High-frequency vehicle track map matching method

Abstract

The invention discloses a high-frequency vehicle track map matching method, which comprises the following steps of: (1) dividing a track T to be matched into a plurality of sub-tracks based on a reference line distance, and extracting key track points in the sub-tracks; (2) extracting a road where the key track point is located and a road type of the road, filtering the extracted road based on the road type, and storing filtered road data by adopting an R-Tree structure; and (3) matching an optimal matching path of the track T to be matched from the road data stored in the R-Tree structure through a hidden Markov probability matching model. From the perspective of reducing the scale of original data, trajectory and road network data are simplified. Then, space and direction analysis is carried out on the simplified trajectory, a micro-region splicing strategy is adopted to accelerate shortest path search, and a path deviation function is designed to improve the topology of a matching result; in addition, the extraction process of track motion key points is refined by fusing more road semantic information, so that a better matching result is obtained.

Description

Technical Field

[0001] The present invention belongs to the technical field of trajectory matching, and more specifically, the present invention relates to a high-frequency vehicle trajectory map matching method. Background Art

[0002] In the field of intelligent transportation systems, map matching algorithms are used as a preprocessing step for trajectory data. They aim to match the actual observed location data with the road network to determine the real location of the moving object and provide relevant guidance information. This algorithm is widely used in vehicle navigation systems, intelligent traffic management, path planning, and location services.

[0003] Due to the popularity of global positioning system (GPS) devices, the scale of floating vehicle trajectory data has grown rapidly. Although high-frequency sampling data can obtain more fine-grained location information of mobile vehicles, massive trajectory data is prone to high storage and processing costs. Existing map matching algorithms limit the computational overhead by generating a limited number of road candidates, or use incremental matching algorithms to dynamically update the matching positions of newly added data, thereby avoiding repeated calculations of the entire data set and reducing resource waste.

[0004] In order to further solve the performance overhead problem caused by dense high-frequency trajectory data, a trajectory compression method is used to reduce the data size to improve the matching speed. However, it relies on a fixed threshold for compression in the global structure and ignores the balance of compression ratios of trajectory segments of different shapes, resulting in over-simplification or insufficient retention of trajectory morphological features, thereby causing a decrease in matching accuracy. Summary of the invention

[0005] The present invention provides a high-frequency vehicle trajectory map matching method, aiming to improve the above-mentioned problem.

[0006] The present invention is implemented as follows: a high-frequency vehicle trajectory map matching method, the method comprising the following steps:

[0007] (1) Segment the to-be-matched trajectory T into several sub-trajectories based on the baseline distance, and extract the key trajectory points in the sub-trajectories;

[0008] (2) Extract the roads where the key trajectory points are located and their road types, filter the extracted roads based on the road types, and then use the R-Tree structure to store the filtered road data;

[0009] (3) The best matching path of the trajectory T to be matched is matched from the road data stored in the R-Tree structure through the hidden Markov probability matching model.

[0010] Furthermore, the sub-trajectory division method is as follows:

[0011] (11) The straight line where the fixed point and the floating point are located is taken as the reference line;

[0012] (12) Calculate the distance between the trajectory point and the baseline between the fixed point and the floating point. If the distance is less than the set distance threshold Then the floating point moves back one trajectory point and executes step (11). If there is a distance greater than the set distance threshold Then execute step (13);

[0013] (13) Taking the previous trajectory point of the floating point as the fixed point, the floating point moves back one trajectory point and executes step (11);

[0014] (14) Traverse the to-be-matched trajectory T, extract fixed points in the to-be-matched trajectory T, use the fixed points as segmentation points, and segment the to-be-matched trajectory T into a number of sub-trajectories.

[0015] Furthermore, the key trajectory point extraction process in the sub-trajectory is as follows:

[0016] (15) Take the first trajectory point p in the sub-trajectory s and the last trajectory point p k is the endpoint of the baseline;

[0017] (16) Traverse the trajectory point p s With trajectory point p k Other trajectory points p between z , calculate all other trajectory points p z The offset distance in the direction of the reference line;

[0018] (17) Calculate all trajectory points p z The average value of the offset distance relative to the baseline is used as the average offset distance of the corresponding sub-track, and other track points with offset distances greater than the average offset distance are used as key track points of the corresponding sub-track.

[0019] Furthermore, the trajectory point p z Offset distance in the direction of the baseline The calculation formula is as follows:

[0020]

[0021] Among them, ||p z-1 , p z || great circle is the trajectory point p z-1 , trajectory point p z The straight-line distance between represents the trajectory point p z-1 , trajectory point p z Straight line With baseline The angle between.

[0022] Furthermore, the road type determination process is as follows:

[0023] The road closest to each key trajectory point is found in the surrounding area of ​​each key trajectory point, the road type of the nearest road is put into the road list, and the road type that is not passable by the vehicle is removed from the road list.

[0024] Furthermore, R-Tree is used to store the road segment identification ID of each road in the road list and the minimum bounding rectangle MBR of each road, and a dictionary structure is introduced to store the geometric information and related attributes of the corresponding road, with the road segment identification ID as the key value for association.

[0025] Furthermore, the total probability of the path is calculated based on the observation probability and the transition probability, and the path with the highest total probability is taken as the best matching path of the trajectory T to be matched. The transition probability is the product of the transition probability obtained based on the speed analysis and the transition probability obtained based on the path deviation analysis.

[0026] Transition probability from velocity analysis The calculation formula is as follows:

[0027]

[0028] Among them, v k Indicates road segment e k The speed limit on the road, Indicates candidate points To the candidate point The average driving speed, is the trajectory point p i On section e k The candidate points on is the trajectory point p i+1 On section e j The candidate point, m represents the number of candidate points To the candidate point The total number of segments contained in the shortest path.

[0029] Furthermore, the transition probability obtained from the path deviation analysis The calculation formula is as follows:

[0030]

[0031] in, From the candidate point arrive The average driving speed, From the candidate point To the candidate point The driving time, From the candidate point To the candidate point The shortest path of Indicates candidate points To the candidate point The Euclidean distance of .

[0032] Furthermore, from the candidate point To the candidate point The shortest path search method is as follows:

[0033] According to the candidate point The corresponding trajectory point p i , the next candidate point The corresponding trajectory point p i+1 The average driving speed and time difference determine the driving distance D d , with the current trajectory point p i and the next trajectory point p i+1 The midpoint p m Centered by D d Add the error parameter ε d As the length of the micro-region, a directed graph G is constructed for the road network nodes in the micro-region, and the Dijkstra algorithm is used to search for candidate points in the directed graph G. To the candidate point The shortest path.

[0034] Furthermore, the driving distance D d The calculation formula is as follows:

[0035]

[0036] in, is the trajectory point p i , trajectory point p i+1 The average speed, (p i+1 .tp i .t) is the trajectory point p i To the trajectory point p i+1 The driving time, p i+1 .t、p i .t represent the trajectory point p i , trajectory point p i+1 Timestamp, len(p i →p i+ 10 represents the trajectory point p i To the trajectory point p i+1 The straight-line distance, len(p i-1 →p i ) represents the trajectory point p i-1 To the trajectory point p iThe straight-line distance, ε d is the error parameter.

[0037] The high-frequency vehicle trajectory map matching method proposed in this paper has the following beneficial effects:

[0038] (1) Trajectory simplification method based on context information simplifies trajectory data: The trajectory segmentation method based on baseline distance is used to divide the sub-trajectory, and the motion key points are obtained using the average offset distance of each sub-trajectory, which reduces the matching data input while maintaining the trajectory shape;

[0039] (2) Filtering road network data based on road network semantics: By matching the key points of motion to the neighboring road types, the road types of the neighboring road sections are detected. On this basis, the roads that are logically inaccessible to vehicles are eliminated to reduce the retrieval overhead caused by invalid road sections. At the same time, based on the R-tree index, an auxiliary storage structure is added to further improve the spatial query efficiency.

[0040] (3) Adopt a micro-region splicing strategy to improve the efficiency of the shortest path query. Introduce a path deviation function to improve the topology of the result, fully consider the relationship between the shortest path length and the actual driving distance, and weight the transition probability of adjacent sampling points to ensure the robustness of the intersection section matching.

[0041] In summary, the present invention improves efficiency while maintaining matching quality through trajectory simplification and road network filtering. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 A flowchart of a high-frequency vehicle trajectory map matching method provided by an embodiment of the present invention;

[0043] Figure 2 A schematic diagram of the baseline distance provided by an embodiment of the present invention;

[0044] Figure 3 A schematic diagram of micro-region splicing provided by an embodiment of the present invention;

[0045] Figure 4 An example diagram of incorrect matching of turning sections provided by an embodiment of the present invention;

[0046] Figure 5 The distance threshold provided by the embodiment of the present invention Schematic diagram of the impact on running time;

[0047] Figure 6 The distance threshold provided by the embodiment of the present invention Schematic diagram of the impact on matching accuracy;

[0048] Figure 7 An example diagram of semantic filtering of a road network provided by an embodiment of the present invention, wherein (a) is the original road network and (b) is the filtered road network;

[0049] Figure 8 The efficiency of the algorithm based on trajectory semantic road network filtering provided by the embodiment of the present invention, wherein (a) is a single point matching time comparison diagram, and (b) is a comparison diagram of different trajectory running times;

[0050] Fig. 9 A comparison chart of matching accuracy of different algorithms provided by the embodiments of the present invention;

[0051] Fig.10 A comparison of the running time of different algorithms provided in the embodiments of the present invention, wherein (a) is a comparison diagram of the running time of the original road network, and (b) is a comparison diagram of the running time of the filtered road network;

[0052] Fig.11 Visualization results of different algorithms for a circular section provided by an embodiment of the present invention:

[0053] Fig.12 Visualization results of different algorithms for turning sections provided by embodiments of the present invention:

[0054] Fig.13 Visualization results of different algorithms for parallel road sections provided by embodiments of the present invention. DETAILED DESCRIPTION

[0055] 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.

[0056] Road network: A road network is defined as a directed graph G(V,E), where V = {v1,v2,…,v n} is the vertex set consisting of road endpoints and intersections in the road network, E = {e1, e2..., e j} represents the directed edge set of road segments in the road network, where e j It is a road segment between two nodes in the road network, and each road segment has a corresponding road type attribute.

[0057] Path P: Path P is a sequence of connected road segments, defined as: P = {e1 → e2 → e3 → ... → e j}.

[0058] Track point: A track point is a sampling point collected by a GPS device to record the location information and time information of a moving object, defined as p i = {lat i ,lon i ,t i}, where lat i ,loni Respectively represent the latitude and longitude of the location coordinates, t i Indicates the current sampling time.

[0059] Trajectory: Trajectory refers to a sequence of trajectory points obtained by continuous sampling of the same moving object at a certain time interval, defined as T = {p1, p2, ..., p n}, where n represents the number of sampling points contained in the trajectory T.

[0060] Baseline: The baseline is the line between two specific reference trajectory points p m 、p n The straight line formed by connecting is used to evaluate the position deviation of other points, denoted as

[0061] Baseline distance: track point p i With baseline Distance, recorded as Take trajectory point p m 、p n The straight line is the reference line. When the trajectory point p i Project to baseline The obtained projection point is located at the trajectory point p m With trajectory point p n On the line segment where is the trajectory point p i To Baseline The projection distance; when the trajectory point p i Project to baseline The obtained projection point is not located at the trajectory point p m With trajectory point p n On the line segment where The value is the trajectory point p i From the trajectory point p m 、p n The shortest great circle distance is like Figure 2 shown.

[0062] The trajectory map matching method provided by the present invention includes: trajectory data processing, road network data processing and matching analysis. In the trajectory data processing stage, the noise trajectory is preliminarily processed, the baseline distance of the trajectory point is used to segment the trajectory, and the average offset distance of the sub-trajectory is obtained to screen the effective motion key points; in the road network data processing stage, the global road network is filtered using the road network semantic information and R-Tree index, and an auxiliary storage structure is designed to shorten the candidate road section retrieval time; in the matching analysis stage, an innovative micro-area splicing strategy is adopted to construct a shortest path search micro-area to accelerate the route distance search, and secondly, the speed analysis function is defined in combination with the road speed limit, and based on the actual situation that the actual driving path is consistent with the correct matching path length, a path deviation function is constructed to calculate the weighted transfer probability. Finally, the Viterbi algorithm is used to generate the best path, such as Figure 1 As shown;

[0063] (1) Segment the to-be-matched trajectory T into several sub-trajectories based on the baseline distance, and extract the key trajectory points in the sub-trajectories;

[0064] The collected original track T to be matched is preprocessed to remove redundant and abnormal track data. In order to ensure high adaptability in different road scenes (circular, turning or straight road sections), the present invention uses the baseline distance of context track points to segment the track, so that the continuous GPS points in each sub-track have similar headings and small position deviations. The removal process of abnormal track points based on the baseline distance is as follows:

[0065] (11) The straight line where the fixed point and the floating point are located is taken as the reference line;

[0066] (12) Calculate the distance between the trajectory point and the baseline between the fixed point and the floating point. If the distance is less than the set distance threshold Then the floating point moves back one trajectory point and executes step (11). If there is a distance greater than the set distance threshold Then execute step (13);

[0067] (13) Taking the previous trajectory point of the floating point as the fixed point, the floating point moves back one trajectory point and executes step (11);

[0068] (14) Traverse the to-be-matched trajectory T, extract fixed points in the to-be-matched trajectory T, use the fixed points as segmentation points, and segment the to-be-matched trajectory T into a number of sub-trajectories.

[0069] For the trajectory to be matched T = {p i ,p i+1 ,...,p m}, with trajectory point p i is a fixed point, and the trajectory point p i+2 For floating point, with fixed point p iand floating point p i+2 The straight line where is located is the baseline, and the trajectory T to be matched is traversed based on the above steps (11) to (14).

[0070] Existing trajectory compression algorithms usually use a single distance threshold as a simplification parameter, which makes it difficult to balance the distribution of trajectory points under different road shapes, resulting in sparse trajectories on straight sections and dense trajectories on circular sections. If the threshold is set too large, the simplified trajectory may lose key geometric features (such as turning points). If it is set too small, too many trajectory points will be retained, failing to effectively reduce the amount of data and speed up the calculation process. Therefore, in order to overcome the problem of unbalanced trajectory simplification ratio under different road shapes, the present invention extracts key trajectory points by calculating the average offset distance of sub-trajectories, simplifies the trajectory while ensuring the original structure of each sub-trajectory, and the key trajectory point extraction process in the sub-trajectory is as follows:

[0071] (15) Take the first trajectory point p in the sub-trajectory s and the last trajectory point p k is the endpoint of the baseline;

[0072] (16) Traverse the trajectory point p s With trajectory point p k Other trajectory points p between z , calculate all other trajectory points p z Offset distance in the direction of the baseline The calculation formula is as follows:

[0073]

[0074] Among them, ||p z-1 , p z || great circle is the trajectory point p z-1 , trajectory point p z The straight-line distance between represents the trajectory point p z-1 , trajectory point p z Straight line With baseline The angle between.

[0075] (17) Calculate all trajectory points p z The average value of the offset distance relative to the baseline d s (p s →p k ), the average value is taken as the average offset distance of the corresponding sub-track, and other track points with offset distances greater than the average offset distance are taken as the key track points of the corresponding sub-track.

[0076]

[0077] Among them, u is the adjustment factor, n is the number of points in the sub-trajectory except the trajectory point p s , trajectory point p k The number of other trajectory points.

[0078] (2) Extract the roads where the key trajectory points are located and their road types, filter the extracted roads based on the road types, and then use the R-Tree structure to store the filtered road data;

[0079] R-Tree is widely used to improve the efficiency of road network query. However, with the growth of data scale, its query cost and storage overhead increase significantly. Therefore, the present invention introduces road network semantic information on the basis of R-Tree, filters the global road network by road type, and quickly queries candidate nodes around trajectory points. The specific process of determining road type is as follows:

[0080] (21) Find the road closest to each key trajectory point in the surrounding area, put the road type of the nearest road into the road list, and remove the road types that are not passable by vehicles (such as tracks, waterways, sidewalks, etc.) from the road list;

[0081] In an embodiment of the present invention, a distance threshold ε of the surrounding area is set. When the distance between a road section and a key trajectory point is greater than the distance threshold ε, the corresponding road section is ignored, and the road section closest to the key trajectory point is found among the road sections within the distance threshold ε, and the road type of the road section is extracted.

[0082] (22) R-Tree is used to store the road segment identification ID of each road in the road list and the minimum bounding rectangle MBR of each road, and a dictionary structure is introduced to store the geometric information and related attributes (such as speed limit, etc.) of the corresponding road, using the road segment identification ID as the key value for association.

[0083] Finally, in the candidate search phase, the trajectory point coordinates and error radius are used to confirm the candidate road segment set. The projection of the trajectory point on the candidate road segment is calculated. If the projection point is within the valid range of the road segment, the projection point is considered as a candidate point; otherwise, the endpoint closest to the trajectory point on the road segment is selected as a candidate point.

[0084] (3) The best matching path of the trajectory T to be matched is matched from the road data stored in the R-Tree structure through the HMM model (Hidden Markov probability matching model).

[0085] The total probability of the path is calculated based on the observation probability and the transition probability, and the path with the highest total probability is taken as the best matching path for the trajectory T to be matched. The total probability function is as follows:

[0086]

[0087] in, Candidate point To the candidate point The transition probability is, is the trajectory point p i On section e k The candidate points on is the trajectory point p i+1 On section e j The candidate points is the trajectory point p i+1 On section e j The probability of observation on is the trajectory point p1 on section e s The probability of observation on .

[0088] (31) Observation probability;

[0089] The observation probability calculation needs to consider the relationship between the trajectory point and the candidate road section, including the distance position difference and the angle difference between the direction of the trajectory point and the direction of the candidate road section. Therefore, the observation probability is defined by combining the distance analysis function and the direction analysis function. The calculation method and The same, I will not repeat it here, the observation probability for:

[0090]

[0091] Distance analysis function: in, is the trajectory point p i+1 On section e j The candidate points on represents the trajectory point p i+1 and candidate points The great circle distance between d Indicates the standard deviation of the distance.

[0092] Directional analysis function: in, represents the trajectory point p i+1 Driving direction and candidate points The angle difference between the directions of the road sections, μ θ and σ θ They represent the mean and standard deviation of the angle difference respectively.

[0093] (32)Transition probability: is the transition probability based on speed analysis, is the transition probability obtained based on path deviation analysis.

[0094] When the HMM algorithm is applied to complex large-scale road maps, the performance bottleneck occurs in the process of searching for the shortest path. When the Dijkstra algorithm is looking for the shortest path between candidate points corresponding to adjacent trajectory points, it needs to traverse the road sections in the entire road network, even invalid sections far away from the GPS point. Therefore, the search efficiency of the algorithm is limited by the scale of the map, which can easily cause large computational overhead and waste of resources. Therefore, the present invention uses the micro-region splicing method to decompose the road network into smaller parts, obtain the sub-road network of the current front and rear trajectory points being matched, and then construct a micro-region directed graph to search for the shortest path.

[0095] First, according to the current trajectory point p i , the next trajectory point p i+1 Average driving speed The time difference (p i+1 .tp i .t) Determine the driving distance D d , with the current trajectory point p i and the next trajectory point p i+1 The midpoint p m Centered by D d Add the error parameter ε d As the length of the micro-area, a directed graph G is constructed for the road network nodes in the micro-area, and the Dijkstra algorithm is used to search for the shortest path in the directed graph G to reduce the traversal of invalid sections. d The calculation formula is as follows:

[0096]

[0097] in, is the trajectory point p i , trajectory point p i+1 The average speed, (p i+1 .tp i .t) is the trajectory point p i To the trajectory point p i+1 The driving time, p i+1 .t、p i .t represent the trajectory point p i , trajectory point p i+1 Timestamp, len(p i →p i+1 ) represents the trajectory point p i To the trajectory point p i+1 The straight-line distance, len(p i-1 →p i ) represents the trajectory point p i-1 To the trajectory point p i The straight-line distance, ε dis the error parameter, which is set to 100m, which is approximately the distance a car traveling at 75 km / h travels in 5 seconds. Since the trajectory points have been simplified, the search range is expanded by 100m on the original basis to ensure that there is a drivable path between the trajectory points.

[0098] (321) Transition probability obtained from velocity analysis

[0099] In a complex road network containing multiple sections with different speed limits, when the roads are parallel and close, it is difficult to distinguish the actual driving path of the vehicle through spatial features. However, the speed limits of different types of sections are different. Correctly matching sections tend to choose sections with driving speeds more similar to the speed limits. The average speed of adjacent trajectory points and the speed limit of the section included in the shortest path between candidate point pairs are analyzed by accumulation to avoid the influence of a certain path speed limit or a certain candidate point speed that is too large or too small, which leads to deviations in the calculation results. The speed analysis function is defined as:

[0100]

[0101] Among them, v k Indicates road segment e k The speed limit on the road, Indicates candidate points To the candidate point The average driving speed, is the trajectory point p i On section e k The candidate points on is the trajectory point p i+1 On section e j The candidate point, m represents the number of candidate points To the candidate point The total number of segments contained in the shortest path.

[0102] (322) Transition probability obtained from path deviation analysis

[0103] When driving on a straight road, the HMM method uses the straight-line distance between trajectory points as the reference length of the candidate shortest path, which can be applied to most road scenarios. However, some situations cannot be handled by relying solely on the straight-line distance, such as Figure 5 As shown in the figure. e1, e2, and e3 are three road sections, and the actual driving paths of trajectory points p1 and p2 are {e2, e3}, that is, candidate points (Candidate point of trajectory point p1 on road segment e1) and candidate point The shortest path between (the candidate point of trajectory point p2 on segment e2). If the straight-line distance is used as the path reference, e1 will be incorrectly selected as the matching segment because the candidate point (Candidate point of trajectory point p1 on road segment e2) and candidate point The distance between (the candidate point of trajectory point p2 on road segment e1) is more similar to the straight-line distance, which leads to the neglect of the actual driving topology when matching the turning road segment. However, based on the actual situation, the correct candidate shortest path should not only be close to the actual driving path in space, but also correspond to it in length. Therefore, the present invention evaluates the probability that the candidate point is the real position by the similarity between the shortest path and the actual driving distance. The actual driving distance is simulated by the product of the average speed of the moving object and the driving time, and the path deviation function is defined:

[0104]

[0105] in, From the candidate point arrive The average driving speed, From the candidate point To the candidate point The driving time, From the candidate point To the candidate point The shortest path of Indicates candidate points To the candidate point The Euclidean distance of .

[0106] Road network data: The present invention uses the OpenStreetMap digital map covering Beijing and Hefei in China as the road network data source. The road segment classification in this data set includes expressways, main urban roads, urban secondary roads, urban branch roads, residential roads, elevated expressways, sidewalks, rails, waterways, etc.

[0107] Trajectory data: Trajectory data is collected through public datasets and GPS applications on smartphones. The attributes of GPS points include timestamp, longitude, and latitude. The sampling interval of this dataset is 1s. Table 1 shows the detailed information of 7 randomly selected trajectory data, including the number of trajectory points, trajectory length, and the number of trajectory points after context simplification.

[0108] Table 1 Trajectory data information

[0109]

[0110] In order to verify the effectiveness of the high-frequency trajectory matching method provided by the present invention, the high-frequency trajectory matching method (HFIMM for short) provided by the present invention is compared with the HMM, STD-Matching and RCIVMM algorithms, and evaluated according to the following indicators:

[0111] 1) Time efficiency: By comparing the running time of different algorithms, the differences in algorithm execution speed can be reflected;

[0112] 2) Matching efficiency: the ratio of the number of correctly matched trajectory points to the total number of trajectory points

[0113] The present invention sets the baseline distance threshold This parameter directly affects the density of trajectory segmentation points, and thus affects the subsequent trajectory simplification and matching process. Figure 5 , Figure 6 Display different distance thresholds separately The running time and matching accuracy under . Figure 5 Algorithm running time and distance threshold is inversely proportional to As a result, the number of motion key points extracted subsequently decreases, and the data size required for candidate retrieval also decreases accordingly. Figure 6 When the display distance threshold is 0.5 and 3, the trajectory matching accuracy is high, but when When , more trajectory points are retained, compared with The running time increases by about 10%, and the accuracy increases by about 1.4%. Therefore, considering the running time and matching accuracy, the present invention sets the distance threshold Set it to 3m for the experiment.

[0114] Figure 7 This is an example diagram of road network semantic filtering. The effectiveness of road network filtering is proved by testing the single point matching time at different sampling rates and the running time on different trajectories. The results are shown in the figure below. Figure 8 As shown in the figure, the single-point matching time is reduced by about 85%, and the overall matching time is reduced by about 88%. The reason is that the road network filtering process eliminates most of the invalid road segments, which reduces the road segment retrieval cost based on the R-tree index, reduces the number of candidate queries, and then the number of paths for subsequent probability matching calculations also decreases, greatly reducing the computational overhead.

[0115] The present invention compares and analyzes the matching accuracy of the algorithm in this paper with that of HMM, STD-Matching and RCIVMM algorithms from the perspective of overall matching effect. Fig. 9As can be seen from Table 2, the accuracy of the trajectory map matching method provided by the present invention is significantly better than that of the other three algorithms. Since the HMM algorithm relies on the matching result of the previous trajectory point, it may not be robust enough in some complex scenarios, and as the sampling frequency decreases, the matching error increases. Similarly, STD-Matching performs directional analysis on each trajectory point and enumerates all possible candidate points, but in complex urban environments, relying solely on distance and direction analysis can easily lead to skipping of trajectory matching, especially near intersections or parallel roads. RCIVMM mainly analyzes through spatiotemporal features and driving behavior, and does not fully utilize the semantic information of the trajectory, which in turn affects the understanding of complex driving behaviors and the continuity of path selection when processing high-frequency trajectories. The HFIMM method relies on efficient trajectory and road network processing steps to streamline trajectory data, help remove redundant noise points, and eliminate irrelevant sections in candidate searches, so as to maintain high accuracy in complex road networks.

[0116] Table 2 Matching accuracy of different algorithms

[0117]

[0118] In order to evaluate the time efficiency of different map matching algorithms and the necessity of road network filtering, this paper tests the running time of different algorithms on the original road network and the filtered road network. The results are as follows: Fig.10 , as shown in Table 3 and Table 4. Among them, the STD method has the highest time complexity, because the STD-Matching method uses the A* algorithm to find the shortest path, but does not limit the search range and does not simplify the trajectory data, resulting in a large number of redundant calculations when enumerating candidate points one by one. The HMM algorithm is based on the state transition model and is often computationally inefficient when faced with high-density trajectory data. RCIVMM uses an interactive voting algorithm based on a weighted matrix to globally select the best matching result, but its computational complexity increases with the increase in the number of candidate paths. In contrast, HFIMM extracts motion change key points through the trajectory structure to reduce trajectory matching input, and based on the R-tree index, combines road network semantic filtering processing, introduces an auxiliary storage structure to improve spatial query efficiency. In addition, Fig.10 (b) It shows that road network filtering processing can effectively improve the operating rate of different algorithms, proving the applicability of HFIMM's road network preprocessing method.

[0119] Table 3 Running time of different algorithms on the original road network

[0120]

[0121]

[0122] Table 4 Running time of different algorithms on the filtering network

[0123]

[0124] Fig.11 This is the visualization result of the round-trip road segment matching. In the HMM map matching method, the current GPS point matching depends on the distance between the previous GPS point and the current GPS point candidate, which is prone to mismatching at the intersection of the round-trip road segment. The STD-Matching method selects candidates based on the spatial and directional probabilities of GPS points, but this method does not fully utilize the correlation between trajectory points, resulting in mismatching to illogical driving routes. The RCIVMM method considers the influence of adjacent points and matches based on the weighted voting information of each candidate path, but does not perform directional analysis on the GPS points, resulting in mismatching to a closer section.

[0125] Fig.12 This is the visualization result of the matching of the turning section. The HMM method, STD-Matching method, and RCIVMM method incorrectly match parallel track sections. The reason is that the matching errors of multiple previous track points are accumulated, causing the algorithm to tend to keep matching to the previous section, even if the actual distance is shorter, so it incorrectly matches to the adjacent parallel section and deviates from the correct driving path. In the process of road network semantic filtering, HFIMM eliminates track sections that are not passable by vehicles to avoid matching track points to logically incorrect sections.

[0126] Fig.13 It is the visualization result of parallel road segment matching. The HMM method matches well under this road condition, but it matches more trajectory points and has a higher overhead than HFIMM. The STD-Matching method calculates the matching probability based on the prior road topology and trajectory distance. It is affected by the fact that the previous trajectory point is closer to the parallel road segment on the other side, resulting in skipping matching. The RCIVMM method is mainly based on distance analysis. When the trajectory data is offset, the vehicle's driving direction is ignored and the road segment with a closer distance is given priority. Compared with the other three methods, the HFIMM method simplifies the original trajectory, eliminates some minor deviation points, and makes full use of the road speed limit information, so it achieves better matching results.

[0127] The present invention improves the map matching algorithm for high-frequency GPS data, greatly improving the accuracy and timeliness of high-frequency trajectory matching. First, from the perspective of reducing the size of the original data, the trajectory and road network data are simplified. Then, the simplified trajectory is analyzed in space and direction, a micro-region splicing strategy is used to accelerate the shortest path search, and a path deviation function is designed to improve the topology of the matching result. In addition, the extraction process of trajectory motion key points will be refined by integrating more road semantic information to obtain better matching results.

[0128] 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 high-frequency vehicle trajectory map matching method, characterized in that: The method comprises the following steps: (1) Segment the to-be-matched trajectory T into several sub-trajectories based on the baseline distance, and extract the key trajectory points in the sub-trajectories; (2) Extract the roads where the key trajectory points are located and their road types, filter the extracted roads based on the road types, and then use the R-Tree structure to store the filtered road data; (3) The best matching path of the trajectory T to be matched is matched from the road data stored in the R-Tree structure through the hidden Markov probability matching model.

2. The high-frequency vehicle trajectory map matching method according to claim 1, characterized in that: The sub-trajectory division method is as follows: (11) The straight line where the fixed point and the floating point are located is taken as the reference line; (12) Calculate the distance between the trajectory point and the baseline between the fixed point and the floating point. If the distance is less than the set distance threshold Then the floating point moves back one trajectory point and executes step (11). If there is a distance greater than the set distance threshold Then execute step (13); (13) Taking the previous trajectory point of the floating point as the fixed point, the floating point moves back one trajectory point and executes step (11); (14) Traverse the to-be-matched trajectory T, extract fixed points in the to-be-matched trajectory T, use the fixed points as segmentation points, and segment the to-be-matched trajectory T into a number of sub-trajectories.

3. The high-frequency vehicle trajectory map matching method according to claim 1, characterized in that: The key trajectory point extraction process in the sub-trajectory is as follows: (15) Take the first trajectory point p in the sub-trajectory s and the last trajectory point p k is the endpoint of the baseline; (16) Traverse trajectory point p s With trajectory point p k Other trajectory points p between z , calculate all other trajectory points p z The offset distance in the direction of the reference line; (17) Calculate all trajectory points p z The average value of the offset distance relative to the baseline is used as the average offset distance of the corresponding sub-track, and other track points with offset distances greater than the average offset distance are used as key track points of the corresponding sub-track.

4. The high-frequency vehicle trajectory map matching method as claimed in claim 3, characterized in that: Trajectory point p z Offset distance in the direction of the baseline The calculation formula is as follows: Among them, ||p z-1 , p z || great circle is the trajectory point p z-1 , trajectory point p z The straight-line distance between represents the trajectory point p z-1 , trajectory point p z Straight line With baseline The angle between.

5. The high-frequency vehicle trajectory map matching method according to claim 1, characterized in that: The process of determining the road type is as follows: The road closest to each key trajectory point is found in the surrounding area of ​​each key trajectory point, the road type of the nearest road is put into the road list, and the road type that is not passable by the vehicle is removed from the road list.

6. The high-frequency vehicle trajectory map matching method as claimed in claim 5, characterized in that: R-Tree is used to store the segment identification ID of each road in the road list and the minimum bounding rectangle MBR of each road, and a dictionary structure is introduced to store the geometric information and related attributes of the corresponding road, with the segment identification ID as the key value for association.

7. The high-frequency vehicle trajectory map matching method as claimed in claim 1, characterized in that: The total probability of the path is calculated based on the observation probability and the transition probability, and the path with the highest total probability is taken as the best matching path of the trajectory T to be matched. The transition probability is the product of the transition probability obtained based on the speed analysis and the transition probability obtained based on the path deviation analysis; Transition probability from velocity analysis The calculation formula is as follows: Among them, v k Indicates road segment e k The speed limit on the road, Indicates candidate points To the candidate point The average driving speed, is the trajectory point p i On section e k The candidate points on is the trajectory point p i+1 On section e j The candidate point, m represents the number of candidate points To the candidate point The total number of segments contained in the shortest path.

8. The high-frequency vehicle trajectory map matching method as claimed in claim 7, characterized in that: Transition probabilities from path deviation analysis The calculation formula is as follows: in, From the candidate point arrive The average driving speed, From the candidate point To the candidate point The driving time, From the candidate point To the candidate point The shortest path of Indicates candidate points To the candidate point The Euclidean distance of .

9. The high-frequency vehicle trajectory map matching method according to claim 7 or 8, characterized in that: From the candidate point To the candidate point The shortest path search method is as follows: According to the candidate point The corresponding trajectory point p i , the next candidate point The corresponding trajectory point p i+1 The average driving speed and time difference determine the driving distance D d , with the current trajectory point p i and the next trajectory point p i+1 The midpoint p m Centered by D d Add the error parameter ε d As the length of the micro-region, a directed graph G is constructed for the road network nodes in the micro-region, and the Dijkstra algorithm is used to search for candidate points in the directed graph G. To the candidate point The shortest path.

10. The high-frequency vehicle trajectory map matching method as claimed in claim 9, characterized in that: Driving distance D d The calculation formula is as follows: in, is the trajectory point p i , trajectory point p i+1 The average speed, (p i+1 .tp i .t) is the trajectory point p i To the trajectory point p i+1 The driving time, p i+1 .t、p i .t represent the trajectory point p i , trajectory point p i+1 Timestamp, len(p i →p i+1 ) represents the trajectory point p i To the trajectory point p i+1 The straight-line distance, len(p i-1 →p i ) represents the trajectory point p i-1 To the trajectory point p i The straight-line distance, ε d is the error parameter.

Images