Track path identification method and device, equipment, medium and program product

Abstract

The invention provides a trajectory path identification method, device and equipment, a medium and a program product, and the method comprises the steps: positioning an original trajectory point in vehicle trajectory data into a target lane according to a matching result between the vehicle trajectory data (such as the trajectory data of an intersection scene) and road data (such as the road data of the intersection scene) and a preset lane model; and for the track points in the target lane, training a preset Gaussian mixture model so as to obtain Gaussian distribution of the track points in the target lane according to the trained Gaussian mixture model, and when a distribution result has multi-peak distribution, determining that a plurality of driving paths exist in the target lane. The accuracy of track path identification, especially the accuracy of track path identification in a complex intersection scene, can be effectively improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent driving, and in particular to a trajectory path identification method and device, equipment, medium and program product. BACKGROUND

[0002] In the field of intelligent driving and intelligent transportation, the identification of trajectory paths is a key to realizing safe driving and efficient traffic management. In particular, in complex road scenarios such as intersections, crossroads, and roundabouts, the identification of trajectory paths is particularly important.

[0003] Due to the dramatic fluctuations in vehicle speed and frequent changes in driving direction in these complex road scenarios, it is difficult to accurately capture and predict the actual driving path of the vehicle. In particular, at intersections without clear physical lane divisions, the driving trajectory of the vehicle is often more complex and variable, further increasing the difficulty of trajectory path identification, resulting in a large gap between the identified trajectory path and the actual driving path of the vehicle.

[0004] Therefore, how to improve the accuracy of trajectory path identification, especially in complex intersection scenarios, has become a problem to be solved. SUMMARY

[0005] The present application provides a trajectory path identification method, device, equipment, medium and program product to improve the accuracy of trajectory path identification, especially in complex intersection scenarios.

[0006] According to an aspect of the present application, a driving path identification method is provided, comprising: positioning original trajectory points in the vehicle trajectory data to a target lane according to the matching results between the vehicle trajectory data and the road data and the pre-set lane model, the target lane being a lane matched with the original trajectory point position information; training a pre-set Gaussian mixture model for the trajectory points in the target lane to obtain the Gaussian distribution of the trajectory points according to the trained Gaussian mixture model to obtain a distribution result; wherein the pre-set Gaussian mixture model is a Gaussian mixture model containing a pre-set number of Gaussian components; when the distribution result has a multi-peak distribution, it is determined that there are multiple driving paths in the target lane.

[0007] In an embodiment, the matching result comprises a correspondence between the original trajectory point in the vehicle trajectory data and a corresponding road; the lane model comprises at least one lane included in each road; and the positioning of the original trajectory point in the vehicle trajectory data into the target lane according to the matching result between the vehicle trajectory data and the road data and the preset lane model comprises: for each original trajectory point, obtaining a boundary distance of the original trajectory point to each lane in the road to which the original trajectory point corresponds according to the road to which the original trajectory point corresponds and position information of the original trajectory point; and determining the target lane in which the original trajectory point is located according to the boundary distance of the original trajectory point to each lane in the road to which the original trajectory point corresponds.

[0008] In an embodiment, the training of the preset Gaussian mixture model comprises: initializing parameters of the Gaussian mixture model, the parameters comprising a mixing weight, a mean value and a variance; and training the parameters of the Gaussian mixture model according to an expectation maximization algorithm, so as to obtain the trained Gaussian mixture model according to the trained parameters.

[0009] In an embodiment, the training of the parameters of the Gaussian mixture model according to the expectation maximization algorithm comprises: in an expectation step, calculating a first posterior probability of each trajectory point belonging to each Gaussian component according to a lateral coordinate of each trajectory point; in a maximization step, updating the parameters of the Gaussian mixture model based on the first posterior probability; and repeating the expectation step and the maximization step until the parameters of the Gaussian mixture model change less than a preset convergence threshold or a maximum iteration number is reached, so as to obtain the trained parameters.

[0010] In an embodiment, the method further comprises: dividing the target lane into a plurality of lane segments according to a preset segmentation rule, and preprocessing the original trajectory points in each lane segment to obtain preprocessed trajectory points; and the distribution result comprises a distribution result of each lane segment; and the training of the preset Gaussian mixture model for the trajectory points in the target lane comprises: training the preset Gaussian mixture model for all trajectory points in each lane segment, so as to obtain a Gaussian distribution of the trajectory points in each lane segment according to the trained Gaussian mixture model, and determining that there are a plurality of driving paths in the corresponding lane segment when the distribution result of the lane segment has a multimodal distribution.

[0011] In an embodiment, the obtaining, according to the trained Gaussian mixture model, the Gaussian distribution of the trajectory points in each lane segment comprises: for each lane segment, calculating, according to the lateral coordinates of each trajectory point, a second posterior probability of each trajectory point in the trained Gaussian mixture model; assigning each trajectory point to a corresponding Gaussian component according to the second posterior probability to obtain an assignment result of the Gaussian components; and determining the Gaussian distribution of the trajectory points in each lane segment according to the assignment result of the Gaussian components.

[0012] In an embodiment, the dividing the target lane into a plurality of lane segments according to the preset segmentation rule comprises: dividing the target lane into a plurality of lane segments according to a reference line curvature of a road reference line of the target lane, wherein the lane segments are lane segments with equal lengths or lane segments with unequal lengths; or dividing the target lane into a plurality of lane segments with equal lengths according to a predetermined fixed segmentation step, wherein the lane segments are lane segments with equal lengths.

[0013] In an embodiment, the preprocessing the original trajectory points in each lane segment to obtain preprocessed trajectory points comprises: for each lane segment, if there is no original trajectory point in the lane segment, determining a position coordinate of a to-be-inserted trajectory point in the lane segment according to a preset linear interpolation algorithm; and inserting a new trajectory point in the lane segment according to the position coordinate of the to-be-inserted trajectory point to obtain the preprocessed trajectory points.

[0014] In an embodiment, the method further comprises: screening, from the plurality of lane segments, lane segments with multimodal distributions as multimodal lane segments; for each multimodal lane segment, determining a forward continuity rate and / or a backward continuity rate between each Gaussian distribution in the multimodal lane segment and a corresponding Gaussian distribution in an adjacent multimodal lane segment; screening, from the multimodal lane segments, reliable multimodal lane segments according to the forward continuity rate and / or the backward continuity rate; and determining the final plurality of driving paths according to the reliable multimodal lane segments.

[0015] In an embodiment, the determining the forward continuity rate and / or the backward continuity rate between each Gaussian distribution in the multimodal lane segment and a corresponding Gaussian distribution in an adjacent multimodal lane segment comprises: determining a backward continuity of the multimodal lane segment according to an overlapping number between all trajectory points of each Gaussian distribution in the multimodal lane segment and all trajectory points of a corresponding Gaussian distribution in a previous multimodal lane segment; and / or determining a forward continuity of the multimodal lane segment according to an overlapping number between all trajectory points of each Gaussian distribution in the multimodal lane segment and all trajectory points of a corresponding Gaussian distribution in a next multimodal lane segment.

[0016] In an embodiment, the determining the final plurality of driving paths according to the reliable multi-modal lane segments comprises: obtaining, for each reliable multi-modal segment, an average trajectory position of each Gaussian distribution in the reliable multi-modal segment; clustering all the average trajectory positions in each reliable multi-modal lane segment to obtain a trajectory grouping for each reliable multi-modal lane segment; and determining the final plurality of driving paths according to the trajectory grouping.

[0017] In an embodiment, the obtaining the average trajectory position of each Gaussian distribution in the reliable multi-modal segment comprises: obtaining a second posterior probability of each trajectory point in the reliable multi-modal segment in each Gaussian distribution; and calculating the average trajectory position of each Gaussian distribution in the reliable multi-modal segment according to the second posterior probability of each trajectory point in each Gaussian distribution.

[0018] According to a second aspect of the present disclosure, there is provided a trajectory path recognition device, comprising: a lane positioning module configured to position original trajectory points in vehicle trajectory data into a target lane according to a matching result between the vehicle trajectory data and road data and a preset lane model, the target lane being a lane matched with position information of the original trajectory points; a model processing module configured to train a preset Gaussian mixture model for the trajectory points in the target lane to obtain Gaussian distributions of the trajectory points according to the trained Gaussian mixture model to obtain a distribution result, wherein the preset Gaussian mixture model is a Gaussian mixture model containing a preset number of Gaussian components; and a multi-modal recognition module configured to determine that there are a plurality of driving paths in the target lane when the distribution result has a multi-modal distribution.

[0019] In an embodiment, the matching result comprises original trajectory points in the vehicle trajectory data being associated with corresponding roads; the lane model comprises at least one lane contained in each road; and the lane positioning module comprises: a distance obtaining unit configured to obtain, for each original trajectory point, a boundary distance of the original trajectory point to each lane in an associated road according to the associated road and position information of the original trajectory point; and a lane determining unit configured to determine a target lane in which the original trajectory point is located according to the boundary distance of the original trajectory point to each lane in the associated road.

[0020] In an embodiment, the model processing module comprises: an initialization unit configured to initialize parameters of the Gaussian mixture model, the parameters comprising a mixing weight, a mean and a variance; and a training unit configured to train the parameters of the Gaussian mixture model according to an expectation maximization algorithm to obtain the trained Gaussian mixture model according to the trained parameters.

[0021] In an implementation, the training unit is specifically configured to: in an expectation step, calculate a first posterior probability of each trajectory point belonging to each Gaussian component according to a lateral coordinate of each trajectory point; in a maximization step, update parameters of the Gaussian mixture model based on the first posterior probability; and repeatedly perform the expectation step and the maximization step until the parameters of the Gaussian mixture model change less than a preset convergence threshold or a maximum iteration number is reached, to obtain the trained parameters.

[0022] In an implementation, the method further includes: a lane segmentation module configured to divide the target lane into a plurality of lane segments according to a preset segmentation rule, and to pre-process the original trajectory points in each lane segment to obtain pre-processed trajectory points; the distribution result includes a distribution result of each lane segment; and the model processing module is specifically configured to: for all trajectory points in each lane segment, train a preset Gaussian mixture model to obtain a Gaussian distribution of the trajectory points in each lane segment according to the trained Gaussian mixture model, and determine that there are a plurality of driving paths in the corresponding lane segment when the distribution result of the lane segment has a multi-peak distribution.

[0023] In an implementation, the model processing module includes: a first calculation unit configured to calculate, for each lane segment, a second posterior probability of each trajectory point in the trained Gaussian mixture model according to a lateral coordinate of each trajectory point; an assignment module configured to assign each trajectory point to a corresponding Gaussian component according to the second posterior probability to obtain an assignment result of the Gaussian components; and a distribution determination module configured to determine a Gaussian distribution of the trajectory points in each lane segment according to the assignment result of the Gaussian components.

[0024] In an implementation, the lane segmentation module includes: a first division unit configured to divide the target lane into a plurality of lane segments according to a reference line curvature of a road reference line of the target lane to determine a variable segmentation step, the lane segments being lane segments with equal lengths or unequal lengths; or a second division unit configured to divide the target lane into a plurality of lane segments with equal lengths according to a predetermined fixed segmentation step, the lane segments being lane segments with equal lengths.

[0025] In an implementation, the lane segmentation module includes: a position determination unit configured to, for each lane segment, determine a position coordinate of a to-be-inserted trajectory point in the lane segment according to a preset linear interpolation algorithm if there is no original trajectory point in the lane segment; and an interpolation unit configured to insert a new trajectory point in the lane segment according to the position coordinate of the to-be-inserted trajectory point to obtain the pre-processed trajectory points.

[0026] In an embodiment, the device further comprises: a segment screening module configured to screen out, from the plurality of lane segments, lane segments with multimodal distribution as multimodal lane segments; a continuity rate determining module configured to determine, for each multimodal lane segment, a forward continuity rate and / or a backward continuity rate between each Gaussian distribution in the multimodal lane segment and a corresponding Gaussian distribution in an adjacent multimodal lane segment; the segment screening module is further configured to screen out reliable multimodal lane segments from the multimodal lane segments according to the forward continuity rate and / or the backward continuity rate; and a path determining module configured to determine a plurality of final driving paths according to the reliable multimodal lane segments.

[0027] In an embodiment, the continuity rate determining module comprises: a backward continuity rate determining unit configured to determine a backward continuity of a multimodal lane segment according to an overlap quantity between all trajectory points of each Gaussian distribution in the multimodal lane segment and all trajectory points of a corresponding Gaussian distribution in a previous multimodal lane segment; and / or a forward continuity rate determining unit configured to determine a forward continuity of the multimodal lane segment according to an overlap quantity between all trajectory points of each Gaussian distribution in the multimodal lane segment and all trajectory points of a corresponding Gaussian distribution in a next multimodal lane segment.

[0028] In an embodiment, the path determining module comprises: a trajectory position obtaining unit configured to obtain, for each reliable multimodal segment, an average trajectory position of each Gaussian distribution in the reliable multimodal segment; a clustering grouping unit configured to cluster all average trajectory positions in each reliable multimodal lane segment to obtain a trajectory grouping for each reliable multimodal lane segment; and a path generating unit configured to determine a plurality of final driving paths according to the trajectory groupings.

[0029] In an embodiment, the trajectory position obtaining unit is specifically configured to: obtain second posterior probabilities of all trajectory points in the reliable multimodal segment in each Gaussian distribution; and calculate the average trajectory position of each Gaussian distribution in the reliable multimodal segment according to the second posterior probabilities of all trajectory points in each Gaussian distribution.

[0030] According to a third aspect of the present application, an electronic device is provided, comprising: a memory and a processor; the memory stores computer-executable instructions; and the processor executes the computer-executable instructions stored in the memory, so that the electronic device performs the trajectory path identification method provided in any one of the first aspect.

[0031] According to a fourth aspect of the present application, a computer readable storage medium is provided, which stores computer execution instructions, and the computer execution instructions are used to implement the trajectory path identification method according to any one of the first aspect when executed by a processor.

[0032] According to a fifth aspect of the present application, a computer program product is provided, which comprises a computer program, and the computer program is used to implement the trajectory path identification method according to any one of the first aspect when executed by a processor.

[0033] The trajectory path identification method, device, equipment, medium and program product provided by the present application can locate the original trajectory points in the vehicle trajectory data (such as the trajectory data of the intersection scene) to the target lane according to the matching result between the vehicle trajectory data (such as the trajectory data of the intersection scene) and the road data (such as the road data of the intersection scene) and the preset lane model. For the trajectory points in the target lane, the preset Gaussian mixture model is trained to obtain the Gaussian distribution of the trajectory points in the target lane according to the trained Gaussian mixture model, and the distribution result is obtained. When the distribution result has multiple peak distributions, it is determined that there are multiple driving paths in the target lane. In this process, the trajectory in the intersection is located by lane-level matching, the Gaussian mixture model is trained using the trajectory data in the target lane, and it is identified whether there are multiple trajectory paths in the target lane. The identification of the trajectory path based on the lane level is realized, so as to improve the accuracy of the intersection trajectory path identification and adapt to various complex road scenes. BRIEF DESCRIPTION OF DRAWINGS

[0034] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0035] Figure 1 A possible application scenario provided by the embodiment of the present application is shown in the following figure;

[0036] Figure 2a One of the flowcharts of the trajectory path identification method provided by the embodiment of the present application is shown in the following figure;

[0037] Figure 2b The flowchart of another trajectory path identification method provided by the embodiment of the present application is shown in the following figure;

[0038] Figure 3 An example diagram for preprocessing the trajectory points in the lane segment in the embodiment of the present application is shown in the following figure;

[0039] Figure 4a The flowchart of another trajectory path identification method provided by the embodiment of the present application is shown in the following figure;

[0040] Figure 4bA flowchart of a trajectory path recognition method provided for an example embodiment of the present application is shown in the following figure;

[0041] Figure 5 A schematic diagram of a road intersection scene provided for an embodiment of the present application is shown in the following figure;

[0042] Figure 6 A schematic diagram of trajectory data of an intersection scene provided for an embodiment of the present application is shown in the following figure;

[0043] Figure 7a An example diagram of trajectory grouping provided for an embodiment of the present application is shown in the following figure;

[0044] Figure 7b An example diagram of multiple trajectory paths identified provided for an embodiment of the present application is shown in the following figure;

[0045] Figure 8 A structural diagram of a trajectory path recognition device provided for an embodiment of the present application is shown in the following figure;

[0046] Figure 9 A structural diagram of an electronic device provided for an embodiment of the present application is shown in the following figure.

[0047] The specific embodiments of the present application have been shown in the above figures, and will be described in more detail hereinafter. These figures and the written description are not intended to limit the scope of the present application in any way, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0048] The example embodiments will be described in detail herein with reference to the accompanying drawings. When the following description refers to the accompanying drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following example embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of apparatuses and methods consistent with some aspects of the present application as detailed in the appended claims.

[0049] The embodiments of the present application will be explained in the following with reference to application scenarios. The trajectory path recognition scheme provided by the embodiments of the present application can be applied to the application scenario of intelligent driving, more specifically, to the application scenario of automatic driving based on vehicle cloud computing. For example, the execution subject of the method provided by the embodiments of the present application can be a server, for example, a server of a navigation map manufacturer. The following will introduce the server as the execution subject of the method provided by the embodiments of the present application.

[0050] Figure 1 A schematic diagram of a trajectory path recognition scenario provided for an embodiment of the present application is shown in the following figure, for example, Figure 1As shown, the server 110 and the intelligent vehicle 120 are connected through a network, the server 110 obtains trajectory data (which can be for intersection scenarios), analyzes whether there are multiple trajectory paths in the intersection scenario according to the trajectory data, and can transmit the recognition result of the trajectory path to the intelligent vehicle 120. The intelligent vehicle 120 can better understand the current traffic environment according to the recognition result of the trajectory path, identify potential traffic conflicts and obstacles, and thus enhance its environmental perception capability. Alternatively, the server 110 can dynamically adjust its path planning to select the optimal path and improve the safety of vehicle driving. The server 110 can be a physical server, a server cluster or a distributed system composed of multiple physical servers, or a cloud server providing cloud services, cloud databases, cloud computing, and other basic cloud computing services. In some application scenarios, the server 110 can also provide detailed intersection path data for high-precision map construction, enrich map scene details, and help autonomous vehicles to plan intersection passing routes in advance and accurately, thereby improving driving safety.

[0051] In related technologies, when trajectory data is used for path analysis, only a single mainstream path is usually extracted by a clustering algorithm, and multiple trajectory paths derived from driving habits, road obstacles, and other factors are ignored. Due to the complex road scenes such as intersections, which are usually accompanied by variable traffic flow, complex vehicle interaction, and uncertain environmental factors, especially intersections without clear physical lane division, it is difficult to handle the complex scenarios of large trajectory speed fluctuations, variable directions, and the diversity of intersection traffic behaviors by extracting a single mainstream path. For example, one way is to use a graph-based clustering algorithm to construct a similarity matrix between data points, map it to a low-dimensional space, and then use a clustering algorithm to cluster the data points. This process does not consider the possibility of multiple driving paths, and is dependent on the construction of the similarity matrix. Noise or outliers (such as noise introduced due to sensor errors or environmental interference) can significantly affect similarity calculation and thus affect clustering results.

[0052] Therefore, an embodiment of the present application proposes a trajectory path recognition method, device, equipment, medium and program product. The original trajectory points in the vehicle trajectory data are positioned into the target lane according to the matching result between the vehicle trajectory data (such as the trajectory data of an intersection scene) and the road data (such as the road data of an intersection scene) and the preset lane model. For the trajectory points in the target lane, the preset Gaussian mixture model is trained to obtain the Gaussian distribution of the trajectory points in the target lane according to the trained Gaussian mixture model, and a distribution result is obtained. When the distribution result has multiple peak distributions, it is determined that there are multiple driving paths in the target lane. In this process, the trajectory in the intersection is positioned and matched at the lane level, the Gaussian mixture model is trained using the trajectory data in the target lane, and it is recognized whether there are multiple trajectory paths in the target lane, the recognition of the trajectory path is realized based on the lane level, thereby improving the accuracy of the intersection trajectory path recognition and adapting to diversified complex road scenes.

[0053] The technical solutions of the present application and how the technical solutions of the present application solve the above technical problems will be described in detail below in combination with the drawings and specific embodiments. It should be noted that these specific embodiments can be combined with each other, and the same or similar concepts or processes can not be described in detail in some embodiments.

[0054] It should be noted that the user information (including but not limited to user equipment information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or authorized by all parties, and the collection, use and processing of related data need to comply with relevant laws, regulations and standards, and provide corresponding operation portal for user to choose authorization or refusal.

[0055] Figure 2a One of the flowcharts of a driving path recognition method provided by an embodiment of the present application is shown in FIG. 1, which can include steps S201-S204. Figure 2a

[0056] Step S201: According to the matching result between the vehicle trajectory data and the road data and the preset lane model, the original trajectory points in the vehicle trajectory data are positioned into the target lane. The target lane is a lane matched with the position information of the original trajectory points.

[0057] For example, the vehicle trajectory data can be trajectory data for a specific road scene, for example, trajectory data for an intersection road scene. These trajectory data can include the timestamp and coordinate point of the original trajectory point, for example, the time, lateral coordinate and longitudinal coordinate of the i-th original trajectory point can be represented as .

[0058] ​Exemplarily, the road data can be acquired through a high-definition map, and can include road topology information and the like. By matching the trajectory data with the road data, the trajectories can be associated to corresponding roads.

[0059] Exemplarily, the matching result between the trajectory data and the road data can be obtained through the following matching manner: for example, a range can be demarcated with the trajectory point in the trajectory data as the center, and candidate roads can be filtered in combination with the heading angle of the trajectory point. By calculating the distance and / or heading angle deviation between the trajectory point and the candidate roads, a matching degree is obtained to determine the candidate road matched by the trajectory point, so as to obtain the road associated with the trajectory point.

[0060] Exemplarily, a preset lane model can be used to describe detailed information of lanes on each road, such as lane boundary, lane shape and the like. A person skilled in the art can select a suitable lane model in combination with actual application or experience value.

[0061] In this embodiment, the matching result can include that the original trajectory point in the vehicle trajectory data is associated to a corresponding road, and the lane model includes at least one lane contained in each road. In the above step, the original trajectory point in the vehicle trajectory data is positioned into a target lane according to the matching result between the vehicle trajectory data and the road data and the preset lane model, which can be implemented in the following manner:

[0062] For each original trajectory point, the boundary distance of the original trajectory point to each lane in the associated road is obtained according to the associated road of the original trajectory point and the position information of the original trajectory point.

[0063] The target lane of the original trajectory point is determined according to the boundary distance of the original trajectory point to each lane in the associated road.

[0064] Exemplarily, for each original trajectory point, the transverse boundary distance of the original trajectory point to the lane can be obtained according to the vertical distance between the original trajectory point and the left and right boundaries of the lane, and the longitudinal boundary distance of the original trajectory point can be obtained according to the driving distance of the original trajectory point in the lane. In this embodiment, the target lane of the original trajectory point can be determined according to the transverse boundary distance. For example, when the boundary distance of the original trajectory point to a certain lane is the smallest and the original trajectory point is located between the boundaries of the lane (within the lane range), it is considered that the original trajectory point is located in the lane.

[0065] Through the above manner, the original trajectory point of the vehicle can be efficiently positioned into a specific target lane, so that the trajectory point data is matched to an appropriate lane in a complex road environment, to facilitate subsequent driving path analysis.

[0066] In some embodiments, in addition to the lane-level positioning of the trajectory points that can be achieved in the above manner, it can also be achieved in other ways. For example, by utilizing the detailed lane information in the high-precision map, the vehicle trajectory points are aligned with the lanes on the map through a map matching algorithm to determine the lane in which the trajectory points are located. Specifically, first, the original trajectory data of the vehicle is obtained, and the lane layer data (which can include the accurate coordinates of the lane lines, the lane width, the lane ID, the lane curvature, the lane connection relationship, and the road speed limit, etc. static features) of the target area in the high-precision map is loaded. Then, according to the latitude and longitude range of the trajectory points, a candidate lane area is delimited in the high-precision map (such as a range with the trajectory point as the center and a preset distance as the radius), and all possible matching lanes in the area are screened out (excluding lanes with a heading angle deviation from the trajectory point exceeding a threshold), forming a candidate lane set; then the matching degree of the trajectory points and each lane in the candidate lane set is calculated, for example, by calculating the vertical distance of the trajectory point to the lane center line, the angle between the trajectory point heading angle and the lane direction, the adaptation degree of the vehicle speed and the lane speed limit, etc. Multi-dimensional indicators, and a weighted sum algorithm can be used to quantify the matching degree, and the lane with the highest matching degree is selected as the matching result of the trajectory point. Further, the time correlation of the continuous trajectory points and the lane connection relationship can also be combined for dynamic verification. If the matching lanes of adjacent trajectory points have reasonable connection logic (such as conforming to the lane changing rules of normal vehicle driving, no abnormal jump across multiple non-connected lanes), the final matching result is confirmed, and the lane ID corresponding to each trajectory point and the lateral position in the lane (such as the distance from the left lane line) are output, thereby completing the lane-level positioning.

[0067] In step S202, the preset Gaussian mixture model is trained for the trajectory points in the target lane to obtain the Gaussian distribution of the trajectory points according to the trained Gaussian mixture model, and a distribution result is obtained. The preset Gaussian mixture model is a Gaussian mixture model containing a preset number of Gaussian components.

[0068] The Gaussian mixture model (Gaussian Mixture Model, GMM) is a probability model used to represent a mixture of multiple Gaussian distributions, which can be used for clustering and modeling of data, and is particularly suitable for processing data with multi-modal distribution. In this embodiment, for a large number of trajectory points, the Gaussian distribution of the trajectory points can be efficiently identified through the modeling of the Gaussian mixture model, so as to facilitate the identification of the trajectory path.

[0069] Exemplarily, the preset Gaussian mixture model is a Gaussian mixture model with a predetermined number of Gaussian components, i.e., a preset number of Gaussian components, so as to facilitate subsequent model training and processing. A person skilled in the art can determine the preset number in combination with an actual road scene. For example, as mentioned in the following embodiments, for an intersection scene, there are usually 2-5 possibilities of driving paths, and the value of the preset number can be determined within this range.

[0070] Exemplarily, the embodiment can be to train a Gaussian mixture model on the trajectory points in the target lane and obtain the Gaussian distribution of each trajectory point. Alternatively, the target lane can be segmented into lane segments, a Gaussian mixture model can be trained in each lane segment, and the Gaussian distribution of each trajectory point in each lane segment can be obtained. The embodiment does not particularly limit this.

[0071] Step S03: When the distribution result exists a multi-peak distribution, it is determined that there are multiple driving paths in the target lane.

[0072] In the embodiment, whether the trajectory data exists a multi-peak distribution, i.e., whether there are multiple Gaussian distributions corresponding to multiple different trajectory point distribution groups, can be determined by the trained Gaussian mixture model, which can efficiently identify whether there are multiple driving paths, and is particularly suitable for complex road segment scenes such as intersections.

[0073] Alternatively, to further improve the recognition accuracy of driving paths in complex road environments such as intersections, the lane is segmented, and the trajectory points in each segment are analyzed and recognized to provide a more fine-grained recognition method. Specifically, Figure 2b is a flowchart of another embodiment of a driving path recognition method provided by the application. Based on the above embodiment, the embodiment adds step S202', and further divides the original step S202 into step S2021 and divides the original step S203 into step S2031.

[0074] As shown in Figure 2b Step S202', the target lane is divided into multiple lane segments according to a preset segmentation rule, and the original trajectory points in each lane segment are preprocessed to obtain preprocessed trajectory points.

[0075] In the embodiment, the target lane is divided into multiple lane segments according to the preset segmentation rule, which can be divided by a fixed segmentation step or a variable segmentation step.

[0076] Method one: the target lane is divided into multiple lane segments with equal lengths according to a predetermined fixed segmentation step, and the lane segments are lane segments with equal lengths.

[0077] It should be noted that the fixed segmentation step can be determined by the person skilled in the art according to actual application or experience value, for example, the fixed segmentation step can be determined according to attributes such as road type or traffic flow. For example, for a highway, the segmentation step can be longer because the vehicle speed is faster and the lane change is less. For a complex intersection, the segmentation step can be shorter to adapt to frequent lane changes and complex traffic environments.

[0078] By segmenting the target lane by a fixed segmentation step, the efficiency of lane segmentation can be improved for lanes with small changes (such as reference line curvature changes).

[0079] Method two: determining a variable segmentation step according to the reference line curvature of the road reference line of the target lane, and dividing the target lane into a plurality of lane segments, the lane segments being lane segments with equal lengths or lane segments with unequal lengths.

[0080] The reference line curvature is an index describing the degree of lane curvature, which can be calculated by the geometric shape of the road (such as the radius of the curve).

[0081] In this embodiment, the length of the lane segmentation can be dynamically adjusted according to the calculated reference line curvature information. A shorter segmentation step is used in areas with large curvature (such as sharp bends), and a longer segmentation step is used in areas with small curvature (such as straight roads or gentle bends). According to the adjusted step, the segments can be equal in length (in areas with little change in curvature) or unequal in length (in areas with significant change in curvature). Compared with the fixed lane segmentation method, the lane segmentation length is dynamically adjusted using the lane line curvature, making the lane segmentation more flexible to better adapt to the geometric characteristics and actual needs of the road.

[0082] In some embodiments, in addition to the above-mentioned lane segmentation methods, other methods can also be used for lane segmentation. For example, lanes can be segmented according to their functions (such as main roads, auxiliary roads, ramps), and different functions of lanes may require different segmentation strategies to meet their specific traffic management needs. Alternatively, historical traffic data can be used to analyze and determine the segmentation strategy. For example, by analyzing historical traffic patterns and events, lane segmentation can be optimized to facilitate subsequent trajectory path recognition.

[0083] In this embodiment, the original trajectory points in each lane segment are preprocessed to obtain preprocessed trajectory points, which can be obtained in the following way: for each lane segment, if there are no original trajectory points in the lane segment, the position coordinates of the trajectory points to be inserted in the lane segment are determined according to a pre-set linear interpolation algorithm. According to the position coordinates of the trajectory points to be inserted, new trajectory points are inserted in the lane segment to obtain preprocessed trajectory points.

[0084] In this embodiment, the distribution of the trajectory points is calculated based on the lane-level distribution and through the division manner of the lane segments. However, it is considered that in the actual collection process, not every segment contains trajectory points. If there is no trajectory point in the lane segment, it will lead to the difficulty in identifying the trajectory path in the lane segment or the discontinuity of the trajectory path generated subsequently. Based on this, for the lane segment without trajectory points (hereinafter referred to as the distribution segment), the linear interpolation method can be used to process the lane segment without trajectory points. Specifically, the slope and offset can be used to calculate the position coordinates of the trajectory points to be inserted, and then new trajectory points are inserted in the lane segment.

[0085] Exemplarily, the linear interpolation algorithm is used to determine the position coordinates of the trajectory points to be inserted in the lane segment. The longitudinal coordinate y of the trajectory point to be inserted can be obtained by the following formula:

[0086]

[0087] In the formula, may be the longitudinal coordinate of the original trajectory point in the last lane segment of the lane segment, may be the transverse coordinate of the original trajectory point in the last lane segment of the lane segment, may be the longitudinal coordinate of the original trajectory point in the next lane segment of the lane segment, may be the transverse coordinate of the original trajectory point in the next lane segment of the lane segment. Correspondingly, the transverse coordinate of the trajectory to be inserted can take the average value between and or a fixed value.

[0088] Exemplarily, the lane segment can contain trajectory points corresponding to multiple trajectories. The above interpolation method can be used for interpolation of each trajectory in the multiple trajectories. In each lane segment, one trajectory point can be selected for each trajectory (for example, the trajectory point closest to the center point of the lane segment is selected, and other trajectory points are deleted, that is, in addition to the pre-processing method of inserting trajectory points, deletion of trajectory points can also be performed). When inserting the trajectory points, for each trajectory (whether it is the same trajectory can be identified according to the vehicle speed, vehicle identification, etc.), the trajectory points in the last lane segment and the trajectory points in the next lane segment are combined to perform interpolation calculation, and the new trajectory points in the lane segment can be obtained. Not only can the trajectory point interpolation be more accurate, but the calculation amount can also be effectively reduced. Exemplarily, the insertion and deletion of the trajectory points are as shown in Figure 3 .

[0089] Through the above method, the trajectory point data can be completed in the case of sparse or missing data points in the lane segment when the lane segment is divided, and the continuity and integrity of the trajectory are improved.

[0090] Step S2021, for each trajectory point in each lane segment, a preset Gaussian mixture model is trained to obtain a Gaussian distribution of the trajectory points in each lane segment according to the trained Gaussian mixture model, and a distribution result is obtained. The Gaussian mixture model is a Gaussian mixture model containing a preset number of Gaussian components.

[0091] In an implementation, the training of the preset Gaussian mixture model can be performed by initializing parameters of the Gaussian mixture model, including the mixing weight, the mean and the variance, and setting the number of Gaussian components. According to the expectation maximization algorithm, the parameters of the Gaussian mixture model are trained to obtain the trained Gaussian mixture model according to the trained parameters.

[0092] In this embodiment, through the lane-level distribution, the lateral offset (i.e., the distance of the vehicle in the direction perpendicular to the lane center line or reference line, which describes the lateral position of the trajectory point in the lane, i.e., the lateral coordinate of the trajectory point) and the longitudinal offset (i.e., the distance of the vehicle in the direction along the lane center line or reference line, which describes the longitudinal position of the vehicle in the lane, i.e., the longitudinal coordinate of the trajectory point) of each trajectory point in the corresponding distribution can be calculated, wherein the lateral distribution of the trajectory points in a single sample (i.e., a single lane segment) can be expressed as .

[0093] Taking the trajectory lateral distribution as a sample set, K Gaussian components are used (a person skilled in the art can adaptively set the value of K according to actual application or experience value. For example, for the intersection road scene, there may be 2-5 driving paths, and the value of K can be determined in this range). The GMM modeling training process can be: input: the trajectory point lateral distribution in the trajectory lateral distribution set , wherein , xi represents the vehicle coordinates (i.e., the trajectory point), R 2 is a two-dimensional vector.

[0094] Gaussian mixture model modeling: assuming that the trajectory data is generated by K Gaussian distributions, the Gaussian mixture model, i.e., the probability density function, can be expressed as:

[0095]

[0096] In the formula, , represents the probability density of the trajectory point x (i.e., xi, the i-th trajectory point), , represents the parameter set of the model, including the mean, variance and mixing weight of all Gaussian components. , is the mixing weight of the k-th Gaussian component, , is the mean of the k-th Gaussian component, , represents the variance of the k-th Gaussian component. is the probability density function of the k-th Gaussian component, i.e., a Gaussian distribution. Wherein,

[0097]

[0098] In the training process, the model parameters can be initialized first For example:

[0099] .

[0100] Through the input trajectory point sample , the preset number of Gaussian components K, and the above initial model parameters, and the convergence threshold value ε (which can be determined according to the actual application or experience value and ε). Using the expectation maximization algorithm for training, outputting the estimated model parameters , obtaining the trained Gaussian mixture model.

[0101] Specifically, the above training of the Gaussian mixture model parameters according to the expectation maximization algorithm can be performed by calculating the first posterior probability of each trajectory point belonging to each Gaussian component according to the horizontal coordinate of each trajectory point in the expectation step (Expectation Step, E step). In the maximization step (Maximization Step, M step), the parameters of the Gaussian mixture model are updated based on the first posterior probability. Repeat the expectation step and the maximization step until the parameters of the Gaussian mixture model change less than the preset convergence threshold or reach the maximum iteration number, to obtain the trained parameters.

[0102] Exemplarily, in the E step, for each observed trajectory point (also referred to as data points below) and each Gaussian component k, the first posterior probability (responsibility) of the trajectory point belonging to the k-th Gaussian component is calculated:

[0103]

[0104] In the formula, xi belongs to the k-th Gaussian component, and j represents an index variable for traversing all Gaussian components.

[0105] In the M step, the model parameters are updated based on the responsibility calculated in the E step:

[0106] The mixing weight is updated , wherein . represents the number of effective data points of the k-th Gaussian component in the t-th iteration. ​denotes the first posterior probability of the data point xi in the t-th iteration belonging to the k-th Gaussian component.

[0107] updating the mean .

[0108] updating the variance .

[0109] convergence judgment: calculate the parameter change or the change of the likelihood function value between two iterations, if it meets (convergence threshold, such as 1e-4) or reaches the preset maximum iteration number (such as 100 times), stop iteration and output the final model parameters. Otherwise, return the E-step.

[0110] The GMM model is trained by the above-mentioned EM algorithm to iteratively update the parameters of the GMM model, so that the model can better fit the distribution of the data.

[0111] In some embodiments, in addition to training the GMM model by the EM algorithm, the GMM model can also be trained in any other way, and the training method is not particularly limited in the present embodiment. For example, the GMM model can also be trained by variational inference, which is a technique for approximating posterior distribution, capable of handling large-scale data sets and complex models, and by optimizing the variational lower bound, the uncertainty in the model can be better handled.

[0112] In one embodiment, the above-mentioned obtaining the Gaussian distribution of the trajectory points in each lane segment according to the trained Gaussian mixture model can be performed as follows: for each lane segment, the second posterior probability of each trajectory point in the trained Gaussian mixture model is calculated according to the lateral coordinate of each trajectory point. And according to the second posterior probability, each trajectory point is assigned to the corresponding Gaussian component to obtain the assignment result of the Gaussian component. Further, the Gaussian distribution of the trajectory points in each lane segment is determined according to the assignment result of the Gaussian component.

[0113] After the iterative training of the Gaussian mixture model is completed, for each trajectory point xi in each lane segment, the second posterior probability of the trajectory point in the trained Gaussian mixture model, i.e. the probability of the trajectory point belonging to each Gaussian component, is calculated. The calculation method of the second posterior probability is similar to that of the first posterior probability, which will not be described here. According to the calculation result of the second posterior probability, the trajectory point is assigned to the Gaussian component k with the maximum probability, and the Gaussian distribution of each trajectory point is obtained.

[0114] It should be noted that the first posterior probability and the second posterior probability in the embodiment are used to represent similar objects, and the first posterior probability and the second posterior probability can be the same posterior probability or different posterior probabilities. For example, the first posterior probability and the second posterior probability used for the last iteration of training the model are the same posterior probability.

[0115] In some embodiments, in addition to determining the Gaussian distribution of the trajectory point by calculating the second posterior probability, the Gaussian distribution of the trajectory point can also be determined by using maximum likelihood estimation (MLE), minimum distance, etc. For example, for each trajectory point, the likelihood value, i.e., the value of the probability density function, under each Gaussian component is calculated, and the trajectory point is assigned to the Gaussian component with the maximum likelihood value. This method is similar to the posterior probability method, but does not consider the influence of the mixing weight. Alternatively, the Euclidean distance of each trajectory point to the mean of each Gaussian component is calculated, and the trajectory point is assigned to the nearest Gaussian component. This method is simpler and is suitable for the case where there is a clear separation between Gaussian components.

[0116] Continuing to refer to Figure 2b , in step S2031, when the distribution result exists a multi-peak distribution, it is determined that there are multiple driving paths in the lane segment.

[0117] In the embodiment, by using the trained Gaussian mixture model, it is determined whether the trajectory data exists a multi-peak distribution, i.e., whether there are multiple Gaussian distributions corresponding to multiple different groups of trajectory points. In the intersection scenario, for the lane segment, these groups of trajectory points correspond to multiple driving paths that can be derived at the intersection according to driving habits, road obstacles, etc.

[0118] By using the above technical solution, the trajectory in the intersection is lane-level matched and positioned, and whether there are multiple trajectory paths in the lane segment is identified by using the trajectory data in each lane segment to train the Gaussian mixture model in the manner of lane segmentation, so as to realize the lane-level trajectory path recognition, thereby improving the accuracy of the intersection trajectory path recognition and being able to adapt to diversified complex road scenarios.

[0119] Figure 4a is a flowchart of another trajectory path recognition method provided by the embodiment of the application, as shown in Figure 4a , can include the following steps S401-S408.

[0120] In step S401, according to the matching result between the vehicle trajectory data and the road data and the preset lane model, the original trajectory point in the vehicle trajectory data is positioned in the target lane, and the target lane is a lane matched with the position information of the original trajectory point.

[0121] Step S402, according to the preset segmentation rule, the target lane is divided into a plurality of lane segments, and the original trajectory points in each lane segment are preprocessed to obtain preprocessed trajectory points.

[0122] Step S403, for the trajectory points in each lane segment, the preset Gaussian mixture model is trained to obtain the Gaussian distribution of the trajectory points in each lane segment according to the trained Gaussian mixture model to obtain a distribution result; wherein the Gaussian mixture model is a Gaussian mixture model containing a preset number of Gaussian components.

[0123] Step S404, when the distribution result has a multi-peak distribution, it is determined that there are multiple driving paths in the lane segment.

[0124] It should be noted that the steps S401-S404 described above are similar to the steps in the embodiments, and the related descriptions will not be repeated here. Figure 2b

[0125] Step S405, the lane segment with a multi-peak distribution is selected from the plurality of lane segments as a multi-peak lane segment.

[0126] In this embodiment, by finely segmenting the lane, for each lane segment, it can be identified whether there is a multi-peak distribution, and by screening out the multi-peak distribution lane segment, it is convenient for subsequent extraction of multiple driving paths.

[0127] Step S406, for each multi-peak lane segment, the forward continuity rate and / or the backward continuity rate between each Gaussian distribution in the multi-peak lane segment and the corresponding Gaussian distribution in the adjacent multi-peak lane segment are determined.

[0128] In one embodiment, the forward continuity rate and / or the backward continuity rate between each Gaussian distribution in the multi-peak lane segment and the corresponding Gaussian distribution in the adjacent multi-peak lane segment can be determined as follows:

[0129] Exemplarily, the calculation method of the backward continuity rate can be that according to the number of overlaps between all trajectory points of each Gaussian distribution in the multi-peak lane segment and all trajectory points of the corresponding Gaussian distribution in the previous multi-peak lane segment, the backward continuity of the multi-peak lane segment is determined.

[0130] ​Wherein, the backward continuity rate is for each multi-peak lane segment, the continuity between each Gaussian distribution in the segment and the corresponding Gaussian distribution in the adjacent previous multi-peak lane segment is calculated. In this embodiment, the continuity between the Gaussian distribution and the corresponding Gaussian distribution in the previous multi-peak lane segment can be calculated by obtaining the set of all trajectory points in the multi-peak lane segment, and calculating the number of overlaps between the set of trajectory points of the current Gaussian distribution and the set of trajectory points of the corresponding Gaussian distribution in the previous multi-peak lane segment (the corresponding Gaussian distribution can be determined in combination with the position or similarity between the Gaussian distributions). The number of overlaps can be obtained by calculating the intersection of the two sets, and the number of overlaps is used as a measure of the backward continuity rate.

[0131] Further, the backward continuity rate r1 can be expressed by the formula: , is the set of trajectory points of the current Gaussian distribution, is the set of trajectory points in the previous multi-peak lane segment. Wherein, the determination of the trajectory point overlap can be determined by the vehicle speed, vehicle identification and other information whether it is the same trajectory point.

[0132] For example, the calculation method of the forward continuity rate can be that according to the number of overlaps between all trajectory points of each Gaussian distribution in the multi-peak lane segment and all trajectory points of the corresponding Gaussian distribution in the subsequent multi-peak lane segment, the forward continuity of the multi-peak lane segment is determined.

[0133] Correspondingly, the forward continuity rate is for each multi-peak lane segment, the continuity between each Gaussian distribution in the segment and the corresponding Gaussian distribution in the adjacent next multi-peak lane segment is calculated. The forward continuity rate which can be expressed by the formula: , is the set of trajectory points in the subsequent multi-peak lane segment.

[0134] Step S407, according to the forward continuity rate and / or the backward continuity rate, the reliable multi-peak lane segment is screened out from the multi-peak lane segment.

[0135] In this embodiment, the forward continuity rate (for example, there is no multi-peak lane segment behind the multi-peak lane segment), or the backward continuity rate (for example, there is no multi-peak lane segment in front of the multi-peak lane segment), or the combination of the forward continuity rate and the backward continuity rate (for example, there are lane segments in front of and behind the multi-peak lane segment) can be used alone to screen out the reliable multi-peak lane segment. For example, the multi-peak lane segment with a bidirectional continuity rate exceeding 80% can be screened out as the reliable multi-peak lane segment.

[0136] In this process, the reliable multi-peak lane segment is screened out by calculating the forward and backward continuity rates and combining the forward and backward continuity rates, so as to extract the reliable multiple driving paths on the road.

[0137] Step S408, determining a plurality of final driving paths according to the reliable multi-peak lane segments.

[0138] In an embodiment, the determination of the plurality of final driving paths according to the reliable multi-peak lane segments can be implemented in the following manner: for each reliable multi-peak segment, the average trajectory position of each Gaussian distribution in the reliable multi-peak segment is obtained; all average trajectory positions in each reliable multi-peak lane segment are clustered to obtain trajectory groups for each reliable multi-peak lane segment; and the plurality of final driving paths are determined according to the trajectory groups.

[0139] In this embodiment, for each Gaussian distribution in each reliable multi-peak lane segment, the average trajectory position of all trajectory points in the Gaussian distribution is calculated. The average trajectory position can be calculated by the average position of the trajectory points, can be calculated by the average position of the trajectory points, or can be obtained by the calculation method based on the posterior probability in the example below. By calculating the average trajectory position, the average trajectory position can represent the center position of the Gaussian distribution, and a more simplified representation is provided for subsequent clustering analysis.

[0140] In this embodiment, all average trajectory positions in each reliable multi-peak lane segment are clustered, and a clustering algorithm (such as K-means, DBSCAN or hierarchical clustering) is used to cluster similar average trajectory positions together to obtain trajectory groups, so that each trajectory group represents a potential driving path, and each group contains average trajectory positions belonging to the same driving path, and different driving paths existing in the lane segment can be identified.

[0141] By integrating the average trajectory positions in each group according to the trajectory groups, a complete driving path is formed. For example, according to the trajectory groups, the trajectory points in each trajectory group are processed by stringing, and a plurality of driving paths corresponding to a plurality of trajectory groups can be obtained. By this method, possible driving paths can be effectively extracted from complex traffic environment, and these driving paths can correspond to a plurality of path sets under different driving modes or obstacle facilities, so as to facilitate high-precision map mapping and path optimization of autonomous driving.

[0142] In combination with the actual road scene, the reliable multi-peak lane segment is usually continuous. Considering special road scenes or too many noise points in the trajectory data, the reliable multi-peak lane segment is not continuous (for example, there are unreliable multi-peak lane segments between the reliable multi-peak lane segments), and the average trajectory position data can be smoothed by interpolation of the average trajectory position, so that the extracted trajectory paths are continuous and reliable.

[0143] Further, the process of obtaining the average trajectory position of each Gaussian distribution in the reliable multi-peak segment can be: obtaining the second posterior probability of all trajectory points in the reliable multi-peak segment in each Gaussian distribution; and calculating the average trajectory position of each Gaussian distribution in the reliable multi-peak segment according to the second posterior probability of all trajectory points in each Gaussian distribution.

[0144] Exemplarily, the sliding average based on the GMM posterior probability (corresponding to the reliable multi-peak lane segment) realizes trajectory grouping:

[0145] For each trajectory point in the sliding window, the weighted average value of each Gaussian component is calculated according to the posterior probability thereof in the GMM. Specifically, for each trajectory point xi, the posterior probability γik thereof on each Gaussian component k is calculated. Then, the sliding average value of each Gaussian component, i.e., the average trajectory position, is calculated:

[0146]

[0147] wherein, is the average trajectory position of the kth Gaussian component in the current sliding window.

[0148] In this way, the posterior probability provides the probability of each data point belonging to a certain Gaussian component, and by using these probabilities as weights, the average trajectory of the Gaussian distribution can be more accurately calculated.

[0149] For the convenience of understanding the embodiments of the present application, in an exemplary embodiment, as shown in Figure 4b , the process can include trajectory matching, trajectory positioning lane, Gaussian mixture model (GMM) modeling and training, and trajectory grouping (in the grouping process, a spatiotemporal continuity verification mechanism, i.e., a bidirectional continuity rate verification, can be used), etc. As shown in Figure 4b , first, vehicle trajectory data, i.e., original trajectories, are obtained. Then, the trajectory data are matched with road data, and the trajectory points are positioned to the lanes and segmented, and then the trajectory lateral distribution, i.e., the lateral coordinates of the trajectory points, are calculated. Then, Gaussian mixture modeling is performed, the input of the trajectory lateral distribution and the E step (Expectation Step) and M step (Maximization Step) in the EM algorithm are used to iteratively update the Gaussian mixture model, and by judging whether the convergence threshold is reached or the iteration number is reached, the iteration is continued when the above conditions are not met, or the trained Gaussian mixture model is used to identify whether there is a multi-peak distribution in the lane segment when the above conditions are met, and further, the trajectory grouping step is performed to output the multi-trajectory path according to the trajectory grouping.

[0150] Taking the intersection scenario as an example, as shown in Figure 5As shown, in the actual intersection scene, there is a physical facility isolation (barrier) after the intersection turn, and the vehicle has multiple driving modes (such as the two situations shown by the arrows). Figure 6 For the trajectory data in this intersection scene, the conventional clustering method can usually only identify one main path in the straight direction. The embodiment combines trajectory point lane-level positioning, lane segmentation, and Gaussian mixture model recognition, and can accurately identify multiple trajectory paths in the intersection scene through multi-peak distribution recognition, such as Figure 7a and Figure 7b As shown, the two lines pointed to by the arrows in Figure 7b In addition, based on the multi-peak distribution recognition method of lane segmentation in the embodiment, since the multi-peak distribution in each lane segment is recognized, combined with the traffic flow characteristics of the lane, the (multiple trajectory generation) starting point and the (multiple trajectory merging) end point position range where multiple trajectory paths exist can be efficiently located.

[0151] In summary, the technical solution provided by the embodiment binds the trajectory data with the driving road or intersection, obtains the space-time information of the trajectory, and extracts based on the lateral trajectory feature distribution of the road to consolidate the data foundation. The multi-peak distribution of the trajectory sample is accurately captured by using the Gaussian mixture model, and the adaptive control of the parameters is realized by the EM algorithm, which enhances the adaptability of the model. In addition, the rationality of the path topology is guaranteed by the bidirectional continuity rate verification, and the continuity of the path is enhanced by the sliding average grouping. Multiple trajectory paths in complex road scenes such as intersections can be extracted, which provides detailed intersection path data for high-precision map construction, enriches the details of the map scene, and helps the automatic driving vehicle to plan the intersection passing route in advance. Precise planning improves driving safety.

[0152] Figure 8 is a structural schematic diagram of a trajectory path recognition device provided by the embodiment, as shown in Figure 8 The device 800 includes a lane positioning module 801, a model processing module 802, and a multi-peak recognition module 803. The lane positioning module 801 is configured to position the original trajectory points in the vehicle trajectory data to the target lane according to the matching result between the vehicle trajectory data and the road data and the pre-set lane model, and the target lane is the lane matched with the original trajectory point position information. The model processing module 802 is configured to train the pre-set Gaussian mixture model for the trajectory points in the target lane to obtain the Gaussian distribution of the trajectory points according to the trained Gaussian mixture model to obtain the distribution result. The pre-set Gaussian mixture model is a Gaussian mixture model containing a pre-set number of Gaussian components. The multi-peak recognition module 803 is configured to determine that there are multiple driving paths in the target lane when the distribution result has a multi-peak distribution.

[0153] In an embodiment, the matching result comprises a correspondence between an original trajectory point in the vehicle trajectory data and a corresponding road; the lane model comprises at least one lane included in each road; and the lane positioning module 801 comprises: a distance obtaining unit configured to, for each original trajectory point, obtain a boundary distance of the original trajectory point to each lane in the road to which the original trajectory point corresponds according to the road to which the original trajectory point corresponds and position information of the original trajectory point; and a lane determining unit configured to determine a target lane in which the original trajectory point is located according to the boundary distance of the original trajectory point to each lane in the road to which the original trajectory point corresponds.

[0154] In an embodiment, the model processing module 802 comprises: an initialization unit configured to initialize parameters of the Gaussian mixture model, the parameters comprising a mixture weight, a mean value and a variance, and set a number of Gaussian components; and a training unit configured to train the parameters of the Gaussian mixture model according to an expectation maximization algorithm, so as to obtain a trained Gaussian mixture model according to the trained parameters.

[0155] In an embodiment, the training unit is specifically configured to: in an expectation step, calculate a first posterior probability of each trajectory point belonging to each Gaussian component according to a lateral coordinate of each trajectory point; in a maximization step, update the parameters of the Gaussian mixture model based on the first posterior probability; and repeat the expectation step and the maximization step until the parameters of the Gaussian mixture model change less than a preset convergence threshold or a maximum iteration number is reached, so as to obtain the trained parameters.

[0156] In an embodiment, the apparatus further comprises: a lane segmentation module configured to divide the target lane into a plurality of lane segments according to a preset segmentation rule, and pre-process the original trajectory points in each lane segment to obtain pre-processed trajectory points; the distribution result comprises a distribution result of each lane segment; and the model processing module 802 is specifically configured to: for all trajectory points in each lane segment, train the preset Gaussian mixture model, so as to obtain a Gaussian distribution of the trajectory points in each lane segment according to a trained Gaussian mixture model, and determine that there are a plurality of driving paths in the corresponding lane segment when the distribution result of the lane segment has a multimodal distribution.

[0157] In an embodiment, the model processing module 802 comprises: a first calculation unit configured to, for each lane segment, calculate a second posterior probability of each trajectory point in the trained Gaussian mixture model according to a lateral coordinate of each trajectory point; an assignment module configured to assign each trajectory point to a corresponding Gaussian component according to the second posterior probability, so as to obtain an assignment result of the Gaussian components; and a distribution determining module configured to determine a Gaussian distribution of the trajectory points in each lane segment according to the assignment result of the Gaussian components.

[0158] In an embodiment, the lane segmentation module comprises: a first dividing unit configured to divide the target lane into a plurality of lane segments according to a reference line curvature of a road reference line of the target lane, the lane segments being equal-length lane segments or unequal-length lane segments; or a second dividing unit configured to divide the target lane into a plurality of equal-length lane segments according to a predetermined fixed segmentation step, the lane segments being equal-length lane segments.

[0159] In an embodiment, the lane segmentation module comprises: a position determining unit configured to determine, for each lane segment, a position coordinate of a to-be-inserted trajectory point in the lane segment according to a preset linear interpolation algorithm if there is no original trajectory point in the lane segment; and an interpolation unit configured to insert a new trajectory point in the lane segment according to the position coordinate of the to-be-inserted trajectory point to obtain the preprocessed trajectory points.

[0160] In an embodiment, the device further comprises: a segment screening module configured to screen lane segments with a multimodal distribution from the plurality of lane segments as multimodal lane segments; a continuity rate determining module configured to determine, for each multimodal lane segment, a forward continuity rate and / or a backward continuity rate between each Gaussian distribution in the multimodal lane segment and a corresponding Gaussian distribution in an adjacent multimodal lane segment; the segment screening module is further configured to screen reliable multimodal lane segments from the multimodal lane segments according to the forward continuity rate and / or the backward continuity rate; and a path determining module configured to determine the final plurality of driving paths according to the reliable multimodal lane segments.

[0161] In an embodiment, the continuity rate determining module comprises: a backward continuity rate determining unit configured to determine a backward continuity of a multimodal lane segment according to a number of overlaps between all trajectory points of each Gaussian distribution in the multimodal lane segment and all trajectory points of a corresponding Gaussian distribution in a previous multimodal lane segment; and / or a forward continuity rate determining unit configured to determine a forward continuity of the multimodal lane segment according to a number of overlaps between all trajectory points of each Gaussian distribution in the multimodal lane segment and all trajectory points of a corresponding Gaussian distribution in a subsequent multimodal lane segment.

[0162] In an embodiment, the path determining module comprises: a trajectory position obtaining unit configured to obtain, for each reliable multimodal segment, an average trajectory position of each Gaussian distribution in the reliable multimodal segment; a clustering grouping unit configured to cluster all average trajectory positions in each reliable multimodal lane segment to obtain a trajectory grouping for each reliable multimodal lane segment; and a path generating unit configured to determine the final plurality of driving paths according to the trajectory groupings.

[0163] In one implementation, the trajectory position acquisition unit is specifically configured to: acquire the second posterior probability of all trajectory points in the reliable multi-peak segment in each Gaussian distribution; and calculate the average trajectory position of each Gaussian distribution in the reliable multi-peak segment based on the second posterior probability of all trajectory points in each Gaussian distribution.

[0164] The relevant explanations can be understood by referring to the relevant descriptions and effects corresponding to the steps in the method embodiments of this application, and will not be elaborated further here.

[0165] Figure 9 This application also provides an electronic device, such as... Figure 9 As shown, the electronic device can be Figure 1 In the example scenario, the server, the electronic device may include: a processor 901, and a memory 902 communicatively connected to the processor 901; the memory 902 stores computer-executable instructions; the processor 901 executes the computer-executable instructions stored in the memory 902 to implement the trajectory path recognition method in the above embodiments, wherein the memory 902 and the processor 901 are connected via a bus 903.

[0166] The relevant explanations can be understood by referring to the relevant descriptions and effects corresponding to the steps in the method embodiments of this application, and will not be elaborated further here.

[0167] This application also provides a chip, including a memory and a processor. The memory is used to store a computer program, and the processor is used to call and run the computer program from the memory to execute a trajectory path recognition method.

[0168] The relevant explanations can be understood by referring to the relevant descriptions and effects corresponding to the steps in the method embodiments of this application, and will not be elaborated further here.

[0169] One embodiment of this application provides a computer-readable storage medium storing a computer program thereon. The computer program is executed by a processor to implement the trajectory path recognition method provided in any embodiment of the method embodiments of this application. The computer-readable storage medium may be a ROM, random access memory (RAM), CD-ROM, magnetic tape, floppy disk, or optical data storage device, etc.

[0170] The relevant explanations can be understood by referring to the relevant descriptions and effects corresponding to the steps in the method embodiments of this application, and will not be elaborated further here.

[0171] One embodiment of this application provides a computer program product, including a computer program that, when executed by a processor, implements the trajectory path recognition method provided in the method embodiment of this application.

[0172] The related description can be understood by referring to the related description and effects of the steps in the method embodiments of the present application. Details are not described here.

[0173] In several embodiments provided in the present application, it should be understood that the disclosed apparatus and method can be implemented in other ways. For example, the apparatus embodiments described above are only illustrative. For example, the division of the modules is only a logical function division. In actual implementation, another division manner can be used. For example, a plurality of modules or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the displayed or discussed ones can be indirect coupling or communication connection through some interfaces, apparatuses or modules, which can be electrical, mechanical or other forms.

[0174] Other embodiments of the present application will be apparent to those skilled in the art from consideration of the specification and practice of the application disclosed herein. The specification and examples given are exemplary only and the true scope and spirit of the application is indicated by the following claims. It will be appreciated by persons skilled in the art that numerous variations and / or modifications can be made to the application as described above without departing from the scope or spirit of the application. It is intended that all such variations and / or modifications be included within the scope of the application. The specification and examples given are exemplary only and the true scope and spirit of the application is indicated by the following claims.

[0175] It should be understood that the application is not limited to the precise construction that has been described above and shown in the accompanying drawings, and that various modifications and changes can be made by those skilled in the art without departing from the scope of the application. The scope of the application is limited only by the appended claims.

Claims

1. A travel path recognition method characterized by comprising: The method comprises the following steps: According to the matching result between the vehicle trajectory data and the road data and the preset lane model, the original trajectory points in the vehicle trajectory data are positioned in the target lane which is the lane matched with the position information of the original trajectory points; For the trajectory points in the target lane, the preset Gaussian mixture model is trained to obtain the Gaussian distribution of the trajectory points according to the trained Gaussian mixture model, and the distribution result is obtained; wherein the preset Gaussian mixture model is a Gaussian mixture model containing a preset number of Gaussian components; When the distribution result has a multi-peak distribution, it is determined that there are multiple driving paths in the target lane.

2. The method of claim 1, wherein, The matching result includes the association of the original trajectory points in the vehicle trajectory data to the corresponding road; the lane model includes at least one lane contained in each road; According to the matching result between the vehicle trajectory data and the road data and the preset lane model, the original trajectory points in the vehicle trajectory data are positioned in the target lane which is the lane matched with the position information of the original trajectory points; For each original trajectory point, the boundary distance of the original trajectory point to each lane in the associated road is obtained according to the associated road of the original trajectory point and the position information of the original trajectory point; According to the boundary distance of the original trajectory point to each lane in the associated road, the target lane where the original trajectory point is located is determined.

3. The method according to claim 1 or 2, characterized in that, The training of the preset Gaussian mixture model comprises: Initializing the parameters of the Gaussian mixture model, including the mixing weight, the mean and the variance; According to the expectation maximization algorithm, the parameters of the Gaussian mixture model are trained to obtain the trained Gaussian mixture model according to the trained parameters.

4. The method of claim 3, wherein, According to the expectation maximization algorithm, the parameters of the Gaussian mixture model are trained, comprising: In the expectation step, the first posterior probability of each trajectory point belonging to each Gaussian component is calculated according to the lateral coordinate of each trajectory point; In the maximization step, the parameters of the Gaussian mixture model are updated based on the first posterior probability; The expectation step and the maximization step are repeatedly executed until the change of the parameters of the Gaussian mixture model is less than the preset convergence threshold or the maximum iteration number is reached, to obtain the trained parameters.

5. The method according to any one of claims 1 to 4, characterized in that, Further comprising: According to the preset segmentation rule, the target lane is divided into multiple lane segments, and the original trajectory points in each lane segment are preprocessed to obtain the preprocessed trajectory points; The distribution result includes the distribution result of each lane segment; The training of the preset Gaussian mixture model for the trajectory points in the target lane comprises: For all trajectory points in each lane segment, the preset Gaussian mixture model is trained to obtain the Gaussian distribution of the trajectory points in each lane segment according to the trained Gaussian mixture model, and when the distribution result of the lane segment has a multi-peak distribution, it is determined that there are multiple driving paths in the corresponding lane segment.

6. The method of claim 5, wherein, The Gaussian distribution of the trajectory points in each lane segment is obtained according to the trained Gaussian mixture model, comprising: For each lane segment, a second posterior probability of each trajectory point in the trained Gaussian mixture model is calculated according to a lateral coordinate of each trajectory point; Each trajectory point is assigned into a corresponding Gaussian component according to the second posterior probability, to obtain an assignment result of the Gaussian components; A Gaussian distribution of the trajectory points in each lane segment is determined according to the assignment result of the Gaussian components.

7. The method of claim 5, wherein, The target lane is divided into a plurality of lane segments according to the preset segmentation rule, including: The target lane is divided into a plurality of lane segments according to a variable segmentation step determined according to a reference line curvature of a road reference line of the target lane, the lane segments being lane segments with equal lengths or unequal lengths; Or, the target lane is divided into a plurality of lane segments with equal lengths according to a predetermined fixed segmentation step, the lane segments being lane segments with equal lengths.

8. The method of claim 5, wherein, Further comprising: Filtering, from the plurality of lane segments, a lane segment with a multimodal distribution as a multimodal lane segment; For each multimodal lane segment, determining a forward continuity rate and / or a backward continuity rate between each Gaussian distribution in the multimodal lane segment and a corresponding Gaussian distribution in an adjacent multimodal lane segment; Filtering, from the multimodal lane segments, a reliable multimodal lane segment according to the forward continuity rate and / or the backward continuity rate; Determining a final plurality of driving paths according to the reliable multimodal lane segment.

9. The method of claim 8, wherein, The determination of the forward continuity rate and / or the backward continuity rate between each Gaussian distribution in the multimodal lane segment and a corresponding Gaussian distribution in an adjacent multimodal lane segment includes: Determining a backward continuity of the multimodal lane segment according to an overlapping number between all trajectory points of each Gaussian distribution in the multimodal lane segment and all trajectory points of a corresponding Gaussian distribution in a previous multimodal lane segment; And / or, determining a forward continuity of the multimodal lane segment according to an overlapping number between all trajectory points of each Gaussian distribution in the multimodal lane segment and all trajectory points of a corresponding Gaussian distribution in a next multimodal lane segment.

10. The method of claim 8, wherein, The determination of the final plurality of driving paths according to the reliable multimodal lane segment includes: For each reliable multimodal segment, obtaining an average trajectory position of each Gaussian distribution in the reliable multimodal segment; Clustering all average trajectory positions in each reliable multimodal lane segment to obtain a trajectory grouping about each reliable multimodal lane segment; Determining a final plurality of driving paths according to the trajectory grouping.

11. A trajectory path recognition apparatus characterized by comprising: Comprising: A lane positioning module configured to position original trajectory points in vehicle trajectory data into a target lane according to a matching result between the vehicle trajectory data and road data and a preset lane model, the target lane being a lane matched with position information of the original trajectory points; A model processing module configured to train a preset Gaussian mixture model for the trajectory points in the target lane to obtain a Gaussian distribution of the trajectory points according to the trained Gaussian mixture model, to obtain a distribution result; wherein the preset Gaussian mixture model is a Gaussian mixture model containing a preset number of Gaussian components. A multi-peak identification module is configured to determine that there are multiple driving paths in the target lane when the distribution result is a multi-peak distribution.

12. An electronic device / computer readable storage medium / computer program product, characterized in that, Comprise: a memory and a processor; the memory stores computer-executable instructions; the processor executes the computer-executable instructions stored by the memory, so that the electronic device executes the trajectory path identification method in any one of claims 1-10; and / or, the computer-readable storage medium stores computer-executable instructions, and the computer-executable instructions are executed by the processor to implement the trajectory path identification method in any one of claims 1-10; and / or, the computer program product comprises a computer program, and the computer program is executed by the processor to implement the trajectory path identification method in any one of claims 1-10.

Images