of longitude, latitude and timestamp, that is, the spatial position of the trajectory at time t is represented by (x, y, t). In two-dimensional space, the trajectory is represented by T = {(x1, y1, t1), (x2, y2, t2), ..., (x i ,y i ,t i )|1≤i≤n} The trajectory of a moving object is essentially continuous, but due to GPS device errors and signal obstruction, the captured trajectory is a series of discrete trajectory points. The final trajectory T′ is the final result obtained after the original trajectory T is processed, expressed as T′={(x1,y1,t1),(x2,y2,t2),...,(x j ,y j ,t j )|1≤i≤m}.

[0057] Definition 2: Road Segment: A road segment e is a directed edge, which represents a linear road segment between two nodes. It is described by the attributes e.id, el, e.start, and e.end.

[0058] Definition 3: Path: A path P is a series of connected segments, P = {e i ,...,e j}∈G(V,E).

[0059] Definition 4: Road network: A road network is a directed graph G(V, E) representing a digital map of real geographical areas and roads, where V is a vertex set consisting of the start and end points of road segments, and E is a set of edges representing road segments.

[0060] Definition 5: Moving object direction: The moving object direction HD is expressed as the deflection angle of the object's current trajectory point relative to the true north direction, such as Figure 1 As shown, the trajectory point p i-1 and trajectory point p i The directions are HD i-1 and HD i .

[0061] Definition 6: (Angle): The angle SA represents the difference between the direction of the current trajectory point and the direction of the previous trajectory point, such as Figure 2 As shown, the trajectory point p i The turning angle is SA i , the calculation formula is: SA i =|HD i -HD i-1 |.

[0062] Definition 7: (Map Matching): Given a journey trajectory T and a road network G(V, E), map matching is the problem of how to match the trajectory T to a path P on the road network.

[0063] The map matching method provided by the present invention mainly includes three parts: incremental trajectory reconstruction, trajectory structure extraction and map matching. First, the driving state of the trajectory is detected and the missing trajectory is incrementally repaired. The trajectory data is compressed using the average direction and road width of the trajectory point to extract the trajectory structure. In the matching stage, the candidates for trajectory point matching are partitioned to obtain the candidate credible area of ​​the trajectory point to reduce the candidate estimation. Finally, the optimal candidate is obtained by estimating the candidates within the credible area.

[0064] 1. Trajectory Reconstruction

[0065] Due to the missing trajectory segments, the trajectory data structure is incomplete, and the difficulty of trajectory matching increases. Figure 2 An example of this problem is shown in Figure 1, where region 1 is located in the gap between trajectory segments P1-P7 and trajectory segments P8-P9, and it represents all possible positions of the moving object in the missing time interval. If it happens to be an important part of a trajectory (such as region 1 at an intersection), matching this type of trajectory data usually results in unsatisfactory accuracy, that is, the trajectory is fuzzy on road 2 and road 3. Considering the missing trajectories caused by these external factors, it is necessary to find where the discrete trajectory segments may be connected. Because in the case of trajectory interruption, vehicle positioning errors may inevitably occur.

[0066] In order to solve this problem, the present invention proposes a trajectory reconstruction method to repair the missing trajectory. In order to avoid the situation where the reconstructed trajectory points have impulse noise and Gaussian noise, firstly, the erroneous data that obviously does not conform to the vehicle motion law, including data with abnormal speed and position, is identified and deleted; secondly, the missing segment of the trajectory is identified based on the time interval and distance interval between two adjacent trajectory points, see formulas (1) and (2), which are specifically expressed as follows:

[0067] p i+1 (t)-p i (t)≥T max (1)

[0068] p i+1 (d)-p i (d)≥D max (2)

[0069] Among them, p i+1 (t), p i (t) represents the timestamp of the i+1th and ith trajectory points respectively, p i+1 (d) pi (d) represent the positions of the i+1th and ith trajectory points, respectively, T max represents the threshold of the time interval, D max The threshold representing the distance interval is called the detection factor. There is a missing trajectory between trajectory points that satisfy both (1) and (2).

[0070] For missing trajectory segments, check the neighborhood of each missing segment, sort the trajectory points in the field and map them into a one-dimensional vector. Calculate the average of the head and tail of the vector to get the median point, and iterate by looking forward and backward through the trajectory points to calculate the average value until the distance difference and time difference between two consecutive trajectory points are no greater than the threshold. The median point obtained is the incremental trajectory point.

[0071] The interpolation point is taken on the straight line between the two trajectory points to interpolate the missing trajectory segment. Since the interpolation operation is performed between the two preceding and following sample points, the specific process is as follows: Figure 3 In the , it is detected that there is a missing track between the track point p7 and the track point p8. The median point of the track point p7 and the track point p8 is calculated to obtain the incremental point p n′ Similarly, detect the incremental point p n′ When the time threshold and distance threshold are met with trajectory points p7 and p8, an incremental point p is added. n′ and the incremental point p of trajectory point p7 n-1′ , increment point p n′ and the incremental point p of trajectory point p8 n+1′ , check again whether the distance threshold and time threshold are met, and iterate in sequence to obtain the completed trajectory.

[0072] Algorithm 1 gives the pseudo code of the trajectory reconstruction method. Please note that the threshold T of formula (1) and formula (2) needs to be adjusted through experiments. max and D max to set it up.

[0073]

[0074]

[0075] The above T start 、T end That is, there are timestamps of adjacent trajectory points where the trajectory is missing. In summary, the trajectory reconstruction method is as follows:

[0076] (1) Identify and delete erroneous data that clearly does not conform to the laws of vehicle movement;

[0077] (2) Based on formulas (1) and (2), detect whether there is a missing track between two adjacent track points;

[0078] (3) If the detection result is yes, a new trajectory point is inserted between the two adjacent trajectory points. The position of the inserted trajectory point is located at the midpoint of the line connecting the adjacent trajectory points, and the time is the median of the timestamps of the adjacent trajectory points;

[0079] (4) Detect whether there is a missing track between the inserted track point and two adjacent track points, and execute step (3) until there is no missing track between all adjacent track points.

[0080] If the distance and time between two adjacent trajectory points satisfy the above formulas (1) and (2), there is a missing trajectory between the two adjacent trajectory points, and trajectory interpolation is required.

[0081] 2. Trajectory Structure Extraction Algorithm Based on Average Direction and Road Width Compression (ADRW)

[0082] When extracting the structure of the interpolated trajectory, it is necessary to remove redundant points and abnormal points while retaining key GPS points. Therefore, from the perspective of trajectory structure, the invention proposes a trajectory structure extraction method that combines the average direction compression strategy and the road width compression strategy.

[0083] Average directional compression: For some traditional trajectory compression algorithms, such as OpenWindow algorithm and TopDown algorithm, since they do not consider directional information, such as Figure 4 As shown in (b) and (c), many details will be lost after compression, destroying the trajectory structure. Therefore, the main purpose of considering directional information is to retain more motion details and directions of moving objects after compression. At the same time, when using the directional information of moving objects as a compression strategy, when calculating the turning angle of the current trajectory point, only the absolute value of the direction difference between the previous trajectory point and the current trajectory point is used as the turning angle. However, there are often certain errors in trajectory data. This type of method may retain abnormal trajectory data in the compression result. Figure 5 (a) shows an example of this problem, where a trajectory is shown on a straight road, with trajectory point p f The trajectory point p a is the abnormal trajectory point of the starting trajectory. If the above method is used to calculate the turning angle, then p f The angle of the point is SA f When the compression angle threshold θ = 40°, the compression result is as follows: Figure 5 As shown in (b), the abnormal trajectory point p f It is retained and affects the compression results of the two surrounding points, and the trajectory structure cannot be extracted well.

[0084] In order to solve this problem, the present invention proposes an average direction strategy, that is, calculating p a Point to p f-1The average direction of all points between points, and this average direction is compared with p f The difference in the direction of the point is taken as the angle of the point. The calculation formula for the angle is:

[0085]

[0086] Using the above method, we can calculate p f-a Point to p f-1 The average direction between points is Figure 6 As shown by the dashed line in the middle, it is obvious that after adopting the average direction, the abnormal point p f Corner It is much smaller than before, and the extracted trajectory structure is more consistent with the current road structure.

[0087] Spatial compression: If only the average direction compression strategy is used to extract the trajectory structure, new problems will arise, and it will be insensitive to some roads with direction changes. Especially on circular roads, the direction change of each trajectory point is small, and direct compression will also cause serious damage to the trajectory structure. Figure 7 (a) The black dots represent the original trajectory data. If only the direction information of the trajectory points is considered, only Figure 7 The black dots in (b) are obviously unable to effectively reflect the original trajectory structure. At the same time, if the angle threshold is reduced, such as Figure 7 (c) and (d), it can be seen that when the angle threshold is small to a certain extent, the extracted trajectory structure can also well reflect the original trajectory. However, when the angle threshold is too small, some non-feature points or trajectory points with collection anomalies may be retained when extracting the structure of ordinary road sections, resulting in an excessively large compression ratio and affecting subsequent matching.

[0088] Therefore, taking the above into consideration, the present invention introduces a road width compression strategy based on the calculation of the turning angle using the average direction and combining the spatial information of the trajectory points. Figure 8 As shown, the line DL represents the straight line of the current moving object, which is defined as follows:

[0089]

[0090] According to the straight line, the trajectory point p f Distance to the straight line f Defined as:

[0091]

[0092]

[0093]

[0094] Where S represents P a , P f The area of ​​the triangle in which the three are located, p is the intermediate parameter.

[0095] Then according to the angle and the offset to the travel line l f Whether it is greater than the given angle threshold θ and half of the width threshold w is used to determine p f Whether the point is a feature trajectory point, only the feature points are retained. After adding the road width strategy, Figure 7 The original trajectory in (a) is compressed and the result is as follows Fig. 9 As shown in the middle gray dots, it is obvious that the trajectory structure extracted after compression better reflects the original trajectory, and the number of trajectory points after compression is Figure 7 (d) (i.e. only considering θ=10°) is less.

[0096] Given the original trajectory T, angle threshold θ and width threshold w, first select the first point in the original trajectory as the initial feature trajectory point, and then calculate the rotation angle of each subsequent trajectory point pf in turn and trajectory point p f Distance to the straight line f , if the turning angle of the trajectory point Greater than the angle threshold θ, or the distance l f If it is greater than the width threshold w / 2, then the trajectory point p is identified f is the characteristic trajectory point, repeat the above process until all trajectory points are traversed, and finally retain all the characteristic trajectory points, which is the final compression result. Algorithm 2 gives the pseudo code of the trajectory structure extraction algorithm based on average direction and road width compression, as follows:

[0097]

[0098]

[0099] 3. Forward looking-trusted areas

[0100] The present invention proposes a method for forward-looking trusted area division (abbreviated as Fl-trusted areas), which removes candidates in untrusted areas and reduces unnecessary candidates. This method only needs to consider the information of trajectory points and is applicable in dense and complex road networks, aiming to improve the matching accuracy and efficiency of trajectory data in complex road sections.

[0101] Since errors may occur in the process of obtaining candidate trajectory points, especially when vehicles are traveling in dense and complex urban road networks, the trajectories generated by vehicles face intricate roads. Existing map matching algorithms are prone to errors in matching under such road network conditions (especially at intersections), or there are problems such as matching redundancy. Fig.10 As shown in (a), there are 7 candidates for trajectory point P3 within the candidate radius R area. Since the HMM model matching needs to calculate each candidate, unnecessary calculations are generated. The existing method divides the trajectory into intersecting trajectories and non-intersecting trajectories, and formulates different matching rules for each. By setting abnormal adjustments at intersections, the entry point is matched with the entry section, and the exit point is matched with the exit section. Since there are deviations in trajectory data in complex environments, and this method does not pre-process the data before matching, it is difficult for this method to process trajectory data with multiple road network scenarios or multiple sampling frequencies. In view of the above situation, the present invention introduces a forward-looking trusted area model, through which it is obtained that there are 2 candidate sections for trajectory point P3, namely candidates r2 and r3, to reduce the selection of candidates, such as Fig.10 (b) as shown.

[0102] Specifically, the forward-looking trusted region model should first consider the velocity information and allowable error of the trajectory points to determine the radius of the forward-looking trusted region circle. The velocity between any two points can be calculated by dividing the geodesic distance between the two points by the difference in their timestamps. However, due to the "Z" error in the input position (see Fig.11 ), which is noisy when used. Therefore, the current speed of the trajectory is estimated by averaging the speed between the position point and the previous position point window. That is, the formula for the radius of the forward-looking credible area is shown in (10). Since the data is compressed before trajectory matching, there is no need to consider the situation where the speed between two adjacent points is almost 0 when the vehicle is waiting.

[0103]

[0104] Next, the area circle obtained by the forward-looking credible area radius r′ is partitioned. Since a road intersection usually has more than two sections, the candidate trajectory points located there are usually more than those located in the straight section. Based on the real driving scene in reality, according to the characteristics of the trajectory, the area circle is divided into a credible area C1 and an untrustworthy area C0, such as Fig.12 Define a short sliding window to estimate the moving direction of the trajectory points of the sliding trajectory, and the window size is 3, such as Fig.13For trajectory point p4, the trusted area range is calculated by the angle range Δθ from the minimum direction to the maximum direction of three consecutive trajectory points, namely, trajectory point p1, trajectory point p2 and trajectory point p3. However, this division rule is only applicable to the turning section of the intersection where the vehicle driving direction changes greatly. When the vehicle is driving on a straight section or a circular section with a small angle change, the angle range Δθ is small, which may result in no candidates in the trusted area. Considering this situation, the angle range is marginalized and expanded, and the angle θ′ is increased. Fig.13 As shown. The formula for the angle range of the credible area is:

[0105] θ=2θ′+Δθ (11)

[0106] Among them, Δθ is the angle range from the minimum angle direction to the maximum angle direction of the three trajectory points connected front and back, and θ′ is the adjustment angle. The angle range Δθ is expanded to both sides by an angle θ′. The angle area formed is the credible area C1 of the area circle, and the remaining area is the untrustworthy area C0 of the area circle.

[0107] Note that the forward-looking credible region division starts from the third trajectory point of a trajectory, because the angle range Δθ of the credible region division needs to refer to the directions of the first two trajectory points, and the divided credible region is for a trajectory point subsequent to the trajectory point.

[0108] (IV) Feature Analysis

[0109] 1) Spatial feature analysis

[0110] For trajectory point P pi+1 and candidate points Observation probability and sampling point (P pi+1 ) and the corresponding candidate points The distance between them is related and follows the Gaussian distribution N(μ,σ 2 ). Therefore, the observation probability can be defined as:

[0111]

[0112] in, Indicates candidate points The observation probability of is the trajectory point P pi+1 and candidate points The great circle distance between the two points is the distance between the trajectory points P and pi+1 and candidate points Distances on Earth.

[0113] However, when the same-direction roads are parallel, the distances from the track points to the two roads are the same due to measurement errors of GPS devices, etc. The simplest example is the problem of overlapping projections of multiple roads (such as the case where the projections of elevated roads overlap with those of ground roads). Fig.14 In the figure, the projections of the elevated road r1 and the ground road r2 onto a plane overlap, and thus the trajectory points on the elevated road Also projected to This makes map matching difficult, that is, it is difficult to decide which road the trajectory point matches according to formula (12). Considering that the speed limits of the two roads are different, they are and Therefore, the decision of trajectory points can be made by considering the speed limits of different roads and introducing speed characteristic factors, that is, assuming that the driver is unlikely to exceed the road speed limit, when parallel roads in the same direction are close to each other and have different speed limits, the speed factor is defined to reduce the possible candidate sections. The speed characteristic factors are as follows:

[0114]

[0115] In the formula, Represents trajectory points On the road section k The speed limiting factor on It is on section r k The speed limit value on It is the trajectory point The recorded speed values, combined with formulas (12) and (13), are redefined as follows:

[0116]

[0117] Represents road segment r k Upper candidate point The probability of observation.

[0118] Spatial analysis refers to evaluating the vehicle's position from the track point based on road network data and GPS track points. Move to track point When the candidate point is Transfer to The possibility of transition between candidate points has been formalized into a method for analyzing the transition probability of the candidate points. The transition probability is the ratio of the Euclidean distance of the adjacent trajectory points to the Euclidean distance of the corresponding adjacent candidate points. The higher the ratio, the lower the transition probability. However, at the intersection, the adjacent candidate points may overlap. In this case, the Euclidean distance of the corresponding adjacent candidate points may be equal to 0 at the intersection, resulting in a transition probability of 0 or infinity.

[0119] When the sampling point and the candidate point coincide, the shortest path is 0 and the transition probability is 0. Existing methods use the difference d between the distance between adjacent candidate points and the distance between adjacent sampling points to measure the transition probability. In order to reduce the discrete distance, it is corrected using the average speed. This method can accurately evaluate the transition probability.

[0120] However, in special cases, when the sampling point and the candidate point coincide, such as Fig.15 As shown, the trajectory point There are 4 candidate points in the candidate area Sampling point There are 2 candidate points in the candidate area It is worth noting that the candidate points and candidate points When calculating the distance difference d between the path between the candidate point pair and the path between the sampling point pair, the value is 0, that is, the transition probability is 0. In fact, the transition probability of the candidate point coinciding with the trajectory point should be 1.

[0121] Therefore, considering the above situation, the coincidence distance is corrected and the transition probability is defined as follows:

[0122]

[0123] in, It is a trajectory point With trajectory points The Euclidean distance between Indicates candidate points With candidate points The Euclidean distance between The trajectory points Track Points The candidate point of Θ represents a very small correction number. The transition probability suggested in formula (15) is more robust to dense urban road networks. Because the inverse of the distance difference between the two will return a large number, when the distance difference is 0 or close to 0, the candidate point There is a big difference between them. We correct the probability by distance to avoid the wrong choice when the distance difference is 0. Therefore, combining formulas (14) and (15), the spatial analysis function is:

[0124]

[0125] Spatial analysis functions Indicates that from a spatial perspective, the candidate points are Transfer to possibility.

[0126] 2) Time behavior analysis

[0127] Time analysis uses the similarity definition of the speed limit of the road section and the speed limit of the candidate road section, and the calculation formula is as follows:

[0128]

[0129]

[0130] Among them, {r1,r2,...,r m} is a candidate point To the candidate point The sequence of connected segments, {r1·v,r2·v,...,r m ·v} represent the road segments {r1,r2,...,r m} speed limit value, From the candidate point To the candidate point The average speed, Indicates candidate points and candidate points The distance between the connecting sections, t i ,t i+1 Represents trajectory points Track Points The timestamp of the

[0131] 3) Direction analysis

[0132] When there are many candidate sections and the sections are similar, it is difficult to identify the correct path using speed limits. Fig.16 As shown, the trajectory point There are 5 candidate routes that are close to each other, with similar speed limits and inconsistent road directions. It is difficult to get accurate candidate routes from spatial analysis and temporal analysis. Considering the change in road direction, the direction of the candidate road segment is compared with the direction of the trajectory point in order to obtain a more likely choice.

[0133] In real-world scenarios, vehicles will usually maintain roughly the same direction as the road they are crossing. When there are different levels of approaching roads, such as in urban areas or overpasses, the vehicle's direction can be compared with the lane's direction to screen candidate road lanes. The direction factor coefficient is defined as follows:

[0134]

[0135] in, It is the trajectory point The recorded direction, i.e. the vehicle's direction of travel, is the candidate road segment r i kThe direction of the lane is the prescribed direction of travel for the lane. If the difference between the two directions is greater than 90°, it means that the vehicle is traveling perpendicular to the lane or in the opposite direction. In this case, it is unlikely that the vehicle is actually traveling in the lane under consideration, so adjustments are required. When studying the observations corresponding to lane changes in the test dataset, it was determined that during lane changes, the difference between the vehicle's heading and the lane's heading can reach 20°, so no adjustments are made when the difference between the two directions is less than 20°. Since the vehicle's heading is always different from the road direction, Has values ​​in [0,1].

[0136] The present invention uses trajectory points Driving direction and road segment direction The similarity between the direction of the moving object and the direction of the road section is evaluated by the angle difference Δθ between them. The smaller the angle difference, the greater the similarity, and the greater the probability that the corresponding candidate road section will be selected.

[0137]

[0138] The present invention considers the change of the direction of the circular section when calculating the section direction. Considering the reality, the direction of the section in the circular section is constantly changing. More specifically, the directions of two consecutive sections are usually different, and there is a changing direction difference between them. The present invention tracks the random change of the direction and calculates the section direction according to the two section endpoints (F mini, ln g, F mini, lat) and (S min i, ln g, S min i, lat) closest to the sampling GPS point on the candidate section, as shown below:

[0139] y=Math.sin(S min i,ln gF min i,ln g)*Math.cos(S min i,lat) (21)

[0140]

[0141] The calculation formula for the section heading is as follows:

[0142]

[0143] Among them, F min i,ln g and F min i,lat refer to the candidate road segment r i k The latitude and longitude of the endpoint closest to the GPS track point, S min i,ln g and S min i,lat refer to the candidate road segment r i kThe latitude and longitude of the second nearest endpoint of the GPS track point. Math.atan2(y,x) returns the plane angle (in radians) between the line segment from the origin (0,0) to the point (x,y) and the positive direction of the x-axis, that is, the heading of the segment is calculated through the nearest segment points (F min i,ln g, F min i,lat) and (S min i,ln g, S min i,lat).

[0144] Directional analysis function uses the normal distribution N(μ,σ 2 ), normalized angle difference, quantized trajectory points With candidate points The probability of a match is defined as follows:

[0145]

[0146] Δθ is the trajectory point Driving direction and road segment direction The angle difference between them is combined with formula (19) and formula (24) to quantify the direction analysis of the trajectory point and the road, and is calculated as follows:

[0147]

[0148] (4) Matching analysis

[0149] For each candidate path, three probability metrics are calculated, namely spatial probability, temporal probability and directional probability. The F-score function is formed by combining equations (9), (10) and (19) to find the best matching path. The path consists of several candidate segments. The F-score function is as follows:

[0150]

[0151] Select the best candidate path between every two adjacent candidate point sets to obtain a matching path The total score of the matching path F(Path) can be expressed as Finally, the overall scores are ranked and analyzed, and the matching path with the highest vote score is regarded as the best matching answer.

[0152] The map matching method based on forward-looking credible regions proposed in the present invention has the following beneficial technical effects:

[0153] (1) The incremental trajectory reconstruction method is used to solve the abnormal missing of the input position. First, the driving state between consecutive trajectory points in each sliding trajectory is detected, then the missing abnormality of the input trajectory is determined, and finally the missing trajectory is supplemented by the trajectory reconstruction method to restore the trajectory shape.

[0154] (2) A trajectory structure extraction method based on average direction and road width compression is used to obtain key GPS points of trajectory data from the perspective of trajectory structure. At the same time, an average direction compression strategy is used to smoothly collect abnormal trajectory data, reduce the impact of abnormal trajectory data, and reduce the cost of candidate calculation.

[0155] (3) Based on forward-looking trusted areas (Fl-trusted areas), this method can deal with the problem of too many candidates in complex urban road networks. This method combines the directional characteristics of trajectory points and the road network, divides the candidate trusted areas of trajectory points, and adopts a sliding area strategy to reduce trajectory point candidates to provide more accurate candidate estimates.

[0156] The present invention has been described exemplarily. Obviously, the specific implementation of the present invention is not limited to the above-mentioned method. As long as various non-substantial improvements are made using the method concept and technical solution of the present invention, or the concept and technical solution of the present invention are directly applied to other occasions without improvement, they are all within the protection scope of the present invention.

Claims

1. A map matching method based on forward-looking credible regions, characterized in that: The method specifically comprises the following steps: S1, extracting characteristic trajectory points in trajectory T based on average direction and road width to form a compressed trajectory T′; S2, determining the area circle of each trajectory point in the compressed trajectory T′ and its credible area within the area circle, and the road section within the credible area is the candidate road section of the corresponding trajectory point; S3, performing spatial analysis, temporal analysis and directional analysis on the candidate sections of each trajectory point to obtain the best matching path of trajectory T; The method for extracting characteristic trajectory points is as follows: S11, taking the first trace point in the trajectory T as the initial feature trajectory point; S12, calculate each subsequent trajectory point p in turn f The average turning angle and trajectory point p f Distance to the straight line f ; S13, if the average turning angle of the trajectory points Greater than the angle threshold θ, or the distance l f If it is greater than the width threshold w / 2, then the trajectory point p is identified f is the characteristic trajectory point; Trajectory point p f The average turning angle The calculation formula is as follows: Among them, HD f represents the trajectory point p f The direction angle of the driving direction, HD i Track point p i The driving direction, a represents the number of trajectory points; Trajectory point p f The method for determining the radius of the area circle is as follows: Among them, d i,f represents the trajectory point p f With trajectory point p i The distance between f ,t i Represent the trajectory point p f , trajectory point p i The timestamp, w s represents the size of the window, Δt represents the trajectory point The timestamp and trajectory point p f-1 The difference in timestamps.

2. The map matching method based on forward-looking credible regions as claimed in claim 1, characterized in that: Trajectory point p f The method for determining the trusted area in the area circle is as follows: Get the trajectory point p in the area circle f-3 , trajectory point p f-2 and trajectory point p f-1 The angle range Δθ from the minimum direction angle to the maximum direction angle is expanded to both sides by an angle θ′, and the area formed is the trajectory point p f Trusted region within the region circle.

3. The map matching method based on forward-looking credible regions as claimed in claim 1, characterized in that: Before step S1, the method further includes: S4. Interpolate the original trajectory with missing trajectories to form trajectory T.

4. The map matching method based on forward-looking credible regions as claimed in claim 3, characterized in that: The specific method for detecting missing tracks is as follows: Calculate two adjacent trajectory points p i+1 、p i If the time interval and distance interval satisfy the following conditions, then the two trajectory points p are determined. i+1 、p i There are missing trajectories between them; p i+1 (t)-p i (t)≥T max p i+1 (d)-p i (d)≥D max Among them, p i+1 (t), p i (t) represents the timestamp of the i+1th and ith trajectory points respectively, p i+1 (d) p i (d) represent the positions of the i+1th and ith trajectory points, respectively, T max represents the threshold of the time interval, D max Indicates the threshold value of the distance interval.

5. The map matching method based on forward-looking credible regions as claimed in claim 3, characterized in that: The trajectory interpolation method between two trajectory points is as follows: S41, inserting a new trajectory point between two adjacent trajectory points with missing trajectories, the position of the new trajectory point is located at the midpoint of the line connecting the adjacent trajectory points, and the time is the median of the timestamps of the adjacent trajectory points; S42, re-detecting whether there is a missing track between the new track point and two adjacent track points. If the detection result is yes, executing step S41 until there is no missing track between all adjacent track points.

6. The map matching method based on forward-looking credible regions as claimed in claim 3, characterized in that: Before detecting missing tracks, the following steps are also included: Delete the trajectory points with abnormal speed and position in the original trajectory.

Images