A high sampling rate trajectory map matching method based on path increment

Abstract

The present invention discloses a high sampling rate trajectory map matching method based on path increment, formally defines the GNSS trajectory point map matching process based on path increment; adopts a combined filtering method based on constraints such as distance, road connectivity, and direction consistency to extract a candidate road subset from a road network; takes the path between two adjacent intersection points in the candidate subnet as an increment, and performs trajectory point matching calculation based on path increment. In the matching process, each incremental calculation process is a process of determining the subsequent matching section with the intersection point as the starting point. In order to reduce the influence of parallel and unidirectional sections on the matching results, a similarity evaluation index combining Hausdorff distance and curvature is comprehensively considered at each intersection point. The present invention shows higher accuracy and efficiency, can better handle various complex road section conditions, and is very suitable for high sampling rate GNSS trajectory matching application scenarios in complex urban road networks.

Description

Technical Field

[0001] The invention relates to a trajectory map matching method, and in particular to a high sampling rate trajectory map matching method based on path increment. Background Art

[0002] The advancement of location acquisition and mobile computing technology has generated a large amount of spatial trajectory data, which represents the mobility of various moving objects and plays a vital role in many fields such as real-time path planning, road network update, and travel pattern discovery. However, since different mobile devices have positioning errors and the road has been transformed from surface elements in the real world to linear elements under an abstract plane through the map synthesis process, there is a deviation between the trajectory and the actual road it is on. Therefore, before processing and analyzing the trajectory data, map matching should be performed so that the trajectory is correctly positioned on the road. In urban application scenarios, the advancement of Global Navigation Satellite System (GNSS) technology and the development of big data processing technology have enabled the positioning data of urban areas to be acquired at very short time intervals, such as 1 second or 5 seconds. How to match high sampling rate trajectories with complex urban road networks while taking into account efficiency and accuracy has become a challenge.

[0003] Existing map matching algorithms can be divided into three categories: local algorithms, incremental algorithms, and global algorithms. Local map matching algorithms only consider a single GNSS point on the trajectory at a time, without considering the relationship between the current trajectory point and the previous and next points. This type of method has the advantages of high efficiency and easy implementation, but matching errors are very likely to occur in practical applications. Incremental map matching algorithms consider adjacent trajectory points, introduce the spatial distribution and movement rules of trajectory points such as topology, motion state, and transition probability into the matching process, thereby improving the performance of the algorithm. However, in the incremental matching algorithm, incorrect matching results of previous points may accumulate and affect the matching of subsequent points. The global map matching algorithm maps the entire trajectory to a path in the road network based on a similarity metric. Since the GNSS points on the trajectory are considered in batches, the error propagation problem of the incremental method will not occur. Therefore, this type of algorithm is not very sensitive to the sampling rate, ensuring high matching accuracy at different sampling rates, but it also leads to low efficiency problems under high sampling rates.

[0004] According to the characteristics of the above-mentioned algorithms, incremental matching ideas are more widely used in application scenarios with high sampling rates and complex urban road networks. However, the existing related algorithms still have the following two problems when processing high sampling rate trajectories: (1) The algorithm matching time is not only related to the number of trajectory points, but also closely related to the complexity of the vehicle driving section. Higher sampling rates and road complexity lead to more time-consuming calculation processes; (2) At complex sections such as intersections, since moving objects usually travel at low speeds, the distance between trajectory points is smaller than in other areas, which is more likely to cause incorrect matching. Summary of the invention

[0005] In order to solve the shortcomings of the above technologies, the present invention provides a high sampling rate trajectory map matching method based on path increment, which further improves the calculation efficiency and the matching accuracy in complex road sections.

[0006] In order to solve the above technical problems, the technical solution adopted by the present invention is: a high sampling rate trajectory map matching method based on path increment, comprising the following steps:

[0007] Step S1, formally defining a GNSS trajectory point map matching process based on path increment;

[0008] Step S2, extracting candidate road subnetworks from the road network using a multi-constraint combination filtering method;

[0009] Step S3: Taking the path between two adjacent intersection points in the candidate road subnet as an increment, a trajectory point matching calculation based on the path increment is performed.

[0010] Preferably, in step S1, the GNSS trajectory point map matching process based on path increment is defined as: according to the trajectory T and the road network G(V,E), a candidate road subnetwork G′(V′, E′) is screened out, and then the path between two adjacent intersection points in the candidate road subnetwork is used as an increment to perform trajectory matching calculation, thereby gradually determining the actual motion path of the object.

[0011] Preferably, the specific process of step S2 is: combining the spatial distribution and movement rules of two adjacent trajectory points with the characteristics of their nearby roads, and removing a large number of irrelevant road sections by adopting a series of filtering conditions such as distance constraints, road connectivity constraints, direction consistency constraints and single-point connected road section "pruning", so as to obtain a candidate road subnet G′(V′, E′).

[0012] Preferably, in step S2, the distance constraint is implemented by a dynamically adjusted buffer zone, that is, taking the trajectory point as the center and r as the center. p0 Build a buffer for the initial radius. If the candidate segment set is empty, expand the radius by a fixed step size r. p.step, and repeat this cycle until the candidate segment set contains at least one segment.

[0013] Preferably, in step S2, the road connectivity constraint includes a topological connectivity constraint and a directional connectivity constraint, wherein the topological connectivity constraint: for any point p in the trajectory i , its candidate road segment set C(p i ) has segment e i and the candidate road segment set C(p i+1 ) in a certain section e j The endpoints are connected or overlap;

[0014] Directional connectivity constraint: For any point p in the trajectory i , candidate road segment set C(p i ) has segment e i and the candidate road segment set C(p i+1 ) in a certain section e j connected or overlapping, and when e i With e j When connected, the directions of the two sections should be connected end to end at the connection point.

[0015] Preferably, in step S2, the direction consistency constraint includes a length ratio constraint and an azimuth constraint, wherein the length ratio constraint: if the length ratio between the distance between two consecutive trajectory points and the distance between their projection points on the candidate road section is less than a threshold value of 0.8, the candidate road section is eliminated;

[0016] Azimuth constraint: Calculate the angle between the trajectory segment and its projection segment on the candidate path. If the angle is greater than 90°, the candidate segment is eliminated.

[0017] Preferably, in the azimuth constraint, the path in the opposite direction is eliminated by calculating the difference between the vector azimuth between the trajectory and the projection point of the trajectory on the candidate path, where the vector azimuth is the horizontal angle passed by the current vector from the north direction to the clockwise rotation, and its calculation formula is: Among them, α is the azimuth of the current vector, (x1, y1) and (x2, y2) are the coordinates of the first and last points of the vector respectively.

[0018] Preferably, in step S2, "pruning" of single-point connected road segments refers to: traversing each road segment in turn, calculating the degree of the starting point and the end point of each road segment respectively, and deleting other single-point connected road segments except the first and last road segments.

[0019] Preferably, the specific process of step S3 is:

[0020] First, determine the initial segment, that is, select the segment with the highest similarity to the trajectory line from the candidate segments of the starting trajectory point as the starting segment;

[0021] After the initial road segment is determined, the incremental calculation process begins. Each incremental calculation process starts from the intersection point and gradually determines the subsequent matching road segments.

[0022] When the vertex connected by the matching road segment is a transition point, the current road segment is recorded and the next road segment is determined;

[0023] When the vertex connected by the matched road segment is an intersection point, the next incremental calculation process is entered; until a single connected point appears, it is used as the end point of the trajectory and the matching process ends.

[0024] Preferably, in step S3, when the incremental forward direction is determined at the intersection, a similarity evaluation index combining the Hausdorff distance and the curvature is used to reduce the influence of parallel sections in the same direction on the matching result.

[0025] The present invention not only improves the matching efficiency, but also takes into account the processing of complex road sections, and better solves the contradiction between accuracy and efficiency in the matching of high sampling rate data in complex road networks. The method eliminates the influence of disconnected road sections and reverse road sections on the matching process by combined filtering of the road network, and uses path increments to replace the trajectory increments in the traditional incremental matching method, so that the matching results are more consistent with the actual vehicle driving route. At the intersection point, the combination of curvature integral and Hausdorff distance is used to replace the geometric metric in the traditional matching algorithm to improve the matching accuracy at the starting section and the intersection point. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 It is a flow chart of the present invention.

[0027] Figure 2 This is a schematic diagram of the road network filtering result in step S2.

[0028] Figure 3 This is a flowchart of filtering the road network in step S2.

[0029] Figure 4 Schematic diagram of the directional connectivity constraint in step S2.

[0030] Figure 5 Schematic diagram of filtering the road network by the direction consistency constraint in step S2.

[0031] Figure 6 FIG. 4 is a flowchart of trajectory matching in step S3.

[0032] Figure 7 This is the pseudo code of the algorithm for determining the starting section in step S3.

[0033] Figure 8 This is the pseudo code for the incremental matching algorithm in step S3.

[0034] Fig. 9 It is the pseudo code of the incremental matching algorithm of the road segment at the intersection point in step S3.

[0035] Fig.10 Schematic diagram of the matching process at the intersection in step S3.

[0036] Fig.11 Schematic diagram of candidate point position determination in step S3.

[0037] Fig.12 FIG. 4 is a display diagram of the test track data of the embodiment in a map.

[0038] Fig.13 It is the accuracy evaluation and comparison chart of the map matching experiment in the embodiment.

[0039] Fig.14 This is a matching result diagram of the test track 1 in the embodiment.

[0040] Fig.15 This is a matching result diagram of test track 2 in the embodiment.

[0041] Fig.16 This is a matching result diagram of test track 3 in the embodiment.

[0042] Fig.17 This is a matching result diagram of test track 4 in the embodiment.

[0043] Fig.18 Schematic diagram for comparing the efficiency of various algorithms in the embodiments.

[0044] Fig.19 Schematic diagram of the number of calls to each algorithm matching unit in the embodiment. DETAILED DESCRIPTION

[0045] The present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0046] like Figure 1 A high sampling rate trajectory map matching method based on path increment is shown, comprising the following steps:

[0047] Step S1, formally define the GNSS trajectory point map matching process based on path increment as: using trajectory T to preliminarily screen out candidate road subnetworks G′(V′, E′) from the road network G(V, E), and then use the path between two adjacent intersection points in the subnetwork as an increment to perform trajectory matching calculation, thereby gradually determining the actual motion path of the object.

[0048] The GNSS trajectory point map matching process based on path increment is formally defined as follows:

[0049] The trajectory T is the set of location points collected by the mobile device, which can be expressed as T = {p i |i=1,...,N},where p i is a trajectory point. The road network is a digital expression of the real road system and can be described by a directed graph G(V, E), where V is a set of vertices {v j |j=1,...,M}, E is the set of edges {e k |k=1,...,S}, vertex v j The number of associated edges is called vertex v j The degree of j ). For any vertex v j , if d(v j )=1, called v j is a simply connected point; if d(v j )=2, called v j is the transition point; if d(v j )≥3, called v j is a road intersection. The edges in G(V, E) are called road segments, and the path P is a set of connected road segments, represented by P = {e o ,…,e j}. The candidate trajectory point is the point that falls on the vertex v j The set of trajectory points in the neighboring area is denoted as I(v j ), referred to as candidate points. The candidate road segment set is the set of i The set of sections where the error distribution range intersects is represented by C(p i ). The road subnetwork G′(V′, E′) is the road network used for map matching after filtering, that is, satisfy The path between two adjacent intersection points is defined as an increment. In the present invention, the trajectory point map matching process based on the path increment is defined as: according to the trajectory T and the road network G(V, E), the road subnetwork G′(V′, E′) is screened out, and then the incremental calculation is performed to finally determine the actual motion path of the object.

[0050] Step S2: extracting candidate road subnetworks from the road network using a multi-constraint combined filtering method.

[0051] The difficulty of map matching in complex urban road networks lies in matching intersections and parallel roads. Trajectories in this area often have multiple candidate sections with similar shapes and close distances, which can easily lead to mismatches. Therefore, the spatial distribution and movement patterns of two adjacent trajectory points are combined with the connectivity, direction and other characteristics of the nearby roads to filter the candidate sections and eliminate a large number of irrelevant sections. Figure 2The figure shows a schematic diagram of the road network filtering results, where the original road network is filtered through a series of conditions such as distance constraint, road connectivity constraint, direction consistency constraint and "single point connected road segment" pruning to obtain a candidate road subnetwork that matches the trajectory point.

[0052] Specifically, Figure 3 As shown, the execution steps of the multi-constraint combined filtering method for the road network include:

[0053] S21, distance constraint is realized by a dynamically adjusted buffer zone, i.e., taking the trajectory point as the center and r as the center. p0 Build a buffer for the initial radius. If the candidate segment set is empty, expand the radius by a fixed step size r. p.step , and repeat this cycle until the candidate segment set contains at least one segment.

[0054] S22. Road connectivity constraints include topological connectivity constraints and directional connectivity constraints.

[0055] Topological connectivity constraint: For any point p in the trajectory i , its candidate road segment set C(p i ) has segment e i and the candidate road segment set C(p i+1 ) in a certain section e j The endpoints are connected or overlap;

[0056] Directional connectivity constraint: For any point p in the trajectory i , candidate road segment set C(p i ) has segment e i and the candidate road segment set C(p i+1 ) in a certain section e j connected or overlapping, and when e i With e j When connected, the directions of the two sections should be connected end to end at the connection point. Figure 4 As shown in (a), for a one-way path, the road segment can be abstracted as a set of directed line segments. If the directions of this set of line segments are consistent, the directional connectivity constraint is satisfied. Figure 4 (b) reflects two types of situations that meet topological connectivity but do not have directional connectivity. Such road segments should be filtered out. Since bidirectional road segments have no directional restrictions, we only need to consider the directional connectivity constraints when unidirectional road segments are connected to each other.

[0057] S23. Directional consistency constraints include length ratio constraints and azimuth constraints.

[0058] Length ratio constraint: If the length ratio between the distance between two consecutive trajectory points and the distance between their projection points on the candidate segment is less than the threshold of 0.8, the candidate segment is eliminated.

[0059] like Figure 5 As shown in (a), p1 and p2 are two consecutive trajectory points, P1, P2 and P3 are three paths that contain candidate sections, and these three paths can all be used as matching paths for p1 and p2. The projections of p1 and p2 on their corresponding candidate sections are the actual positions of the trajectory points. For trajectories with high sampling rates, their distribution is usually approximately parallel to the actual sections. Based on this understanding, before comparing the trajectory direction with the section direction, the horizontal sections are first eliminated through the length ratio of the distance between consecutive trajectory points and the distance between their projection points on the candidate sections, while avoiding the situation where the trajectory order is inconsistent with the corresponding candidate section order. According to experiments, when the ratio is set to <0.8, a better elimination effect can be achieved. Figure 4 In (a), the length Len3 after the trajectory point is projected to P3 is much smaller than the distance between the trajectory points, so the path can be eliminated.

[0060] Azimuth constraint: Calculate the angle between the trajectory segment and its projection segment on the candidate path. If the angle is greater than 90°, the candidate segment is eliminated.

[0061] like Figure 5 As shown in (b), after completing the length ratio constraint filtering, what remains are two paths in opposite directions. The paths in opposite directions can be eliminated by calculating the difference between the trajectory and the vector azimuth between the projection points of the trajectory on the path. The vector azimuth is the horizontal angle from the north direction to the current vector in clockwise rotation. The calculation formula is: Among them, α is the azimuth of the current vector, (x1, y1) and (x2, y2) are the coordinates of the first and last points of the vector respectively.

[0062] The specific implementation process of direction filtering is as follows: ① First, obtain the trajectory points p i , p i+1 On path P j ② Based on the direction of the trajectory and path, calculate The azimuth angle α0 and the azimuth angle of its projection point ③When When the path P j Be eliminated. Figure 5 As shown in (b), we calculate The azimuth angle can eliminate the reverse path P1, Figure 5 (c) shows the result after elimination, where only the paths with the same direction as the trajectory are retained.

[0063] S24, "pruning" of single-point connected sections means: traversing each section in turn, calculating the degree of the starting point and the end point of each section respectively, and deleting other single-point connected sections except the first and last sections. In the subnet composed of the remaining sections, in addition to the first and last sections, if there are other sections with vertex degrees of 1, they can be "pruned" because such single-point connected sections cannot be used as intermediate sections on the final matching path. Figure 2 There are still single-point connected sections in the road network after filtering by road connectivity and direction consistency constraints, that is, "pruned" sections. They will increase the computational complexity of subsequent map matching and need to be eliminated.

[0064] Step S3, taking the path between two adjacent intersection points in the candidate road subnet as an increment, a trajectory point matching calculation based on the path increment is performed. Each incremental calculation process starts from the intersection point and determines the process of subsequent matching sections. When determining the incremental forward direction at the intersection point, the two similarity evaluation indicators of Hausdorff distance and curvature are comprehensively considered to reduce the influence of parallel sections in the same direction on the matching results.

[0065] Taking the path between two adjacent intersection points in the candidate subnet as the increment, the trajectory point matching calculation based on the path increment is performed, such as Figure 6 As shown, it specifically includes the following steps:

[0066] S31, determine the initial section: use the "forward-looking" approach to determine the initial section, that is, select a section with the highest similarity to the trajectory line from the candidate sections of the starting trajectory point as the starting section. Figure 7 The pseudo code of the algorithm for determining the initial road segment is shown.

[0067] S32, determine the next road section: after determining the initial road section, enter the incremental calculation process. Figure 8 As shown in the figure, each incremental calculation process in the algorithm starts from the intersection point to determine the subsequent matching section. When the vertex connecting the matching section is a transition point, the current section is recorded and the next section is determined; when the vertex connecting the matching section is an intersection point, the next incremental calculation process is entered; until a single connected point appears, the matching process ends. During the matching process, the similarity evaluation of the section and the trajectory line at each intersection point uses the similarity evaluation index combining the Hausdorff distance and curvature.

[0068] S33, output matching results: The matching at the intersection is used to determine the direction of the incremental matching. The algorithm pseudo code is as follows: Fig. 9 First, a candidate point set I is constructed based on the intersection points and the search radius, then the candidate points and their projection points on the corresponding road sections are obtained, the similarity between the trajectory and the road section is calculated, and finally the incremental matching direction is determined by comparing the similarity.

[0069] Fig.10 Describes the matching process of intersection point v1. Fig.10 As shown in (a), e1 is the road segment that has been matched in the previous increment, and v1 is the starting point of the next incremental matching. At this time, it is necessary to determine the matching direction of the subsequent increment, that is, to determine the path that matches the trajectory from the paths e1→e2 and e1→e3. Fig.10 (b) and Fig.10 (c) depicts the matching process of two different paths. By calculating the similarity between the trajectory and its projection points, it is easy to determine e3 as the subsequent matching section. Among them, the candidate point set I should contain both trajectory points that have passed and those that have not passed the intersection. Otherwise, the step length r is used to select the candidate point set I. c.step Gradually increase the search radius r c length until the condition is met.

[0070] The positional relationship between the trajectory points and the intersection points in the candidate point set I is determined by the projection of the intersection points on the line connecting the adjacent trajectory points. Fig.11 As shown in (a), p i , p i+1 are two consecutive trajectory points, v k is the intersection point, let v k ′ is v k In p i , p i+1 The projection point on . The determination method is as follows: From v k Towards line segment p i , p i+1 Draw a vertical line. If the foot of the vertical line falls on the line segment p i , p i+1 On, take the foot of the vertical as v k The projection point, otherwise take p i , p i+1 Middle distance v k The closer endpoint is the projection point. Then we can judge that when v k ′ is located at p i , p i+1 When p i Not passing the intersection point v k , p i+1 Passed v k ; When v k ′ and p i When overlapping, p i , p i+1 All have passed v k ,like Fig.11 (b) As shown; when v k ′ and p i+1 When overlapping, p i , p i+1 None passed vk ,like Fig.11 (c) as shown.

[0071] The similarity at the intersection is a combination of the Hausdorff distance and curvature, and its calculation formula is: Among them, H dif is the absolute value of the difference in Hausdorff distance between the trajectory and the corresponding path, K dif is the absolute value of the curvature integral difference between the trajectory and the corresponding road. The range of similarity is (0, 1]. The larger the value, the more similar the current path is to the trajectory.

[0072] Hausdorff distance is a measure of the similarity between two sets of points. Suppose there are two sets A = {a1, a2, ..., a p}, B = {b1, b2, ..., b q}, the Hausdorff distance between the two groups of points is: H(A, B) = max(h(A, B), h(B, A)) where h(A, B) = maxmin||ab||, (a∈A, b∈B), h(B, A) = maxmin||ba||, (a∈A, b∈B), ||*|| is the distance paradigm between point set A and point set B.

[0073] The curvature is used to describe the geometric features of the candidate points at the intersection and the road section to be matched. In the map matching problem, the trajectory and the road are discretized into a sequence of points. Therefore, the vector formed by the two adjacent points in the discretized point sequence is used to approximate the tangent vector of the position. Suppose the smooth curve γ is discretized into a point sequence [a1, a2, ..., a n ], then the curvature integral calculation formula of the discretized curve is: When calculating the curvature of each point, the north direction is used as the reference.

[0074] The present invention will be further described below with reference to the embodiments.

[0075] Example

[0076] To verify the effectiveness of the algorithm, some taxi trajectories were selected from the GeoLife 1.3 public dataset provided by Microsoft Research Asia for experiments. The trajectory sampling interval was 1s. The road network data is the road traffic network of Beijing. The selected trajectories pass through special sections such as a large number of parallel sections and complex intersections. The test trajectory is as follows: Fig.12 shown.

[0077] Before matching, the trajectory data is preprocessed to remove possible abnormal trajectory points based on the moving speed. Abnormal trajectory points can be divided into two categories: ① points where the trajectory moving speed is much lower than the normal driving speed or stagnant, and ② points that deviate from the normal trajectory range due to positioning errors. Both types of abnormal points are manifested as abnormal speeds and can be removed by speed thresholds. In the experiment, abnormal points with speeds less than 1m / s or greater than 35m / s were removed.

[0078] The experimental host environment is AMD Ryzen 536006-Core Processor CPU@3.60GHz, 16.0GB RAM, WIN10 x64; the programming environment used is Python 3.6.

[0079] In order to verify the effect of the algorithm, the algorithm proposed in the present invention is compared with the map matching algorithm constrained by curvature integral (hereinafter referred to as curvature integral algorithm, the relevant reference documents are: Zhe Zeng, Tong Zhang, Qingquan Li, Zhongheng Wu, Haixiang Zou & Chunxian Gao. Curvedness feature constrained map matching for low-frequency probe vehicle data [J], International Journal of Geographical Information Science, 2016, 30: 4, 660-690) and the HMM algorithm (the relevant reference documents are: Newson, P., Krumm, J.: Hidden markov map matching through noise and sparseness [J]. In: ACM SIGSPATIAL International Conference on Advances in Geographic Information Systems, 2009, 336–343). The quantitative evaluation is mainly carried out from the perspective of the accuracy and efficiency of the matching results. The evaluation indicators of the algorithm matching accuracy are: Rount Mismatch Fraction (RMF), Precision, and Recall. The calculation formula of RMF is: Where, L + L is the length of the error increase in the algorithm calculation result compared with the true path, - It is the length of the error reduction in the algorithm calculation result compared with the true path; the calculation formulas for accuracy and recall are: and Where L matched L is the total length of the correct matching paths in the algorithm calculation results. inferred is the total length of the algorithm calculation results, L real is the total length of the actual path.

[0080] Considering the GNSS positioning error and road width in the experimental dataset, the buffer radius of the candidate road section is r p and the candidate point buffer radius r c The initial length of is set to 30m, and the corresponding step length is set to 5m. At the same time, in order to compare the matching performance of related algorithms under different sampling frequencies, the sampling interval is gradually increased to 10 seconds during the experiment. The corresponding algorithm is implemented according to the description in the literature, and the above indicators are used for comparative experiments.

[0081] Matching result statistics are as follows: Fig.13 As shown in , in the test data set, when the sampling interval is 1s, the accuracy of the algorithm provided by the present invention reaches 97.70%, the recall rate reaches 97.10%, and the RMF is 5.23%, which is significantly improved compared with the other two algorithms, indicating that the algorithm of the present invention can ensure a high matching accuracy rate at a high sampling rate. The reasons are mainly two aspects: (1) The algorithm filter filters out most of the wrong sections, greatly reducing the possibility of wrong matching. Fig.14 , Fig.15 , Fig.16 , Fig.17 is the matching result of each data set when the sampling interval is 1s. It can be seen from the figure that the HMM algorithm has a large number of cases where the trajectory is matched to the opposite direction of the road section, while the algorithm of the present invention has filtered out the opposite direction section before matching; (2) The matching process based on the incremental algorithm provided by the present invention ensures the uniqueness of the path between the intersection points and reduces the adverse effects that parallel sections may have on the matching results. Fig.15 As shown in the figure, the curvature integral algorithm has two parallel sections selected. The reason is that this method selects matching sections based on point pairs. When the trajectory sampling rate is high, two parallel sections in the same direction may be selected in different point pairs, resulting in wrong matching. The algorithm of the present invention avoids the wrong selection of parallel sections. At the same time, the algorithm also shows good results in matching complex scenes with changing driving directions using multi-layer interconnections. The results are shown in Figure 2. Fig.17 However, as the sampling interval increases, the number of candidate points in the intersection point matching will gradually decrease, causing the algorithm accuracy and recall rate to gradually decrease. When the sampling interval is about 10s, there is no obvious advantage, as shown in Fig.13 (a) and Fig.13(b) shows that the algorithm of the present invention is more sensitive to the change of sampling rate. However, due to the uniqueness of the algorithm acquisition path, RMF is still significantly better than other algorithms.

[0082] The efficiency of the above algorithms is statistically analyzed, and the results are as follows: Fig.18 As shown in the figure. The shorter the sampling interval, the more obvious the efficiency advantage of this algorithm. The reason is that the number of calls to the matching unit in the two comparison algorithms is closely related to the number of trajectory points, and frequent calls to the matching unit will undoubtedly lead to higher time costs. However, this algorithm uses the path as an increment, so that the number of calls to the calculation unit is only related to the number of intersections in the road network. After the road network is combined and filtered, the complexity is greatly reduced, thereby achieving further improvement in efficiency.

[0083] Fig.19 is the number of calls to the matching unit in each algorithm when the sampling interval is 1s. When the trajectory sampling interval is short, the number of trajectory points is much larger than the number of intersection points that need to be matched, so the number of calls to the matching unit of this algorithm is much lower than that of the other two algorithms, which significantly improves the efficiency of the algorithm. As the sampling interval increases, the number of trajectory points gradually decreases, but the number of intersection points remains unchanged, so the efficiency advantage of this algorithm will gradually disappear, and eventually it will be close to the efficiency of the global matching algorithm.

[0084] Compared with the existing map matching algorithm with curvature integral constraint and the map matching algorithm based on hidden Markov model, the algorithm provided by the present invention shows higher accuracy and efficiency in solving the map matching problem of high sampling rate GNSS trajectory points, and can better handle various complex road sections. Therefore, the present invention is very suitable for high sampling rate GNSS trajectory matching application scenarios in complex urban road networks.

[0085] The above implementation modes are not limitations of the present invention, and the present invention is not limited to the above examples. Changes, modifications, additions or substitutions made by technicians in this technical field within the scope of the technical solution of the present invention also belong to the protection scope of the present invention.

Claims

1. A high sampling rate trajectory map matching method based on path increment, characterized in that: The method comprises the following steps: Step S1, formally defining a GNSS trajectory point map matching process based on path increment; The GNSS trajectory point map matching process based on path increment is defined as: based on the trajectory T and the road network G(V, E), the candidate road subnetwork G′(V′, E′) is screened out, and then the path between two adjacent intersection points in the candidate road subnetwork is used as the increment to perform trajectory matching calculation, thereby gradually determining the actual motion path of the object; Step S2, extracting candidate road subnetworks from the road network using a multi-constraint combination filtering method; Step S3: Taking the path between two adjacent intersection points in the candidate road subnet as an increment, a trajectory point matching calculation based on the path increment is performed. The specific process is as follows: First, determine the initial segment, that is, select the segment with the highest similarity to the trajectory line from the candidate segments of the starting trajectory point as the starting segment; After the initial road segment is determined, the incremental calculation process begins. Each incremental calculation process starts from the intersection point and gradually determines the subsequent matching road segments. When the vertex connected by the matching road segment is a transition point, the current road segment is recorded and the next road segment is determined; When the vertex connected by the matched road segment is an intersection point, the next incremental calculation process is entered; until a single connected point appears, it is used as the end point of the trajectory and the matching process ends.

2. The high sampling rate trajectory map matching method based on path increment according to claim 1 is characterized in that: The specific process of step S2 is: combining the spatial distribution and movement rules of two adjacent trajectory points with the characteristics of their nearby roads, and removing a large number of irrelevant road sections by adopting a series of filtering conditions such as distance constraints, road connectivity constraints, direction consistency constraints and "pruning" of single-point connected road sections to obtain a candidate road subnet G′(V′, E′).

3. The high sampling rate trajectory map matching method based on path increment according to claim 2 is characterized in that: In step S2, the distance constraint is implemented by dynamically adjusting the buffer zone, that is, taking the trajectory point as the center and r as the center. p0 Build a buffer for the initial radius. If the candidate road segment set is empty, expand the radius by a fixed step size r. p.step , and repeat this cycle until the candidate segment set contains at least one segment.

4. The high sampling rate trajectory map matching method based on path increment according to claim 3 is characterized in that: In step S2, the road connectivity constraints include topological connectivity constraints and directional connectivity constraints, wherein the topological connectivity constraints are: for any point p in the trajectory i , its candidate road segment set C(p i ) has segment e i and the candidate road segment set C(p i+1 ) in a certain section e j The endpoints are connected or overlap; Directional connectivity constraint: For any point p in the trajectory i , candidate road segment set C(p i ) has segment e i and the candidate road segment set C(p i+1 ) in a certain section e j connected or overlapping, and when e i With e j When connected, the directions of the two sections should be connected end to end at the connection point.

5. The high sampling rate trajectory map matching method based on path increment according to claim 3 is characterized in that: In step S2, the direction consistency constraint includes a length ratio constraint and an azimuth constraint, wherein the length ratio constraint is: if the length ratio between the distance between two consecutive trajectory points and the distance between their projection points on the candidate road section is less than a threshold value of 0.8, the candidate road section is eliminated; Azimuth constraint: Calculate the angle between the trajectory segment and its projection segment on the candidate path. If the angle is greater than 90°, the candidate segment is eliminated.

6. The high sampling rate trajectory map matching method based on path increment according to claim 5 is characterized in that: In the azimuth constraint, the path in the opposite direction is eliminated by calculating the difference between the vector azimuth of the trajectory and the projection point of the trajectory on the candidate path, where the vector azimuth is the horizontal angle passed by the current vector from the north direction to the clockwise rotation, and its calculation formula is: Among them, α is the azimuth of the current vector, (x1, y1), (x2, y2) are the coordinates of the first and last points of the vector respectively.

7. The high sampling rate trajectory map matching method based on path increment according to claim 3 is characterized in that: In step S2, "pruning" of single-point connected road segments refers to: traversing each road segment in turn, calculating the degree of the starting point and the end point of each road segment respectively, and deleting other single-point connected road segments except the first and last road segments.

8. The high sampling rate trajectory map matching method based on path increment according to claim 1 is characterized in that: In step S3, when determining the incremental forward direction at the intersection, a similarity evaluation index combining the Hausdorff distance and the curvature is used to reduce the influence of parallel sections in the same direction on the matching result.

Images