A method and apparatus for extracting traffic networks from high-noise trajectory data

Abstract

This invention provides a method and apparatus for extracting traffic networks from high-noise trajectory data. Based on the density raster map of the trajectory data, an iterative path extraction process is executed to extract multiple path centerlines to obtain the traffic network. In the path extraction process, a Markov probability model is used to model the tracking process, and the tracking step size is adaptively calculated based on the curvature of the trajectory lines to address the Markov chain interruption problem, thus providing stronger robustness when tracking path structures with different widths and curvatures. Then, for each local path center point candidate point set, the deviation of the candidate points from the path center is adaptively quantified based on the density raster map to solve the path center estimation problem under uneven trajectory distribution. Finally, the Viterbi algorithm is introduced to account for local measurement errors caused by noise, and the path centerlines are accurately derived with optimal global joint probability, overcoming the problem of complex parameter settings faced by existing technologies in different scenarios and improving the accuracy of extraction.

Description

Technical Field

[0001] This invention relates to the field of trajectory data mining and analysis technology, specifically to a method and apparatus for extracting traffic networks from high-noise trajectory data. Background Technology

[0002] With the widespread adoption of portable positioning devices and the continuous improvement of communication infrastructure, aviation, maritime, and urban land transportation have accumulated massive amounts of low-cost trajectory data rich in mobile information, such as flight paths, ship trajectories, and floating car trajectories. Extracting the corresponding transportation networks from this data can effectively support research on trajectory behavior discrimination, path planning, and intelligent monitoring and management.

[0003] In existing technologies, the focus is often on raw data trajectory points or trajectory lines. Clustering algorithms are used to obtain trajectory clusters to form a representation path. This may lead to a serious burden on computing resources during the similarity measurement stage. Furthermore, trajectory data is usually characterized by low frequency, high noise, and spatial heterogeneity. When extracting traffic networks in detail, it faces the challenge of setting complex parameters (such as achieving a balance between noise reduction and preservation of path details), resulting in poor extraction accuracy. Summary of the Invention

[0004] In view of this, embodiments of the present invention provide a method and apparatus for extracting traffic networks from high-noise trajectory data, so as to improve the accuracy of traffic network extraction.

[0005] To achieve the above objectives, the embodiments of the present invention provide the following technical solutions:

[0006] The first aspect of this invention discloses a method for extracting traffic networks from high-noise trajectory data, the method comprising:

[0007] The pre-collected trajectory data is processed into a density raster to obtain a density raster map;

[0008] Based on the density raster map, an iterative path extraction process is executed to extract multiple path centerlines, resulting in a traffic network composed of these multiple path centerlines; the path extraction process includes:

[0009] Arbitrarily select a seed point from the density grid and obtain the tracking direction based on the directional trend of the trajectory lines around the seed point;

[0010] The width of the local path where the seed point is located is estimated based on the image gray-level gradient change. The local region where the seed point is located is determined based on the width of the local path where the seed point is located. The tracking step size is determined based on the curvature of the trajectory line in the local region where the seed point is located and the width of the local path where the seed point is located.

[0011] Based on the tracking direction and the tracking step size, tracking is performed starting from the seed point until the trajectory intersection point or sharp turning point is reached, and then the tracking stops, resulting in multiple tracking points. For each tracking point, multiple target trajectory points are searched along the direction perpendicular to the trajectory at the tracking point. The tracking points and the corresponding target trajectory points are used as candidate points for the local path center point to construct a candidate point set.

[0012] For each candidate point set, the deviation degree corresponding to each candidate point is quantified based on the trajectory density at each candidate point in the candidate point set.

[0013] The path centerline is calculated based on the preset Viterbi algorithm, each set of candidate points, and the degree of deviation corresponding to each candidate point.

[0014] Optionally, the step of performing density rasterization processing on the pre-collected trajectory data to obtain a density raster map includes:

[0015] Map the latitude and longitude coordinates of each trajectory point in the pre-collected trajectory data to the raster map;

[0016] For each grid cell of the grid image, the number of trajectory points is counted. If the number of trajectory points is greater than a threshold, the threshold is used as the grid value of the grid cell. If the number of trajectory points is less than or equal to the threshold, the number of trajectory points is used as the grid value of the grid cell.

[0017] The grid values ​​of the trajectory points in each grid cell are normalized so that the gray value range of each grid cell is within a preset target range, thus obtaining a density grid map.

[0018] Optionally, determining the tracking step size based on the curvature of the trajectory line within the local region where the seed point is located and the width of the local path where the seed point is located includes:

[0019] For each trajectory point within the local region where the seed point is located, calculate the direction angle difference between the previous trajectory point and the next trajectory point of the trajectory point.

[0020] The average angle difference corresponding to each trajectory point is averaged to obtain the average angle difference that characterizes the curvature of the trajectory line in the local area where the seed point is located.

[0021] The product of the average angle difference and the width of the local path where the seed point is located is determined as the tracking step size.

[0022] Optionally, for each candidate point set, quantifying the deviation degree corresponding to each candidate point based on the trajectory density at each candidate point in the candidate point set includes:

[0023] For each candidate point set, a corresponding slice image is obtained from the density raster image, and the slice image is flipped with each candidate point in the candidate point set as the center to obtain multiple flipped images.

[0024] Using a preset objective function, the similarity measure between each flipped image and the sliced ​​image is calculated;

[0025] The similarity measure between each flipped image and the sliced ​​image is used as the degree of deviation corresponding to each candidate point.

[0026] Optionally, the calculation of the path centerline based on a preset Viterbi algorithm, each candidate point set, and the deviation degree corresponding to each candidate point includes:

[0027] For each candidate point set, based on each candidate point in the candidate point set, the deviation degree corresponding to each candidate point, and a preset emission probability calculation formula, the emission probability corresponding to each candidate point is calculated; the emission probability calculation formula is: ;in, Let represent the deviation degree corresponding to the candidate points in the i-th candidate point set. These are candidate points in the i-th candidate point set;

[0028] The transition probability is calculated based on a preset transition probability calculation formula; the transition probability calculation formula is: ;in, For the (i+1)th candidate point set, for and The transition probability between them Indicates connection and vector lines, express and The direction of the local path between them ), for and The angle difference between them;

[0029] Based on a preset path scoring formula, the path score corresponding to each candidate tracking path is calculated; the path scoring formula is: ; The transition probability between the candidate point in the first candidate point set and the seed point;

[0030] The candidate tracing path with the highest path score is used as the path centerline.

[0031] A second aspect of this invention discloses a traffic network extraction device suitable for high-noise trajectory data, the device comprising:

[0032] The density rasterization processing unit is used to perform density rasterization processing on the pre-collected trajectory data to obtain a density raster map;

[0033] An iterative unit is used to iteratively execute a path extraction process based on the density raster map, extract multiple path centerlines, and obtain a traffic network composed of the multiple path centerlines; the path extraction process includes:

[0034] Arbitrarily select a seed point from the density grid and obtain the tracking direction based on the directional trend of the trajectory lines around the seed point;

[0035] The width of the local path where the seed point is located is estimated based on the image gray-level gradient change. The local region where the seed point is located is determined based on the width of the local path where the seed point is located. The tracking step size is determined based on the curvature of the trajectory line in the local region where the seed point is located and the width of the local path where the seed point is located.

[0036] Based on the tracking direction and the tracking step size, tracking is performed starting from the seed point until the trajectory intersection point or sharp turning point is reached, and then the tracking stops, resulting in multiple tracking points. For each tracking point, multiple target trajectory points are searched along the direction perpendicular to the trajectory at the tracking point. The tracking points and the corresponding target trajectory points are used as candidate points for the local path center point to construct a candidate point set.

[0037] For each candidate point set, the deviation degree corresponding to each candidate point is quantified based on the trajectory density at each candidate point in the candidate point set.

[0038] The path centerline is calculated based on the preset Viterbi algorithm, each set of candidate points, and the degree of deviation corresponding to each candidate point.

[0039] Optionally, the density rasterization processing unit is specifically used for:

[0040] Map the latitude and longitude coordinates of each trajectory point in the pre-collected trajectory data to the raster map;

[0041] For each grid cell of the grid image, the number of trajectory points is counted. If the number of trajectory points is greater than a threshold, the threshold is used as the grid value of the grid cell. If the number of trajectory points is less than or equal to the threshold, the number of trajectory points is used as the grid value of the grid cell.

[0042] The grid values ​​of the trajectory points in each grid cell are normalized so that the gray value range of each grid cell is within a preset target range, thus obtaining a density grid map.

[0043] Optionally, determining the tracking step size based on the curvature of the trajectory line within the local region where the seed point is located and the width of the local path where the seed point is located includes:

[0044] For each trajectory point within the local region where the seed point is located, calculate the direction angle difference between the previous trajectory point and the next trajectory point of the trajectory point.

[0045] The average angle difference corresponding to each trajectory point is averaged to obtain the average angle difference that characterizes the curvature of the trajectory line in the local area where the seed point is located.

[0046] The product of the average angle difference and the width of the local path where the seed point is located is determined as the tracking step size.

[0047] Optionally, for each candidate point set, quantifying the deviation degree corresponding to each candidate point based on the trajectory density at each candidate point in the candidate point set includes:

[0048] For each candidate point set, a corresponding slice image is obtained from the density raster image, and the slice image is flipped with each candidate point in the candidate point set as the center to obtain multiple flipped images.

[0049] Using a preset objective function, the similarity measure between each flipped image and the sliced ​​image is calculated;

[0050] The similarity measure between each flipped image and the sliced ​​image is used as the degree of deviation corresponding to each candidate point.

[0051] Optionally, the calculation of the path centerline based on a preset Viterbi algorithm, each candidate point set, and the deviation degree corresponding to each candidate point includes:

[0052] For each candidate point set, based on each candidate point in the candidate point set, the deviation degree corresponding to each candidate point, and a preset emission probability calculation formula, the emission probability corresponding to each candidate point is calculated; the emission probability calculation formula is: ;in, Let represent the deviation degree corresponding to the candidate points in the i-th candidate point set. These are candidate points in the i-th candidate point set;

[0053] The transition probability is calculated based on a preset transition probability calculation formula; the transition probability calculation formula is: ;in, For the (i+1)th candidate point set, for and The transition probability between them Indicates connection and vector lines, express and The direction of the local path between them ), for and The angle difference between them;

[0054] Based on a preset path scoring formula, the path score corresponding to each candidate tracking path is calculated; the path scoring formula is: ; The transition probability between the candidate point in the first candidate point set and the seed point;

[0055] The candidate tracing path with the highest path score is used as the path centerline.

[0056] Based on the above embodiments of the present invention, a method and apparatus for extracting traffic networks from high-noise trajectory data are provided. The trajectory data is processed by density rasterization to obtain a density raster map. Based on the density raster map, a path extraction process is iteratively executed to extract multiple path centerlines to obtain the traffic network. In the path extraction process, a Markov probability model is used to model the tracking process. The tracking step size is adaptively calculated based on the curvature of the trajectory line to address the problem of Markov chain interruption, thus providing stronger robustness when tracking path structures with different widths and curvatures. Then, for each local path center point candidate point set, the deviation of the candidate points from the path center is adaptively quantified based on the density raster map to solve the problem of path center estimation under uneven trajectory distribution. Finally, the Viterbi algorithm is introduced to account for local measurement errors caused by noise, and the path centerline is accurately derived with optimal global joint probability. This overcomes the problem of complex parameter settings faced by existing technologies in different scenarios and improves the accuracy of extraction. Attached Figure Description

[0057] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0058] Figure 1 This is a flowchart of a traffic network extraction method for high-noise trajectory data disclosed in an embodiment of the present invention;

[0059] Figure 2 This is a flowchart of a path extraction process disclosed in an embodiment of the present invention;

[0060] Figure 3 This is a schematic diagram of the trajectory point density distribution on a path profile disclosed in an embodiment of the present invention;

[0061] Figure 4 This is a comparison image before and after image flipping centered on candidate points, as disclosed in an embodiment of the present invention;

[0062] Figure 5 This is a schematic diagram illustrating the connection possibilities between two consecutive candidate points as disclosed in an embodiment of the present invention;

[0063] Figure 6 This is a structural diagram of a traffic network extraction device suitable for high-noise trajectory data, as disclosed in an embodiment of the present invention. Detailed Implementation

[0064] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0065] In this application, the terms "comprising," "including," or any other variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0066] As can be seen from the background technology, existing technologies mostly focus on raw data trajectory points or trajectory lines, using clustering algorithms to obtain trajectory clusters to form a representation path. This may bring a serious burden on computing resources in the similarity measurement stage. Moreover, trajectory data usually has the characteristics of low frequency, high noise, and spatial heterogeneity, which poses a complex parameter setting problem when extracting traffic networks in detail (such as achieving a balance between denoising and preserving path details), resulting in poor extraction accuracy.

[0067] Therefore, this invention discloses a method and apparatus for extracting traffic networks from high-noise trajectory data. In this scheme, a Markov probability model is used to model the tracking process, and the tracking step size is adaptively calculated based on the curvature of the trajectory line to address the problem of Markov chain interruption, thereby exhibiting stronger robustness when tracking path structures with different widths and curvatures. Then, for each local path center point candidate point set, the deviation of the candidate points from the path center is adaptively quantized based on the density raster map to solve the problem of path center estimation under uneven trajectory distribution. Finally, the Viterbi algorithm is introduced to take into account local measurement errors caused by noise, and the path center line is accurately derived with optimal global joint probability, thereby overcoming the problem of complex parameter settings faced by existing technologies in different scenarios and improving the accuracy of traffic network extraction.

[0068] like Figure 1 The diagram shows a flowchart of a traffic network extraction method for high-noise trajectory data disclosed in an embodiment of the present invention. The method includes the following steps:

[0069] Step S1: Perform density rasterization processing on the pre-collected trajectory data to obtain a density raster map.

[0070] In step S1, the density grid map can clearly reflect the backbone structure of the traffic network.

[0071] It should be noted that the trajectory data is pre-processed and includes, but is not limited to, trajectory data generated by vehicles, ships, and aircraft. The specific methods of collecting data for vehicles, ships, and aircraft are different. For example, floating cars are equipped with GPS receivers and collect data at an uncertain frequency when the vehicle is in motion.

[0072] The specific implementation of step S1 includes the following steps:

[0073] Step S1.1: Map the latitude and longitude coordinates of each trajectory point in the pre-collected trajectory data to the raster map.

[0074] In step S1.1, the latitude and longitude coordinates of all trajectory points in the collected trajectory data are... The trajectory location information is mapped onto a raster image.

[0075] Step S1.2: For each grid cell of the raster image, count the number of trajectory points. If the number of trajectory points is greater than the number threshold, use the number threshold as the raster value of the grid cell. If the number of trajectory points is less than or equal to the number threshold, use the number of trajectory points as the raster value of the grid cell.

[0076] In step S1.2, the number of trajectory points in each grid cell is initially counted as the grid value, and each grid value is truncated. Specifically, for grid cells with more than a set threshold number of trajectory points, their grid value is truncated to that threshold value, while for grid cells with more than a set threshold number of trajectory points, the value remains unchanged. The specific formula is as follows:

[0077] (1)

[0078] In equation (1), Indicates the number of trajectory points in a grid cell. This is a preset quantity threshold.

[0079] Step S1.3: Normalize the grid values ​​of each grid cell so that the gray value range of each grid cell is within the preset target range, thus obtaining a density grid map.

[0080] In step S1.3, the raster values ​​of each raster unit are normalized so that the grayscale value range of all raster units in the raster image is (0, 255). The density raster image uses a function... Represents the grayscale value at any position, where and These are the row and column numbers in the density raster image, respectively. The mapping relationship between the latitude and longitude coordinates of the trajectory points and the pixels in the density raster image is as follows:

[0081] (2)

[0082] (3)

[0083] in, and These represent the minimum longitude and maximum latitude of the study area (i.e., the area corresponding to the trajectory data), respectively. and These represent the number of pixels corresponding to each longitude and the number of pixels corresponding to each latitude, respectively.

[0084] In this embodiment of the invention, density rasterization can effectively reduce the impact of noise and drift of individual trajectory points, enhancing the ability of low-quality, high-noise trajectory data to describe the geometry of the route. Simultaneously, to avoid the problem of most low-density values ​​being compressed into a smaller range due to excessive differences between the maximum and minimum values ​​after image normalization, thus leading to the loss of details in local areas, especially in areas with relatively sparse trajectory distribution (such as road edges or local locations in the middle of a road segment), the preliminary statistical raster values ​​are truncated before normalization.

[0085] Step S2: Based on the density raster map, iteratively execute the path extraction process to extract multiple path centerlines and obtain a traffic network composed of multiple path centerlines.

[0086] In step S2, the path extraction process is iterative, which means that the path extraction process will be executed repeatedly until the preset conditions are met, such as extracting enough paths (a preset number of path centerlines) or the traffic network composed of multiple path centerlines can achieve a certain accuracy.

[0087] like Figure 2 The diagram shown is a path extraction flowchart disclosed in an embodiment of the present invention, including the following steps:

[0088] Step S2.1: Select a seed point arbitrarily from the density raster map, and obtain the tracking direction based on the directional trend of the trajectory lines around the seed point.

[0089] In step S2.1, the area surrounding the seed point refers to a small neighborhood of the seed point, the size of which can be preset. Within this neighborhood, the trend of change in the trajectory line direction is calculated, i.e., the direction of movement of the trajectory line, and the direction of movement of the trajectory line is used as the tracking direction.

[0090] Optionally, multiple seed points can be arbitrarily selected from the density raster map to execute the path extraction process in parallel, thereby improving the speed of traffic network extraction.

[0091] Step S2.2: Estimate the width of the local path where the seed point is located based on the image gray-level gradient change, determine the local region where the seed point is located based on the width of the local path where the seed point is located, and determine the tracking step size based on the curvature of the trajectory line in the local region where the seed point is located and the width of the local path where the seed point is located.

[0092] Optionally, for each trajectory point within the local region where the seed point is located, calculate the directional angle difference between the previous trajectory point and the next trajectory point; average the directional angle differences corresponding to each trajectory point to obtain the average angle difference that characterizes the curvature of the trajectory line within the local region where the seed point is located; and determine the tracking step size by multiplying the average angle difference by the width of the local path where the seed point is located.

[0093] In the specific implementation, based on the initially selected seed point, the width L of the local path where the seed point is located is estimated based on the image gray-level gradient change. Based on L, the local region where the seed point is located is determined (e.g., a region with a length of 1.5L and a width of 2L, centered on the seed point). The set of trajectory points in the local region is obtained, and the direction angle difference between each trajectory point in the set and its preceding and following points is calculated. After filtering out large outliers, the average angle difference is obtained. Set a scaling factor , and Inversely proportional; determine the tracking step size as .

[0094] Step S2.3: Based on the tracking direction and tracking step size, track from the seed point until the trajectory intersection point or sharp turning point is reached, and stop tracking to obtain multiple tracking points. For each tracking point, search along the direction perpendicular to the trajectory at the tracking point to obtain multiple target trajectory points. Use the tracking points and the corresponding target trajectory points as candidate points for the local path center point to construct a candidate point set.

[0095] Specifically, when the trajectory direction angle distribution at the tracking location shows a multi-peak distribution or the difference in turning angle between the front and rear of the trajectory is greater than 45°, the tracking is stopped when the trajectory intersection point or sharp turning point is determined.

[0096] Step S2.4: For each candidate point set, quantify the degree of deviation corresponding to each candidate point based on the trajectory density at each candidate point in the candidate point set.

[0097] In the specific implementation of step S2.4, for each candidate point set, the corresponding slice image is obtained from the density raster image, and the slice image is flipped with each candidate point in the candidate point set as the center to obtain multiple flipped images; using a preset objective function, the similarity measure between each flipped image and the slice image is calculated; the similarity measure between each flipped image and the slice image is used as the deviation degree corresponding to each candidate point.

[0098] like Figure 3 The figure shown is a schematic diagram of the trajectory point density distribution on a path profile disclosed in an embodiment of the present invention.

[0099] It is understandable that the density of trajectory points exhibits a symmetrical trend on both sides of the path centerline. This symmetry is unaffected by the path width and the density or sparseness of the points.

[0100] like Figure 4 The image shown is a comparison before and after image flipping centered on candidate points, as disclosed in an embodiment of the present invention.

[0101] It can be observed that the closer the flip center is to the actual path center, the more similar the flipped image after vertical flipping is to the original slice image.

[0102] Therefore, in this embodiment of the invention, a convex model is constructed, assuming the flip center (candidate point) is... The objective function is a measure of the symmetry similarity between the sliced ​​image and the flipped image obtained after flipping the sliced ​​image. The objective function reaches its maximum value when the flipped center is close to the actual center of the path, thereby quantifying the degree of deviation.

[0103] It should be noted that similarity measurement is a measure that comprehensively evaluates the degree of similarity between two things. The closer two things are, the greater their similarity measurement; conversely, the more distant two things are, the smaller their similarity measurement.

[0104] The objective function is expressed as:

[0105] (4)

[0106] In equation (4), It is the original slice image. At the center of the flip The flipped image obtained by flipping.

[0107] In the specific implementation, the normalized cross-correlation (NCC) coefficient is used to calculate... and The similarity measure between them is used as the flip center (a candidate point in the i-th candidate point set). Corresponding degree of deviation The degree of deviation corresponding to the candidate points in the i-th candidate point set. The calculation formula is as follows:

[0108] (5)

[0109] In equation (5), and These represent the pixel coordinates of the original sliced ​​image and the flipped image, respectively. ,when season . The larger the value, the closer the candidate point is to the actual path centerline; conversely, the smaller the value, the further away it is from the actual path centerline.

[0110] Step S2.5: Calculate the path centerline based on the preset Viterbi algorithm, each candidate point set, and the degree of deviation corresponding to each candidate point.

[0111] The specific implementation process of step S2.5 is as follows:

[0112] For each candidate point set, the emission probability corresponding to each candidate point is calculated based on each candidate point in the candidate point set, the deviation degree corresponding to each candidate point, and the preset emission probability calculation formula.

[0113] The formula for calculating the launch probability is:

[0114] (6)

[0115] in, Let represent the deviation degree corresponding to the candidate points in the i-th candidate point set. Let be a candidate point in the i-th candidate point set. The emission probability of the candidate point. Degree of deviation There is a positive correlation. The emission probability corresponding to the candidate point with the largest deviation is... It equals 1.

[0116] like Figure 5 The diagram shown is a schematic representation of the connection possibilities between two consecutive candidate points as disclosed in an embodiment of the present invention.

[0117] Figure 5 Two lines were shown from arrive The connecting lines, where the fixed tracking step size is set to 1.5 based on experience. (That is, the scaling factor is 1.5). Observation shows that, compared with the blue dashed line, the direction of the orange line segment is closer to that of the blue dashed line. and The local path direction between two candidate points has a geometry more suitable as the centerline of the local path. Therefore, this embodiment of the invention introduces the road segment direction information between two candidate points to calculate the transition probability, as follows:

[0118] The transition probability is calculated based on a preset transition probability calculation formula; the transition probability calculation formula is:

[0119] (7)

[0120] in, For the (i+1)th candidate point set, for and The transition probability between them Indicates connection and vector lines, express and The direction of the local path between them ), for and The angle difference between them; The angle depends on the area (e.g.) Figure 5 The angles of the trajectory points within the red box area are calculated.

[0121] Then, based on the preset path scoring formula, the path score corresponding to each candidate tracking path is calculated; the path scoring formula is:

[0122] (8)

[0123] in, This represents the transition probability between a candidate point and a seed point in the first candidate point set.

[0124] Finally, the candidate tracing path with the highest path score is used as the path centerline.

[0125] Understandably, according to the Viterbi algorithm, the optimal path is the candidate tracking path with the highest score. Therefore, the optimal tracking path... The start and end points (seed points) of the newly added path ; trajectory intersections or sharp turning points After connection, it is considered as the centerline of the desired path, denoted as . . It is expressed as follows:

[0126]

[0127] Based on the above-described embodiments of the present invention, a traffic network extraction method suitable for high-noise trajectory data is proposed. In this scheme, a Markov probability model is used to model the tracking process, and the tracking step size is adaptively calculated based on the curvature of the trajectory line to address the problem of Markov chain interruption, thereby exhibiting stronger robustness when tracking path structures with different widths and curvatures. Then, for each local path center point candidate point set, the deviation of the candidate points from the path center is adaptively quantized based on the density raster map to solve the path center estimation problem under uneven trajectory distribution. Finally, the Viterbi algorithm is introduced to take into account the local measurement errors caused by noise, and the path center line is accurately derived with optimal global joint probability, thereby overcoming the problem of complex parameter settings faced by existing technologies in different scenarios and improving the accuracy of traffic network extraction.

[0128] like Figure 6 The diagram shown is a structural diagram of a traffic network extraction device for high-noise trajectory data disclosed in an embodiment of the present invention. The device includes a density rasterization processing unit 601 and an iteration unit 602.

[0129] The density rasterization processing unit 601 is used to perform density rasterization processing on the pre-collected trajectory data to obtain a density raster map.

[0130] Optionally, the density rasterization processing unit 601 is specifically used for:

[0131] Map the latitude and longitude coordinates of each trajectory point in the pre-collected trajectory data to the raster map;

[0132] For each grid cell in the raster image, the number of trajectory points is counted. If the number of trajectory points is greater than the number threshold, the number threshold is used as the grid value of the grid cell. If the number of trajectory points is less than or equal to the number threshold, the number of trajectory points is used as the grid value of the grid cell.

[0133] The grid values ​​of the trajectory points in each grid cell are normalized so that the gray value range of each grid cell is within the preset target range, thus obtaining a density grid image.

[0134] Iteration unit 602 is used to iteratively execute the path extraction process based on the density raster map, extract multiple path centerlines, and obtain a traffic network composed of multiple path centerlines; the path extraction process includes:

[0135] Select any seed point from the density raster and obtain the tracking direction based on the directional trend of the trajectory lines around the seed point;

[0136] The width of the local path where the seed point is located is estimated based on the image gray-level gradient change. The local region where the seed point is located is determined based on the width of the local path where the seed point is located. The tracking step size is determined based on the curvature of the trajectory line in the local region where the seed point is located and the width of the local path where the seed point is located.

[0137] Based on the tracking direction and tracking step size, tracking starts from the seed point and stops when the trajectory intersection point or sharp turning point is reached, resulting in multiple tracking points. For each tracking point, multiple target trajectory points are searched along the direction perpendicular to the trajectory at the tracking point. The tracking points and their corresponding target trajectory points are used as candidate points for the local path center point to construct a candidate point set.

[0138] For each candidate point set, the deviation degree corresponding to each candidate point is quantified based on the trajectory density at each candidate point in the candidate point set.

[0139] The path centerline is calculated based on the pre-defined Viterbi algorithm, each candidate point set, and the degree of deviation corresponding to each candidate point.

[0140] Optionally, the tracking step size is determined based on the curvature of the trajectory line within the local region where the seed point is located and the width of the local path where the seed point is located, including:

[0141] For each trajectory point within the local region where the seed point is located, calculate the difference in direction angle between the previous trajectory point and the next trajectory point of the trajectory point;

[0142] The average angle difference between each trajectory point is used to obtain the average angle difference that represents the curvature of the trajectory line in the local area where the seed point is located.

[0143] The tracking step size is determined by multiplying the average angle difference by the width of the local path where the seed point is located.

[0144] Optionally, for each candidate point set, the deviation degree corresponding to each candidate point is quantified based on the trajectory density at each candidate point in the candidate point set, including:

[0145] For each candidate point set, the corresponding slice image is obtained from the density raster image, and the slice image is flipped with each candidate point in the candidate point set as the center to obtain multiple flipped images.

[0146] Using a predefined objective function, the similarity measure between each flipped image and the sliced ​​image is calculated;

[0147] The similarity measure between each flipped image and the sliced ​​image is used as the degree of deviation corresponding to each candidate point.

[0148] Optionally, based on a preset Viterbi algorithm, each candidate point set, and the deviation degree corresponding to each candidate point, the path centerline is calculated, including:

[0149] For each candidate point set, based on each candidate point in the set, the deviation degree corresponding to each candidate point, and a preset emission probability calculation formula, the emission probability corresponding to each candidate point is calculated; the emission probability calculation formula is: ;in, Let represent the deviation degree corresponding to the candidate points in the i-th candidate point set. These are candidate points in the i-th candidate point set;

[0150] The transition probability is calculated based on a preset transition probability calculation formula; the transition probability calculation formula is: ;in, For the (i+1)th candidate point set, for and The transition probability between them Indicates connection and vector lines, express and The direction of the local path between them ), for and The angle difference between them;

[0151] Based on a preset path scoring formula, the path score for each candidate tracking path is calculated; the path scoring formula is: ; The transition probability between the candidate point and the seed point in the first candidate point set;

[0152] The candidate tracing path with the highest path score is used as the path centerline.

[0153] Based on the above-described embodiments of the present invention, a traffic network extraction device suitable for high-noise trajectory data is disclosed. In this scheme, a Markov probability model is used to model the tracking process, and the tracking step size is adaptively calculated based on the curvature of the trajectory line to address the problem of Markov chain interruption, thereby exhibiting stronger robustness when tracking path structures with different widths and curvatures. Then, for each local path center point candidate point set, the deviation of the candidate points from the path center is adaptively quantized based on the density raster map to solve the path center estimation problem under uneven trajectory distribution. Finally, the Viterbi algorithm is introduced to take into account the local measurement errors caused by noise, and the path center line is accurately derived with optimal global joint probability, thereby overcoming the problem of complex parameter settings faced by existing technologies in different scenarios and improving the accuracy of traffic network extraction.

[0154] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, for system or system embodiments, since they are basically similar to method embodiments, the description is relatively simple, and relevant parts can be referred to the descriptions in the method embodiments. The systems and system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0155] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0156] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A traffic network extraction method suitable for high noise trajectory data, characterized in that, The method comprises: Density gridding processing is performed on the pre-acquired trajectory data to obtain a density grid map; Based on the density grid map, an iterative path extraction process is performed to extract a plurality of path centerlines and obtain a traffic network composed of the plurality of path centerlines; the path extraction process comprises: An arbitrary seed point is selected from the density grid map, and a tracking direction is obtained according to the direction trend of trajectory lines around the seed point; The width of a local path in which the seed point is located is estimated based on the image gray gradient change, the local area in which the seed point is located is determined based on the width of the local path in which the seed point is located, and a tracking step is determined based on the bending degree of trajectory lines in the local area in which the seed point is located and the width of the local path in which the seed point is located; Based on the tracking direction and the tracking step, tracking is performed from the seed point as a starting point until a trajectory intersection point or a sharp turning corner point is tracked, a plurality of tracking points are obtained, and for each tracking point, a plurality of target trajectory points are searched along a direction perpendicular to the trajectory at the tracking point, the tracking point and each target trajectory point corresponding to the tracking point are taken as candidate points of a local path center point to construct a candidate point set; For each candidate point set, the deviation degree corresponding to each candidate point in the candidate point set is quantified according to the trajectory density of each candidate point in the candidate point set; Based on a preset Viterbi algorithm, each candidate point set and the deviation degree corresponding to each candidate point, a path centerline is calculated and obtained. The method comprises: The latitude and longitude coordinates of each trajectory point in the pre-acquired trajectory data are mapped into a grid map; For each grid cell of the grid map, the number of trajectory points is counted, if the number of trajectory points is greater than a number threshold, the number threshold is taken as the grid value of the grid cell, if the number of trajectory points is less than or equal to the number threshold, the number of trajectory points is taken as the grid value of the grid cell; The grid values of trajectory points in each grid cell are normalized to make the gray value range of each grid cell in a preset target range, and a density grid map is obtained.

2. The method of claim 1, wherein, The method comprises: For each trajectory point in the local area in which the seed point is located, the direction angle difference value between the last trajectory point of the trajectory point and the next trajectory point of the trajectory point is calculated; The direction angle difference values corresponding to each trajectory point are averaged to obtain an average angle difference value representing the bending degree of trajectory lines in the local area in which the seed point is located; A corresponding proportion coefficient is obtained according to the average angle difference value, wherein the proportion coefficient is inversely proportional to the average angle difference value; The product of the proportion coefficient and the width of the local path in which the seed point is located is determined as the tracking step.

3. The method of claim 1, wherein, The method comprises: For each of the candidate point sets, a corresponding slice image is obtained from the density grid map, and the slice image is flipped with each candidate point in the candidate point set as the center to obtain a plurality of flipped images; A similarity measure between each of the flipped images and the slice image is calculated using a preset objective function; The similarity measure between each of the flipped images and the slice image is taken as the deviation degree corresponding to each of the candidate points.

4. The method according to any one of claims 1 to 3, characterized in that, The path center line is calculated based on the preset Viterbi algorithm, each of the candidate point sets, and the deviation degree corresponding to each of the candidate points, and includes: For each of the candidate point sets, a corresponding emission probability of each candidate point is calculated based on each candidate point in the candidate point set, a corresponding deviation degree of each candidate point, and a preset emission probability calculation formula. The emission probability calculation formula is: ; wherein, is a candidate point in the ith candidate point set, is a candidate point in the ith candidate point set. The transition probability is calculated based on a preset transition probability calculation formula, and the transition probability calculation formula is: ; wherein, is a candidate point in an (i+1)th candidate point set, is a transition probability between and , , denotes a vector line connecting and , , denotes a direction of a local path between and , , , is a direction of a vector line connecting and , , is an angle difference between the direction of the vector line and a direction of the local path between and , , . Based on a preset path score formula, a path score corresponding to each candidate tracking path is calculated; the path score formula is: ; is a transition probability between a candidate point in the first candidate point set and the seed point. The candidate tracking path with the highest path score is taken as the path center line.

5. A traffic network extraction apparatus suitable for high noise trajectory data, characterized by, The device includes: A density gridding processing unit configured to perform density gridding processing on pre-acquired trajectory data to obtain a density grid map; An iteration unit configured to iteratively perform a path extraction process based on the density grid map to extract a plurality of path center lines and obtain a traffic network composed of the plurality of path center lines; the path extraction process includes: selecting a seed point from the density grid map at random, and obtaining a tracking direction according to a directional trend of trajectory lines around the seed point; estimating a width of a local path where the seed point is located based on a change in image gray scale gradient, determining a local region where the seed point is located based on the width of the local path where the seed point is located, and determining a tracking step based on a bending degree of trajectory lines in the local region where the seed point is located and the width of the local path where the seed point is located; tracking from the seed point as a starting point based on the tracking direction and the tracking step until a trajectory intersection point or a sharp turning corner point is tracked to, to obtain a plurality of tracking points, and for each of the tracking points, searching for a plurality of target trajectory points along a direction perpendicular to a trajectory at the tracking point, taking the tracking points and the corresponding target trajectory points as candidate points of local path center points to construct a candidate point set; quantifying a deviation degree corresponding to each of the candidate points according to trajectory density at each of the candidate points in each of the candidate point sets; calculating a path center line based on the preset Viterbi algorithm, each of the candidate point sets, and the deviation degree corresponding to each of the candidate points; The density gridding processing unit is specifically configured to: map longitude and latitude coordinates of each trajectory point in the pre-acquired trajectory data to a grid map; for each grid cell of the grid map, count the number of trajectory points, if the number of trajectory points is greater than a number threshold, take the number threshold as a grid value of the grid cell, if the number of trajectory points is less than or equal to the number threshold, take the number of trajectory points as the grid value of the grid cell; normalize the grid values of the trajectory points in each of the grid cells so that the gray scale value range of each of the grid cells is in a preset target range to obtain a density grid map.

6. The apparatus of claim 5, wherein, The method for determining the tracking step based on the bending degree of trajectory lines in the local region where the seed point is located and the width of the local path where the seed point is located includes: calculating, for each trajectory point in the local region where the seed point is located, a difference value of a direction angle between a previous trajectory point of the trajectory point and a next trajectory point of the trajectory point; averaging the direction angle difference values corresponding to the trajectory points to obtain an average angle difference value representing a bending degree of a trajectory line in the local region where the seed point is located; obtaining a proportional coefficient corresponding to the average angle difference value, wherein the proportional coefficient is inversely proportional to the average angle difference value; determining a product of the proportional coefficient and a width of the local path where the seed point is located as a tracking step length.

7. The apparatus of claim 5, wherein, The method further includes: for each candidate point set, quantifying a deviation degree corresponding to each candidate point in the candidate point set according to a trajectory density at each candidate point in the candidate point set, including: for each candidate point set, obtaining a corresponding slice image from the density grid map, and respectively flipping the slice image with each candidate point in the candidate point set as a center to obtain a plurality of flipped images; calculating a similarity measure between each flipped image and the slice image by using a preset target function; taking the similarity measure between each flipped image and the slice image as the deviation degree corresponding to each candidate point.

8. The apparatus of any one of claims 5 to 7, wherein, The method further includes: calculating a path center line based on a preset Viterbi algorithm, each candidate point set, and the deviation degree corresponding to each candidate point, including: For each of the candidate point sets, a corresponding emission probability of each candidate point is calculated based on each candidate point in the candidate point set, a corresponding deviation degree of each candidate point, and a preset emission probability calculation formula; the emission probability calculation formula is: ; wherein, is a candidate point in the i th candidate point set, is a candidate point in the i th candidate point set. The transition probability is calculated based on a preset transition probability calculation formula, and the transition probability calculation formula is: ; wherein, is a candidate point in an (i+1)th candidate point set, is an angle difference between a direction of a vector line connecting a direction of a local path between and​​​​​​​​​​​​​​ Based on a preset path score formula, a path score corresponding to each candidate tracking path is calculated; the path score formula is: ; is a transition probability between a candidate point in the first candidate point set and the seed point. taking the candidate tracking path with the highest path score as the path center line.

Images