A typical lane-changing trajectory determination method, device and computer storage medium based on lane-changing scene data

Abstract

A method, device and computer storage medium for determining a typical lane change trajectory based on lane change scene data, relating to the field of autonomous driving technology. The method solves the problems of low extraction rate of common features of trajectories in existing lane change trajectory extraction technologies and the loss of practical application capability of the obtained lane change trajectory in automatic steering control or autonomous driving due to insufficient consideration of factors during feature extraction. In the method, parameters characterizing the lane change trajectory are extracted based on lane change scene data collected by an autonomous driving vehicle, and a trajectory expression is obtained based on the parameters; the lane change trajectory is then segmented into a number of sub-trajectories; the parameter features of each sub-trajectory are converted and summarized to obtain a lane change trajectory data set; and the data set is clustered to obtain a typical lane change trajectory. The present invention is a lane change trajectory obtained based on an actual lane change scenario, and is more reliable when applied to the field of autonomous driving technology.

Description

Technical Field

[0001] The present invention relates to the field of autonomous driving technology, and in particular to trajectory determination technology for lane change scenarios. Background Art

[0002] In the field of autonomous driving technology, trajectory control and planning during vehicle driving are the key to reliable, safe and stable operation of vehicles. Lane changing is a basic action in the vehicle driving process. Studying typical vehicle lane changing trajectories plays an important role in the research of steering control of autonomous driving vehicles and the development of autonomous driving functions.

[0003] At present, the existing lane change trajectory extraction technology generally uses linear fitting and clustering methods to extract typical lane change trajectories. Such technologies often ignore the local change information of the lane change trajectory and cannot effectively extract the common features of the trajectory. At the same time, the existing technology for extracting typical lane change trajectories often only considers the position feature information of the vehicle during the lane change process, ignoring the influence of lane change time, vehicle speed, turning angle and other features on the lane change trajectory, and cannot effectively extract lane change trajectories with typical features, making the developed automatic steering control and automatic driving functions not typical and losing the ability to be actually applied. Summary of the invention

[0004] The present invention solves the problems that the existing lane change trajectory extraction technology has a low extraction rate of common trajectory features and the lane change trajectory obtained loses the ability to be actually applied in automatic steering control or automatic driving due to insufficient consideration of factors during feature extraction.

[0005] A typical lane-changing trajectory determination method based on lane-changing scene data according to the present invention includes:

[0006] The steps of trajectory parameter extraction and trajectory expression are as follows: extracting parameters representing the lane change trajectory based on the lane change scene data collected by the autonomous driving vehicle, and obtaining the trajectory expression based on the parameters;

[0007] Lane change trajectory segmentation step: segment each lane change trajectory into N equal segments, each segment is a sub-trajectory;

[0008] Sub-trajectory parameter feature conversion step: extract trajectory parameters from each sub-trajectory, and perform feature conversion on the trajectory parameters to obtain the parameter feature of the nth sub-trajectory in the mth trajectory:

[0009] TS' mn =[Angel' mn ,Speed' mn , Position_X' mn , Position_Y m ' n],

[0010] Among them, Angel` mn Represents the steering wheel angle of the nth sub-track in the mth track, Speed` mn Represents the speed of the nth sub-track in the mth track, Position_X` mn Indicates the lateral distance span of the nth sub-track in the mth track, Position_Y` mn represents the longitudinal distance span of the nth sub-track in the mth track; the data after feature conversion are aggregated to obtain the lane change track data set;

[0011] Trajectory clustering step: clustering the lane change trajectory data set using the k-means algorithm, using the cluster center to reflect the typical trajectory features, thereby completing the extraction and determination of the typical lane change trajectory.

[0012] Furthermore, the parameters characterizing the lane change trajectory include four parameters: steering wheel angle Angle, vehicle speed Speed, vehicle longitudinal position Position_X, and vehicle lateral position Position_Y.

[0013] Further, the trajectory-based expression is: TD={TR1, TR1, ..., TRM}, M is the total amount of all lane change trajectory data, TR1 represents the first lane change trajectory, TR2 represents the second lane change trajectory, ..., TRM represents the Mth lane change trajectory.

[0014] Furthermore, the lane change trajectory dataset DR is:

[0015]

[0016] Among them: TS 11 is the trajectory of the first trajectory segment in the first lane change trajectory, and so on;

[0017] The trajectory of the nth segment in the mth lane change trajectory is expressed as:

[0018] TS mn =[Cond m,(n-1)j / N+1 ,Cond m,(n-1)j / N+2 ,......,Cond m,nj / N ],

[0019] Among them, 1≤n≤N, 1≤m≤M, among which Cond m,(n-1)j / N+1 is the feature at time (n-1)j / N+1 in the n-th segment of the m-th lane change trajectory.

[0020] Furthermore, the Angel` mn is based on

[0021]

[0022] Obtained, of which express

[0023] <Angel m,(n-1)j / N+1 , Angel m,(n-1)j / N+2 ,......,Angel m,nj / N >

[0024] The maximum value in ;

[0025] express

[0026] <Angel m,(n-1)j / N+1 , Angel m,(n-1)j / N+2 ,......,Angel m,nj / N >

[0027] The minimum value in ;

[0028] The Speed` mn is through Obtained, of which: express

[0029] <Speed m,(n-1)j / N+1 ,Speed m,(n-1)j / N+2 ,......,Speed m,nj / N >

[0030] The maximum value in ;

[0031] express

[0032] <Speed m,(n-1)j / N+1 ,Speed m,(n-1)j / N+2 ,......,Speed m,nj / N >

[0033] The minimum value in ;

[0034] express

[0035] <Speed m,(n-1)j / N+1 ,Speed m,(n-1)j / N+2 ,......,Speed m,nj / N > the mean value;

[0036] Position_X` mn and Position_Y` mn is through

[0037]

[0038]

[0039] Obtained, of which:

[0040] express

[0041] <Position_X m,(n-1)j / N+1 , Position_X m,(n-1)j / N+2 ,......,Position_X m,nj / N >

[0042] The maximum value in

[0043] express

[0044] <Position_X m,(n-1)j / N+1 , Position_X m,(n-1)j / N+2 ,......,Position_X m,nj / N >

[0045] The minimum value in

[0046] express

[0047] <Position_X m,(n-1)j / N+1 , Position_X m,(n-1)j / N+2 ,......,Position_X m,nj / N >

[0048] The maximum value in

[0049] express

[0050] <Position_X m,(n-1)j / N+1 , Position_X m,(n-1)j / N+2 ,......,Position_X m,nj / N >

[0051] The minimum value in .

[0052] Furthermore, the method of clustering the lane change trajectory dataset using the k-means algorithm to reflect typical trajectory features with cluster centers is as follows: randomly selecting k samples from the feature-converted trajectory feature lane change trajectory dataset as initial k centroid vectors:

[0053] {μ1, μ2, ..., μ k};

[0054] Repeat the following process until convergence:

[0055] A) Divide the clusters into Cj, j = 1, 2, 3, ..., k, k is the total number of clusters;

[0056] B) For each lane change trajectory TRi to each center of mass

[0057] μ j , j = 1, 2, 3, ..., k,

[0058] Where i = 1, 2, 3, ..., M, and then further obtain the distance d between each centroid ij , the TRi class

[0059] It is marked as Cj corresponding to the minimum dij;

[0060] For j = 1, 2, 3, ..., k, for C j Recalculate the new center of mass based on all lane change trajectory data;

[0061]

[0062] If all k centroid vectors have not changed, indicating that the data has converged, go to the next step, otherwise return to step A) to continue;

[0063] Get C = {C1, C2, ..., Ck} and the corresponding centroid {μ1, μ2, ..., μ k}, the centroid is the typical lane change trajectory feature extracted by the present invention.

[0064] Furthermore, the distance dij between the centroids is obtained by the following method:

[0065] a) First, define the weight vector of the steering wheel angle, vehicle speed, vehicle longitudinal position, and vehicle lateral position features as:

[0066] W={W Angel , W Speed , W Position_X , W Position_Y}

[0067] Among them: the weight value of each feature is greater than or equal to zero and satisfies

[0068] W Angel +W Speed +W Position_X +W Position_Y =1,

[0069] b) Calculate the distance between two trajectories, assuming that the two trajectories to be calculated are TR1 and TR2:

[0070] TR1=[TS 11, TS 12 , ......., TS 1N ], TR2 = [TS 21 , TS 22 , ......., TS 2N ],

[0071] The distance between the two trajectories is:

[0072]

[0073] in:

[0074]

[0075] The above-mentioned typical lane change trajectory determination method based on lane change scene data is a technical solution that can be implemented by computer software. Therefore, based on the same design idea, the device description of the solution implemented by computer software is as follows:

[0076] A typical lane-changing trajectory determination device based on lane-changing scene data includes the following modules:

[0077] Trajectory parameter extraction and trajectory expression module: used to extract parameters representing lane change trajectories based on lane change scene data collected by the autonomous driving vehicle, and obtain trajectory expression based on the parameters;

[0078] Lane change trajectory segmentation module: used to segment each lane change trajectory into N equal segments, each segment is a sub-trajectory;

[0079] Sub-trajectory parameter feature conversion module: used to extract trajectory parameters from each sub-trajectory and perform feature conversion on the trajectory parameters to obtain the parameter feature of the nth sub-trajectory in the mth trajectory:

[0080] TS' mn =[Angel' mn ,Speed' mn , Position_X' mn , Position_Y m ' n ],

[0081] Among them, Angel` mn Represents the steering wheel angle of the nth sub-track in the mth track, Speed` mn Represents the speed of the nth sub-track in the mth track, Position_X` mn Indicates the lateral distance span of the nth sub-track in the mth track, Position_Y` mnrepresents the longitudinal distance span of the nth sub-track in the mth track; the data after feature conversion are aggregated to obtain the lane change track data set;

[0082] Trajectory clustering module: used to cluster the lane change trajectory data set using the k-means algorithm, using the cluster center to reflect the typical trajectory characteristics, thereby completing the extraction and determination of the typical lane change trajectory.

[0083] Furthermore, the sub-trajectory parameter feature conversion module further includes the following units: Angel` mn Conversion unit, Speed` mn Conversion unit, Position_X` mn and Position_Y` mn ,in:

[0084] Angel` mn Conversion unit for use with

[0085]

[0086] To achieve the conversion, express

[0087] <Angel m,(n-1)j / N+1 , Angel m,(n-1)j / N+2 ,......,Angel m,nj / N >

[0088] The maximum value in ;

[0089] express

[0090] <Angel m,(n-1)j / N+1 , Angel m,(n-1)j / N+2 ,......,Angel m,nj / N >

[0091] The minimum value in ;

[0092] Speed mn Conversion unit for use with

[0093]

[0094] Implements the conversion, where:

[0095] express

[0096] <Speed m,(n-1)j / N+1 ,Speed m,(n-1)j / N+2 ,......,Speed m,nj / N >

[0097] The maximum value in ;

[0098] express

[0099] <Speed m,(n-1)j / N+1 ,Speed m,(n-1)j / N+2 ,......,Speed m,nj / N >

[0100] The minimum value in ;

[0101] express

[0102] <Speed m,(n-1)j / N+1 ,Speed m,(n-1)j / N+2 ,......,Speed m,nj / N > the mean value;

[0103] Position_X` mn and Position_Y` mn Conversion unit for

[0104]

[0105]

[0106] Implements the conversion, where:

[0107] express

[0108] <Position_X m,(n-1)j / N+1 , Position_X m,(n-1)j / N+2 ,......,Position_X m,nj / N The maximum value among >,

[0109] express

[0110] <Position_X m,(n-1)j / N+1 , Position_X m,(n-1)j / N+2 ,......,Position_X m,nj / N >minimum value among

[0111] express

[0112] <Position_X m,(n-1)j / N+1 , Position_X m,(n-1)j / N+2 ,......,Position_X m,nj / N The maximum value among >,

[0113] express

[0114] <Position_X m,(n-1)j / N+1 , Position_X m,(n-1)j / N+2 ,......,Position_X m,nj / N >the minimum value among .

[0115] Furthermore, the trajectory clustering module further includes the following units:

[0116] Extract centroid vector unit: used to randomly select k samples from the feature-converted trajectory feature lane change trajectory dataset as the initial k centroid vectors:

[0117] {μ1, μ2, ..., μ k};

[0118] Cluster convergence unit: used to repeat the following process until convergence:

[0119] A) Divide the clusters into Cj, j = 1, 2, 3, ..., k, k is the total number of clusters;

[0120] B) For each lane change trajectory TRi to each center of mass

[0121] μ j , j = 1, 2, 3, ..., k,

[0122] Where i = 1, 2, 3, ..., M, and then further obtain the distance d between each centroid ij , mark the TRi category as Cj corresponding to the minimum dij;

[0123] For j = 1, 2, 3, ..., k, for C j Recalculate the new center of mass based on all lane change trajectory data;

[0124]

[0125] If all k centroid vectors have not changed, it means that the data convergence is completed, otherwise continue;

[0126] A typical lane change trajectory feature acquisition unit is used for C = {C1, C2, ..., Ck} and the corresponding centroids {μ1, μ2, ..., μ k}, the centroid is the typical lane change trajectory feature extracted by the present invention.

[0127] Furthermore, the cluster convergence unit further includes a subunit for obtaining the distance dij between centroids, and the subunit includes:

[0128] Feature weight definition subunit: The weight vector used to define the features of steering wheel angle, vehicle speed, vehicle longitudinal position, and vehicle lateral position is:

[0129] W={W Angel , W Speed , W Position_X , W Position_Y}

[0130] Among them: the weight value of each feature is greater than or equal to zero and satisfies

[0131] W Angel +W Speed +W Position_X +W Position_Y =1,

[0132] Distance calculation subunit: used to calculate the distance between two trajectories. Assume that the two trajectories to be calculated are TR1 and TR2:

[0133] TR1=[TS 11 , TS 12 , ......., TS 1N ], TR2 = [TS 21 , TS 22 , ......., TS 2N ],

[0134] The distance between the two trajectories is:

[0135]

[0136] in:

[0137]

[0138] The present invention also claims protection for a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements any one of the typical lane change trajectory determination methods based on lane change scene data described in the present invention.

[0139] The present invention first extracts data parameters that have a relevant impact on the lane changing trajectory from a large amount of lane changing scene data as lane changing trajectory features, including feature information such as vehicle speed, turning angle, and position features, and segments a complete trajectory considering the time characteristics of the lane changing, so as to highlight the local characteristics of the lane changing trajectory, which can meet the development needs of the automatic lane changing algorithm.

[0140] The present invention utilizes local feature conversion and feature weighted distance measurement, and finally realizes the extraction of typical lane change trajectories through the k-means clustering method, thereby improving the scientific rationality of the generation of typical lane change trajectories, and making the trajectories obtained based on the method of the present invention more reliable when applied to automatic steering control technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0141] Figure 1 This is a typical lane change trajectory determination process based on lane change scenario data described in the present invention.

[0142] Figure 2 It is a schematic diagram of lane change trajectory and parameter expression.

[0143] Figure 3 In the present invention, a schematic diagram of lane change trajectory segmentation. Implementation

[0144] A typical lane-changing trajectory determination method based on lane-changing scene data claimed in the present invention is illustrated below with reference to the accompanying drawings.

[0145] Implementation method 1. See Figure 1 To illustrate this embodiment, a typical lane change trajectory determination method based on lane change scene data described in this embodiment includes the following steps:

[0146] The steps of trajectory parameter extraction and trajectory expression are as follows: extracting parameters representing the lane change trajectory based on the lane change scene data collected by the autonomous driving vehicle, and obtaining the trajectory expression based on the parameters;

[0147] Lane change trajectory segmentation step: segment each lane change trajectory into N equal segments, each segment is a sub-trajectory;

[0148] Sub-trajectory parameter feature conversion step: extract trajectory parameters from each sub-trajectory, and perform feature conversion on the trajectory parameters to obtain the parameter feature of the nth sub-trajectory in the mth trajectory:

[0149] TS' mn =[Angel' mn ,Speed' mn , Position_X' mn , Position_Y m ' n ],

[0150] Among them, Angel` mn Represents the steering wheel angle of the nth sub-track in the mth track, Speed` mn Represents the speed of the nth sub-track in the mth track, Position_X` mnIndicates the lateral distance span of the nth sub-track in the mth track, Position_Y` mn represents the longitudinal distance span of the nth sub-track in the mth track; the data after feature conversion are aggregated to obtain the lane change track data set;

[0151] Trajectory clustering step: clustering the lane change trajectory data set using the k-means algorithm, using the cluster center to reflect the typical trajectory features, thereby completing the extraction and determination of the typical lane change trajectory.

[0152] The scheme described in this embodiment includes trajectory parameter extraction and trajectory expression, lane change trajectory segmentation, sub-trajectory parameter feature conversion, and trajectory clustering process to obtain a typical lane change trajectory.

[0153] In this embodiment, the lane change trajectory is divided into N equal segments, see Figure 3 As shown, this kind of trajectory segmentation can effectively highlight the local features of the lane change trajectory without losing the global features of the trajectory. Therefore, all lane change trajectory data are segmented into specific N equal segments to obtain the feature data of each segment trajectory.

[0154] Implementation method 2. This implementation method is a further limitation of the typical lane change trajectory determination method based on lane change scene data described in implementation method 1. In this implementation method, the parameters characterizing the lane change trajectory include: steering wheel angle Angle, vehicle speed Speed, vehicle longitudinal position Position_X, and vehicle lateral position Position_Y.

[0155] This embodiment explains the lane change trajectory parameters. The lane change trajectory parameters of this embodiment take into account four parameters, see Figure 2 As shown, among which: the longitudinal position of the vehicle refers to the displacement of the current position of the vehicle along the lane direction relative to the starting time of lane change, and the lateral position of the vehicle refers to the displacement of the current position of the vehicle perpendicular to the lane direction relative to the starting time of lane change.

[0156] Implementation method three. This implementation method is a further limitation of the typical lane change trajectory determination method based on lane change scene data described in implementation method one. In this implementation method, the trajectory based on the lane change scene data is expressed as: TD = {TR1, TR2, ..., TRM}, M is the total amount of all lane change trajectory data, TR1 represents the first lane change trajectory, TR2 represents the second lane change trajectory, ..., TRM represents the Mth lane change trajectory.

[0157] In this implementation, the lane change trajectory is expressed as:

[0158] Trajectory=[Cond1, Cond2,...,Condj],

[0159] Among them, j represents the end time of the lane change trajectory, and Condi represents the trajectory characteristics at time i

[0160] Condi=<Anglei,Speedi,Position_Xi,Position_Yi> ,

[0161] Where Anglei represents the steering wheel angle at time i, Speedi represents the vehicle speed at time i, Position_Xi represents the lateral position of the vehicle at time i, and Position_Yi represents the longitudinal position of the vehicle at time i.

[0162] In this embodiment, j represents the end time of the lane change trajectory, which actually represents the length of the lane change trajectory.

[0163] In this implementation, a complete lane change trajectory is described in the form of a mathematical expression to facilitate subsequent data processing.

[0164] Embodiment 4. This embodiment is a further limitation of the typical lane change trajectory determination method based on lane change scene data described in Embodiment 1. In this embodiment, the lane change trajectory dataset DR is:

[0165]

[0166] Among them: TS 11 is the trajectory of the first trajectory segment in the first lane change trajectory, and so on;

[0167] The trajectory of the nth segment in the mth lane change trajectory is expressed as:

[0168] TS mn =[Cond m,(n-1)j / N+1 ,Cond m,(n-1)j / N+2 ,......,Cond m,nj / N ],

[0169] Among them, 1≤n≤N, 1≤m≤M, among which Cond m,(n-1)j / N+1 is the feature at time (n-1)j / N+1 in the n-th segment of the m-th lane change trajectory.

[0170] Implementation 5. This implementation is a further limitation of the typical lane change trajectory determination method based on lane change scene data described in Implementation 1. In this implementation, the Angel ` mn is based on

[0171]

[0172] Obtained, of which express

[0173] <Angel m,(n-1)j / N+1 , Angel m,(n-1)j / N+2 ,......,Angel m,nj / N >

[0174] The maximum value in ;

[0175] express

[0176] <Angel m,(n-1)j / N+1 , Angel m,(n-1)j / N+2 ,......,Angel m,nj / N >

[0177] The minimum value in ;

[0178] The Speed` mn is through Obtained, of which:

[0179] express

[0180] <Speed m,(n-1)j / N+1 ,Speed m,(n-1)j / N+2 ,......,Speed m,nj / N >

[0181] The maximum value in ;

[0182] express

[0183] <Speed m,(n-1)j / N+1 ,Speed m,(n-1)j / N+2 ,......,Speed m,nj / N >

[0184] The minimum value in ;

[0185] express

[0186] <Speed m,(n-1)j / N+1 ,Speed m,(n-1)j / N+2 ,......,Speed m,nj / N >

[0187] The mean of

[0188] Position_X` mn and Position_Y` mn is through

[0189]

[0190]

[0191] Obtained, of which:

[0192] express

[0193] <Position_X m,(n-1)j / N+1 , Position_X m,(n-1)j / N+2 ,......,Position_X m,nj / N >

[0194] The maximum value in

[0195] express

[0196] <Position_X m,(n-1)j / N+1 , Position_X m,(n-1)j / N+2 ,......,Position_X m,nj / N >

[0197] The minimum value in ;

[0198] express

[0199] <Position_X m,(n-1)j / N+1 , Position_X m,(n-1)j / N+2 ,......,Position_X m,nj / N >

[0200] The maximum value in

[0201] express

[0202] <Position_X m,(n-1)j / N+1 , Position_X m,(n-1)j / N+2 ,......,Position_X m,nj / N >

[0203] The minimum value in .

[0204] Embodiment 6. This embodiment further limits the method for determining a typical lane-changing trajectory based on lane-changing scene data described in Embodiment 1. In this embodiment, the method for clustering the lane-changing trajectory dataset using the k-means algorithm and using the cluster center to reflect the typical trajectory characteristics is as follows: randomly selecting k samples from the feature-converted trajectory feature lane-changing trajectory dataset as the initial k centroid vectors:

[0205] {μ1, μ2, ..., μ k};

[0206] Repeat the following process until convergence:

[0207] A) Divide the clusters into Cj, j = 1, 2, 3, ..., k, k is the total number of clusters; j is the public intermediate variable,

[0208] B) For each lane change trajectory TRi to each center of mass

[0209] μ j , j = 1, 2, 3, ..., k,

[0210] Where i = 1, 2, 3, ..., M, and then further obtain the distance d between each centroid ij , the TRi class

[0211] It is marked as Cj corresponding to the minimum dij;

[0212] For j = 1, 2, 3, ..., k, for C j Recalculate the new center of mass based on all lane change trajectory data;

[0213]

[0214] If all k centroid vectors have not changed, indicating that the data has converged, go to the next step, otherwise return to step A) to continue;

[0215] Get C = {C1, C2, ..., Ck} and the corresponding centroid {μ1, μ2, ..., μ k}, the centroid is the typical lane change trajectory feature extracted by the present invention.

[0216] Embodiment 7. This embodiment further limits the typical lane change trajectory determination method based on lane change scene data described in Embodiment 1. In this embodiment, the distance d between the centroids is ij It is obtained by the following method:

[0217] a) First, define the weight vector of the steering wheel angle, vehicle speed, vehicle longitudinal position, and vehicle lateral position features as:

[0218] W={W Angel , W Speed , W Position_X , W Position_Y}

[0219] Among them: the weight value of each feature is greater than or equal to zero and satisfies

[0220] W Angel +W Speed +W Position_X +W Position_Y =1,

[0221] b) Calculate the distance between two trajectories, assuming that the two trajectories to be calculated are TR1 and TR2:

[0222] TR1=[TS 11 , TS 12 , ......., TS 1N ], TR2 = [TS 21 , TS 22 , ......., TS 2N ],

[0223] The distance between the two trajectories is:

[0224]

[0225] in:

[0226]

[0227] The above-mentioned embodiments are illustrative examples of the technical solutions claimed for protection by the present invention and do not limit the protection scope of the present invention. Reasonable combinations of the technical features recorded in the above-mentioned embodiments also belong to the technical solutions claimed for protection by the present invention.

Claims

1. A typical lane-changing trajectory determination method based on lane-changing scene data, characterized in that: The method comprises: The steps of trajectory parameter extraction and trajectory expression are as follows: extracting parameters representing the lane change trajectory based on the lane change scene data collected by the autonomous driving vehicle, and obtaining the trajectory expression based on the parameters; Lane change trajectory segmentation step: segment each lane change trajectory into N equal segments, each segment is a sub-trajectory; Sub-trajectory parameter feature conversion step: extract trajectory parameters from each sub-trajectory, and perform feature conversion on the trajectory parameters to obtain the parameter feature of the nth sub-trajectory in the mth trajectory: TS' mn =[Angel' mn ,Speed' mn ,Position_X' mn ,Position_Y m ' n ], Among them, Angel` mn Represents the steering wheel angle of the nth sub-track in the mth track, Speed` mn Represents the speed of the nth sub-track in the mth track, Position_X` mn Indicates the lateral distance span of the nth sub-track in the mth track, Position_Y` mn represents the longitudinal distance span of the nth sub-track in the mth track; the data after feature conversion are aggregated to obtain the lane change track data set; Trajectory clustering step: cluster the lane change trajectory data set using the k-means algorithm, and use the cluster center to reflect the typical trajectory characteristics, thereby completing the extraction and determination of the typical lane change trajectory; the Angel` mn is based on Obtained, of which express <Angel m,(n-1)j / N+1 ,Angel m,(n-1)j / N+2 ,......,Angel m,nj / N > The maximum value in ; express <Angel m,(n-1)j / N+1 ,Angel m,(n-1)j / N+2 ,......,Angel m,nj / N > The minimum value in ; The Speed` mn is through Obtained, of which: express <Speed m,(n-1)j / N+1 ,Speed m,(n-1)j / N+2 ,......,Speed m,nj / N > The maximum value in ; express <Speed m,(n-1)j / N+1 ,Speed m,(n-1)j / N+2 ,......,Speed m,nj / N > The minimum value in ; express <Speed m,(n-1)j / N+1 ,Speed m,(n-1)j / N+2 ,......,Speed m,nj / N > the mean value; Position_X` mn and Position_Y` mn is through Obtained, of which: express <Position_X m,(n-1)j / N+1 ,Position_X m,(n-1)j / N+2 ,......,Position_X m,nj / N > The maximum value in express <Position_X m,(n-1)j / N+1 ,Position_X m,(n-1)j / N+2 ,......,Position_X m,nj / N > The minimum value in ; express <Position_X m,(n-1)j / N+1 ,Position_X m,(n-1)j / N+2 ,......,Position_X m,nj / N > The maximum value in express <Position_X m,(n-1)j / N+1 ,Position_X m,(n-1)j / N+2 ,......,Position_X m,nj / N > The minimum value in .

2. The typical lane-changing trajectory determination method based on lane-changing scene data according to claim 1, characterized in that: The parameters characterizing the lane change trajectory include four parameters: steering wheel angle Angle, vehicle speed Speed, vehicle longitudinal position Position_X, and vehicle lateral position Position_Y.

3. The typical lane-changing trajectory determination method based on lane-changing scene data according to claim 1, characterized in that: The trajectory-based expression is: TD = {TR1, TR2, ..., TRM}, M is the total amount of all lane change trajectory data, TR1 represents the first lane change trajectory, TR2 represents the second lane change trajectory, ..., TRM represents the Mth lane change trajectory.

4. The method for determining a typical lane-changing trajectory based on lane-changing scene data according to claim 1, characterized in that: The lane change trajectory dataset DR is: Among them: TS 11 is the trajectory of the first trajectory segment in the first lane change trajectory, and so on; The trajectory of the nth segment in the mth lane change trajectory is expressed as: TS mn =[Cond m,(n-1)j / N+1 ,Cond m,(n-1)j / N+2 ,......,Cond m,nj / N ], Among them, 1≤n≤N, 1≤m≤M, among which Cond m,(n-1)j / N+1 is the feature at time (n-1)j / N+1 in the n-th segment of the m-th lane change trajectory.

5. The method for determining a typical lane-changing trajectory based on lane-changing scene data according to claim 1, characterized in that: The method of clustering the lane change trajectory dataset using the k-means algorithm and using the cluster center to reflect the typical trajectory characteristics is as follows: randomly selecting k samples from the feature-converted trajectory feature lane change trajectory dataset as the initial k centroid vectors: {μ1,μ2,...,μ k }; Repeat the following process until convergence: A) Divide the clusters into Cj, j = 1, 2, 3, ..., k, k is the total number of clusters; B) For each lane change trajectory TRi to each center of mass m j ,j=1,2,3,...,k, Where i = 1, 2, 3, ..., M, and then further obtain the distance d between each centroid ij , mark the TRi category as Cj corresponding to the minimum dij; For j = 1, 2, 3, ..., k, for C j Recalculate the new center of mass based on all lane change trajectory data; If all k centroid vectors have not changed, indicating that the data has converged, go to the next step, otherwise return to step A) to continue; Get C = {C1, C2, ..., Ck} and the corresponding centroid {μ1, μ2, ..., μ k }, the centroid is the typical lane change trajectory feature extracted by the present invention.

6. The method for determining a typical lane-changing trajectory based on lane-changing scene data according to claim 5, characterized in that: The distance d between the centroids ij It is obtained by the following method: a) First, define the weight vector of the steering wheel angle, vehicle speed, vehicle longitudinal position, and vehicle lateral position features as: In={In Angel ,IN Speed ,IN Position_X ,IN Position_Y } Among them: the weight value of each feature is greater than or equal to zero and satisfies IN Angel +W Speed +W Position_X +W Position_Y =1, b) Calculate the distance between two trajectories, assuming that the two trajectories to be calculated are TR1 and TR2: TR1=[TS 11 ,TS 12 ,.......,TS 1N ],TR2=[TS 21 ,TS 22 ,.......,TS 2N ], The distance between the two trajectories is: in:

7. A typical lane-changing trajectory determination device based on lane-changing scene data, characterized in that: The device comprises the following modules: Trajectory parameter extraction and trajectory expression module: used to extract parameters representing lane change trajectories based on lane change scene data collected by the autonomous driving vehicle, and obtain trajectory expression based on the parameters; Lane change trajectory segmentation module: used to segment each lane change trajectory into N equal segments, each segment is a sub-trajectory; Sub-trajectory parameter feature conversion module: used to extract trajectory parameters from each sub-trajectory and perform feature conversion on the trajectory parameters to obtain the parameter feature of the nth sub-trajectory in the mth trajectory: TS' mn =[Angel' mn ,Speed' mn ,Position_X' mn ,Position_Y m ' n ], Among them, Angel` mn Represents the steering wheel angle of the nth sub-track in the mth track, Speed` mn Represents the speed of the nth sub-track in the mth track, Position_X` mn Indicates the lateral distance span of the nth sub-track in the mth track, Position_Y` mn represents the longitudinal distance span of the nth sub-track in the mth track; the data after feature conversion are aggregated to obtain the lane change track data set; Trajectory clustering module: used to cluster the lane change trajectory data set using the k-means algorithm, using the cluster center to reflect the typical trajectory characteristics, thereby completing the extraction and determination of the typical lane change trajectory. The sub-trajectory parameter feature conversion module further includes the following units: Angel` mn Conversion unit, Speed` mn Conversion unit, Position_X` mn and Position_Y` mn Conversion unit, where: Angel` mn Conversion unit for use with To achieve the conversion, express <Angel m,(n-1)j / N+1 ,Angel m,(n-1)j / N+2 ,......,Angel m,nj / N > The maximum value in ; express <Angel m,(n-1)j / N+1 ,Angel m,(n-1)j / N+2 ,......,Angel m,nj / N > The minimum value in ; Speed mn Conversion unit for use with Implements the conversion, where: express <Speed m,(n-1)j / N+1 ,Speed m,(n-1)j / N+2 ,......,Speed m,nj / N > The maximum value in ; express <Speed m,(n-1)j / N+1 ,Speed m,(n-1)j / N+2 ,......,Speed m,nj / N > The minimum value in ; express <Speed m,(n-1)j / N+1 ,Speed m,(n-1)j / N+2 ,......,Speed m,nj / N > the mean value; Position_X` mn and Position_Y` mn Conversion unit for Implements the conversion, where: express <Position_X m,(n-1)j / N+1 ,Position_X m,(n-1)j / N+2 ,......,Position_X m,nj / N > The maximum value in express <Position_X m,(n-1)j / N+1 ,Position_X m,(n-1)j / N+2 ,......,Position_X m,nj / N > The minimum value in ; express <Position_X m,(n-1)j / N+1 ,Position_X m,(n-1)j / N+2 ,......,Position_X m,nj / N > The maximum value in express <Position_X m,(n-1)j / N+1 ,Position_X m,(n-1)j / N+2 ,......,Position_X m,nj / N > The minimum value in .

8. The typical lane-changing trajectory determination device based on lane-changing scene data according to claim 7, characterized in that: The trajectory clustering module further includes the following units: Extract centroid vector unit: used to randomly select k samples from the feature-converted trajectory feature lane change trajectory dataset as the initial k centroid vectors: {μ1,μ2,...,μ k }; Cluster convergence unit: used to repeat the following process until convergence: A) Divide the clusters into Cj, j = 1, 2, 3, ..., k, k is the total number of clusters; B) For each lane change trajectory TRi to each center of mass m j ,j=1,2,3,...,k, Where i = 1, 2, 3, ..., M, and then further obtain the distance d between each centroid ij , mark the TRi category as Cj corresponding to the minimum dij; For j = 1, 2, 3, ..., k, for C j Recalculate the new center of mass based on all lane change trajectory data; If all k centroid vectors have not changed, it means that the data convergence is completed, otherwise continue; A typical lane change trajectory feature acquisition unit is used for C = {C1, C2, ..., Ck} and the corresponding centroids {μ1, μ2, ..., μ k }, the centroid is the typical lane change trajectory feature extracted by the present invention.

9. The typical lane-changing trajectory determination device based on lane-changing scene data according to claim 8, characterized in that: The cluster convergence unit further includes the distance d between the centroids ij Obtain a subunit, the subunit comprising: Feature weight definition subunit: The weight vector used to define the features of steering wheel angle, vehicle speed, vehicle longitudinal position, and vehicle lateral position is: In={In Angel ,IN Speed ,IN Position_X ,IN Position_Y } Among them: the weight value of each feature is greater than or equal to zero and satisfies IN Angel +W Speed +W Position_X +W Position_Y =1, Distance calculation subunit: used to calculate the distance between two trajectories. Assume that the two trajectories to be calculated are TR1 and TR2: TR1=[TS 11 ,TS 12 ,.......,TS 1N ],TR2=[TS 21 ,TS 22 ,.......,TS 2N ], The distance between the two trajectories is: in:

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method for determining a typical lane change trajectory based on lane change scene data as described in any one of claims 1 to 6 is implemented.

Images