A Heuristic Rule-Based Method for Vehicle Lane Changing Trajectory Recognition

Abstract

This invention discloses a vehicle lane-changing trajectory recognition method based on heuristic rules, comprising the following steps: sorting the original trajectory data by global time; determining whether the original dataset records observation data from different time periods separately; classifying and saving the original dataset according to different recording time periods; smoothing the trajectory coordinate information in the dataset; using heuristic trajectory recognition to obtain the lane-changing trajectory and obtaining the starting and ending points of the lane-changing action; performing the final five-vehicle data matching of the lane-changing model to complete the calculation of higher-order data. Based on the data characteristics of common vehicle trajectory datasets, this invention details various aspects and rules for extracting vehicle lane-changing trajectories, thereby providing data support for subsequent model algorithms based on lane-changing behavior details and laying the foundation for the innovative development of lane-changing models.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of traffic trajectory data processing, and particularly relates to a vehicle lane-changing trajectory recognition method based on heuristic rules. BACKGROUND

[0002] Lane-changing behavior of vehicles is one of the main driving behaviors that affect traffic flow characteristics. In the past three decades, lane-changing models have also been a hot issue in various traffic studies. However, most of the studies focus on the modeling of the lane-changing decision-making process, and often ignore the specific details in the lane-changing behavior. Many current numerical simulation experiments on traffic flow and vehicle lane-changing regard lane changes as instantaneous events or events with constant intervals. This simplification is due to the lack of rich and accurate vehicle trajectories. With the introduction of artificial intelligence technologies such as computer vision into the field of transportation, video-based vehicle trajectory monitoring technology has overcome this difficulty. Typical representatives in this field are the NGSIM (Next Generation SIMulation) and HighD (Highway Drone Dataset) datasets, which contain rich information about individual driver behavior and can be used as learning data for data-driven lane-changing models. However, tests have shown that if the original data is used directly without proper examination, biased conclusions may be drawn. Therefore, how to extract the required lane-changing data from complex and large original datasets becomes a key issue.

[0003] Generally speaking, there are two main problems to be solved. First, non-anomalous lane-changing trajectory data needs to be extracted from a large amount of sampled data, and the trajectory data needs to be made to meet the research needs (such as only free lane-changing data is needed). Second, the starting point and the ending point of the vehicle lane-changing behavior need to be defined based on the filtered trajectory. The present application develops a set of heuristic rules to filter large amounts of trajectory data sets, logically combs and elaborates the processing procedures such as preprocessing of original trajectory data, trajectory extraction, and data twin, and typical trajectory data sets can be processed according to this idea. SUMMARY

[0004] The technical problem to be solved by the present application is to solve the deficiencies of the prior art, and to provide a vehicle lane-changing trajectory recognition method based on heuristic rules. The vehicle lane-changing trajectory recognition method based on heuristic rules first reasonably preprocesses the original data to obtain data with smooth position, speed, and acceleration changes, and then identifies the lane-changing trajectory based on the data. The trajectory data with normal lane-changing duration and appropriate lane-changing distance is selected through rules. The starting point and the ending point of the selected data are divided to obtain lane-changing trajectory data that actually contains most of the effective information. Finally, the data is further calculated and processed, and the vehicle feature data in the lane-changing trajectory data is extracted, laying a foundation for subsequent research on data-driven lane-changing models.

[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0006] A method for recognizing vehicle lane-changing trajectories based on heuristic rules includes the following steps.

[0007] Step 1: Data Segmentation and Recording Judgment: Sort the original trajectory data by global time as the keyword and determine whether the original trajectory data records data observed in different time periods separately. If the same global time in the original trajectory data has a unique shooting frame number, it is determined that the original trajectory data records data observed in different time periods separately, and proceed directly to Step 3; otherwise, it is determined that the original trajectory data does not record data observed in different time periods separately, and proceed to Step 2.

[0008] Step 2, Data Segmentation and Recording: When the global time corresponds to a shooting frame number, the original trajectory data is divided into observation data of a time periods and recorded separately, so that the same global time has a unique shooting frame number.

[0009] Step 3, Data Smoothing: First, sort the observation data for each time period that was recorded separately, using vehicle number n as the primary keyword and the sequence number i of the captured frame as the secondary keyword; then, smooth the location data of the observation data for each time period after sorting, to obtain the smoothed observation data of vehicle number n for each time period.

[0010] Step 4, Lane Change Trajectory Recognition: For the smoothed observation data of each vehicle number n in Step 3, lane change trajectory recognition is performed based on heuristic rules, specifically including the following steps:

[0011] Step 4-1, Lane Change Trajectory Judgment: Preliminarily determine whether the current trajectory is a lane change trajectory by observing the changes in the lane number column in a set of smoothed observation data; if the number of lane number changes in a set of smoothed observation data exceeds one, it is considered as one lane change. Filter out trajectories without lane changes and those with more than 2 lane changes to obtain lane change trajectory data with one lane change.

[0012] Step 4-2: Determine the lane change midpoint: Based on the lane change trajectory data obtained in Step 4-1, determine the lane change midpoint according to the lane line position distribution of the corresponding dataset collection road segment; the set of trajectory frame numbers before vehicle n crosses the lane change midpoint is called... The set of trajectory frame numbers of vehicle n after it crosses the lane change midpoint is called the

[0013] Step 4-3: Maintain the integrity of the lane change trajectory: Filter out vehicle trajectories whose trajectory data duration before or after the lane change is less than the set time retention threshold to ensure the integrity of the obtained lane change trajectory.

[0014] Step 4-4, determining the lane-changing starting point and the ending point, specifically:

[0015] A, starting point: starting point i q Located in the data set , the starting point is the last trajectory frame number i when the vehicle body is about to contact the lane line * Or the last trajectory frame number i' when the angle between the vehicle body and the lane line is less than the set angle threshold a.

[0016] B, ending point: ending point i z Located in the data set , the ending point is the minimum trajectory frame number i when the vehicle body approaches the lane line ** Or the minimum trajectory frame number i" when the angle between the vehicle body and the lane line is less than the set angle threshold a.

[0017] In step 4-4, the discriminant of the lane-changing starting point i q is:

[0018]

[0019] Where:

[0020]

[0021] And

[0022]

[0023]

[0024] And In the formula, is the set of trajectory frame numbers when the vehicle numbered n is straight.

[0025] is the angle between the vehicle body and the lane line when the vehicle numbered n is at the i1th trajectory shooting frame.

[0026] is the set of trajectory frame numbers in the trajectory of the vehicle numbered n when the vehicle body does not contact the lane line.

[0027] is the lateral coordinate of the vehicle numbered n at the i2th trajectory shooting frame.

[0028] X MidLin据 is the lateral coordinate of the lane line crossed by the lane-changing.

[0029] W n The vehicle width of the vehicle numbered n.

[0030] In step 4-4, The calculation formula is:

[0031] A, when the lane changing direction is from left to right, specifically:

[0032]

[0033] B, when the lane changing direction is from left to right, specifically:

[0034]

[0035] In the formula, and are the lateral coordinate and longitudinal coordinate of the vehicle numbered n at the i1th trajectory shooting frame, respectively.

[0036] and are the lateral coordinate and longitudinal coordinate of the vehicle numbered n at the i1+1th trajectory shooting frame, respectively. In step 4-4, the termination point i z The discriminant of the end point i

[0037]

[0038] Among them:

[0039]

[0040] And

[0041]

[0042]

[0043] And

[0044] In the formula, is the angle between the vehicle body and the lane line of the vehicle numbered n at the i3th trajectory shooting frame.

[0045] is the lateral coordinate of the vehicle numbered n at the i4th trajectory shooting frame.

[0046] In step 4, the angle threshold α needs to be set according to the average speed of the original trajectory data collection section, and the angle threshold α is proportional to the average speed of the collection section.

[0047] When the medium-high speed environment is in a medium-high speed environment, the angle threshold α is 1 degree.

[0048] In step 4-3, the calculation formula of the time keeping threshold ΔT is set as:

[0049]

[0050] In the formula, The trajectory shooting frame number is represented as

[0051] It also includes step 5, trajectory data expansion: trajectory data expansion includes primary expansion and secondary expansion, which is specifically:

[0052] A, primary expansion: for the lane change trajectory data of vehicle number n obtained in step 4-1, by matching the keywords of the front and rear vehicle numbers in the current lane before lane changing, the lane change trajectory data contains the trajectory shooting frame data of the current frame of the vehicle number n, the lead vehicle and the following vehicle.

[0053] B, secondary expansion: after the primary expansion is completed, by matching the keywords of the front and rear vehicle numbers in the target lane after lane changing, the lane change trajectory data contains the trajectory shooting frame data of the current frame of the vehicle number n, the lead vehicle before lane changing, the following vehicle before lane changing, the lead vehicle after lane changing, and the following vehicle after lane changing.

[0054] It also includes step 6, calculating high-order data collision time TTC, and the specific calculation formula is:

[0055]

[0056]

[0057]

[0058]

[0059] In the formula,

[0060] And respectively represent the collision time of the vehicle with number n before and after lane changing with the lead vehicle;

[0061] And respectively represent the collision time of the vehicle with number n before and after lane changing with the following vehicle;

[0062] And respectively represent the longitudinal coordinates of the lead vehicle in frame number i before and after lane changing of the vehicle with number n;

[0063] And ​respectively represent the longitudinal coordinates of the following vehicle before and after the lane change of the vehicle numbered n at the frame with serial number i;

[0064] respectively represent the longitudinal coordinates of the vehicle numbered n at the frame with serial number i;

[0065] and respectively represent the longitudinal velocities of the pilot vehicle before and after the lane change of the vehicle numbered n at the frame with serial number i;

[0066] and respectively represent the longitudinal velocities of the following vehicle before and after the lane change of the vehicle numbered n at the frame with serial number i;

[0067] respectively represent the longitudinal velocities of the vehicle numbered n at the frame with serial number i.

[0068] Further comprising the step 7, calculating the normalized fuel consumption rate NFR, and the specific calculation formula is:

[0069] NFR = 1.71 · VSP 0.42

[0070] wherein:

[0071]

[0072] In the formula, and respectively represent the lateral velocity and the longitudinal velocity of the vehicle numbered n at the frame with serial number i.

[0073] and respectively represent the lateral acceleration and the longitudinal acceleration of the vehicle numbered n at the frame with serial number i.

[0074] The present application has the following beneficial effects:

[0075] 1. The present application sorts and combs based on common keywords in the vehicle trajectory data set, smooths the vehicle trajectory, filters abnormal or incomplete vehicle trajectory data, and then filters out high-quality lane change trajectory data using a complete set of heuristic rules, and gives the starting point and the ending point of the lane change behavior, and according to this, obtains various high-order data that may be needed for vehicle lane change model calibration.

[0076] 2. The present application sorts out the processing flow of the typical trajectory data set, details the rules for identifying and extracting the lane change trajectory, and also thinks about various details in the trajectory data processing, which provides support for future data-driven lane change model research. BRIEF DESCRIPTION OF DRAWINGS

[0077] Figure 1 is a flow chart of a vehicle lane-changing trajectory recognition method based on heuristic rules according to an embodiment of the present application.

[0078] Figure 2 is a schematic diagram of basic elements of a vehicle lane-changing model according to an embodiment of the present application.

[0079] Figure 3 is a schematic diagram of lane-changing trajectory recognition according to an embodiment of the present application.

[0080] Figure 4 is a schematic diagram of lane-changing trajectory smoothing results according to an embodiment of the present application.

[0081] Figure 5 is a schematic diagram of lane-changing trajectory recognition extraction results according to an embodiment of the present application. DETAILED DESCRIPTION

[0082] The present application will be further described below in conjunction with the accompanying drawings and specific preferred embodiments.

[0083] In the description of the present application, it should be understood that the terms "left side", "right side", "upper part", "lower part" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the device or element referred to must have a particular orientation, be constructed and operated in a particular orientation, and "first", "second" and the like do not represent the importance of the parts, and therefore cannot be understood as a limitation on the present application. The specific dimensions used in the embodiments are only for the purpose of illustrating the technical solutions and do not limit the protection scope of the present application.

[0084] As shown in Figure 1 , a vehicle lane-changing trajectory recognition method based on heuristic rules comprises the following steps.

[0085] Step 1, data segmentation record judgment

[0086] The original trajectory data is sorted with global time (i is the frame number, n is the vehicle number) as the key word, and it is judged whether the original trajectory data separates the data observed in different time periods; when there is only one frame number in the original trajectory data at the same global time, it is judged that the original trajectory data separates the data observed in different time periods, and directly enters step 3; otherwise, it is judged that the original trajectory data does not separate the data observed in different time periods, and enters step 2.

[0087] Specifically, if the original trajectory data is sorted by the keyword of global time, and there are two or more frame numbers with the same global time, it can be considered that the original data set directly stores data recorded in different time periods. As shown in Table 1, in the embodiment, there are two different frame numbers with the same global time, and the frame numbers are 89 and 2 respectively, so the original data set of the embodiment stores data recorded in different time periods.

[0088]

[0089] Table 1

[0090] Step 2, data segmentation recording

[0091] Since the data recorded in adjacent time periods often has data crossing near the boundary, resulting in chaotic trajectory data and unable to be analyzed and recognized, if the original data is not recorded in different time periods, the data needs to be classified and saved according to different time periods, and the classified data set recorded in different time periods is obtained. For example, when the global time corresponds to a shooting frame number, the original trajectory data is divided into a time period to record the observation data, so that the same global time has only one shooting frame number.

[0092] The above data segmentation observation time period saving method is preferably:

[0093] Since it is impossible to only use the global time to classify the crossing data at the boundary of adjacent time periods, the shooting frame number i is used as an auxiliary keyword, and the shooting frame number column of the trajectory data with the same global time is input into the unique function to remove duplicate elements. If the number of remaining elements exceeds one, it is determined that the data with a larger shooting frame number is the data of the previous time period, and the data with a smaller shooting frame number is the data of the next time period, and the number of times that the number of non-repeating frame numbers changes from 2 to 1 determines (equals) the number of different time periods in the observation of the original data set.

[0094] In the embodiment, the crossing data of different time periods is shown in Table 1, and the shooting frame numbers are 89 and 2 respectively. According to the rule, the data with a larger shooting frame number is determined as the data of the previous time period, and the data with a smaller shooting frame number is determined as the data of the next time period. Therefore, the data of vehicle number 13 and shooting frame number 89 is divided into the previous time period, and the data of vehicle number 1 and shooting frame number 2 is divided into the next time period.

[0095] Step 3, data smoothing

[0096] First, sort each time period observation data by vehicle number n as the primary key and frame number i as the secondary key; then, smooth the position data of each time period observation data after sorting to obtain the smoothed observation data of vehicle number n in each time period.

[0097] Routine sorting of the processed vehicle trajectory data (i.e., sorting by vehicle number n as the primary key and frame number i as the secondary key) ensures the orderliness of the data. Due to the limited accuracy of the camera, some abnormal data will inevitably appear in the measured trajectory data, so it is necessary to smooth the position data in the trajectory. The smoothing method includes average filtering, sliding median filtering, Savitzky-Golay filtering, etc., but it is not recommended to use model-based filters such as Kalman filter. Perform differential calculation on the smoothed position data to obtain the smoothed speed acceleration of vehicle number n at time i, where is the lateral position coordinate and longitudinal position coordinate of vehicle number n at time i, is the lateral speed and longitudinal speed of vehicle number n at time i, is the lateral acceleration and longitudinal acceleration of vehicle number n at time i.

[0098] In this embodiment, the smoothing method preferably uses the Savitzky-Golay filtering method, the core idea of which is to perform k-order polynomial fitting on the data points within a certain length window to obtain the fitted results. The calculation method of the filter value is:

[0099] Let the filter window be L = 2m + 1, and the time axis scale of the longitudinal position y coordinate of each measurement point in the trajectory data within the window can be regarded as τ = (-m, -m + 1, …, -1, 0, 1, …, m - 1, m). A k - 1 order polynomial is used to fit the data points within the window to obtain:

[0100] y = a0 + a1τ + a2τ 2 + … + a k-1 τ k-1

[0101] L such equations can be obtained from the trajectory data within the window, which can be converted into a k-element linear equation set. To have a solution, L should be greater than or equal to k, and generally L > k is selected. The fitting parameters A are determined by least squares fitting, and the smoothed position data of vehicle number n at time i is obtained as follows:

[0102]

[0103] which can be represented by matrix as​

[0104] Y (2m+1)×1 = T (2m+1)×k · A k×1 + E (2m+1)×1

[0105] The least square solution is obtained by optimizing the matrix a k×1 such that ||E (2m+1)×1 ||2 is minimized, and the least square solution can be obtained by matrix calculation = B

[0106]

[0107] The smoothed filter value of the longitudinal coordinate Y of the trajectory is obtained

[0108]

[0109] Wherein, B = T·(T T ·T) -1 ·T T Thus, the correction value of the longitudinal coordinate of the trajectory in the window is obtained The idea of calculating the filter value at the data edge is to expand the data according to the first and last values, and the calculation idea of the horizontal position x coordinate of the trajectory data is the same as above.

[0110] The above speed and acceleration are calculated as follows:

[0111]

[0112]

[0113]

[0114]

[0115] In the formula, ΔT' is the time interval of each frame when the data set is recorded.

[0116] and are the horizontal coordinate and the longitudinal coordinate of the vehicle numbered n at the i-th frame.

[0117] and are the horizontal speed and the longitudinal speed of the vehicle numbered n at the i-1-th frame.

[0118] This example takes a typical trajectory in ngsim as the processing object, first takes the Vehicle_ID (vehicle number) and Frame_ID (shooting frame number) in the ngsim data set as the sorting key to perform routine sorting, and the sorted data becomes the form in Table 2 below:

[0119]

[0120] Table 2

[0121] Then the Savitzky-Golay filtering method is used, and a third-order polynomial is used for fitting, and the window length is set to 21. Since the data amount of a trajectory is large, it is not suitable to directly display, so here the Figure 4 The change of the trajectory data after such smoothing is shown. It can be seen that compared with the original data, the coordinate curve of the smoothed trajectory is obviously smoother, and the unreasonable sudden displacement in the original trajectory is removed.

[0122] Step 4, lane change trajectory recognition

[0123] In this example, the angle threshold is set to 1 degree, and the lane width is set to 3.6576 meters according to the actual situation of the i-80 highway in the United States. The ngsim data set in this example contains the following information as shown in Table 3:

[0124]

[0125] Table 3

[0126] For each vehicle number n of the smoothed observation data in step 3, lane change trajectory recognition is performed based on heuristic rules, which includes the following steps.

[0127] Step 4-1, lane change trajectory judgment: the value change of the lane number column in a group of smoothed observation data is used to preliminarily judge whether the current trajectory is a lane change trajectory; when the number of lane number changes in a group of smoothed observation data exceeds once, it is considered as a lane change, and the trajectories without lane change and the trajectories with lane change more than 2 times are filtered out, and the lane change trajectory data with lane change once is obtained. Among them, the lane change more than 2 times is a continuous lane change vehicle, which may not follow the general mode and belongs to an abnormal situation, which is not discussed in this application.

[0128] In this embodiment, the lane number (LaneID) in the trajectory data is changed from 3 to 2, and there is no second lane change, so the trajectory is a lane change trajectory.

[0129] Step 4-2, determine the lane change midpoint: for the lane change trajectory data obtained in step 4-1, the lane change midpoint is determined according to the lane line position distribution of the corresponding data set collection road section; the trajectory frame number set of the vehicle numbered n before crossing the lane change midpoint is called The trajectory frame sequence number set of the vehicle numbered n after crossing the lane change midpoint is called

[0130] Due to the limited accuracy of the measured sensor, the lane number change point in the data of a trajectory may not be the actual lane change midpoint, so the actual lane change midpoint needs to be determined according to the specific lane line position distribution of the actual sampling road section of the data set, and the following discrimination is made:

[0131]

[0132] In the table, X MidLin据 is the lateral coordinate of the middle lane line of the lane change trajectory, is the trajectory lateral coordinate of the vehicle numbered n before crossing the lane change midpoint, is the trajectory lateral coordinate of the vehicle numbered n after crossing the lane change midpoint, is the trajectory frame sequence number set of the vehicle numbered n before crossing the lane change midpoint, is the trajectory frame sequence number set of the vehicle numbered n after crossing the lane change midpoint.

[0133] In this embodiment, the lane change trajectory is from right to left lane change, and the lane line (lateral coordinate X MidLin据 between lanes 2 and 3 is 3.6576x2=7.3152m) as the judgment limit, the trajectory before lane change and the trajectory after lane change are divided.

[0134] The determination of the lane change midpoint in the present application can obtain accurate vehicle lane change frames (the time when the vehicle center crosses the lane line) as the division points before and after the lane change process from the extraction process, which provides correct learning data for the related intelligent vehicle lane change before and after the behavior research, and lays a foundation for better intelligent lane change model that conforms to human behavior.

[0135] Further, the lane change trajectory with insufficient actual effective duration can be completely removed. In the traditional method, the lane number change frame is the lane change time, but if the measurement equipment or recognition algorithm is insufficient, it will cause a large deviation in the division of the lane change trajectory before and after, which will cause the method to believe that some trajectories after division meet the duration requirement (i.e. rule three), even if these trajectories do not meet the requirement, which causes data pollution.

[0136] Further, the lane change trajectory with insufficient actual effective duration can be completely removed. In the traditional method, the lane number change frame is the lane change time, but if the measurement equipment or recognition algorithm is insufficient, it will cause a large deviation in the division of the lane change trajectory before and after, which will cause the method to believe that some trajectories after division meet the duration requirement (i.e. rule three), even if these trajectories do not meet the requirement, which causes data pollution,

[0137] To provide data basis for the extraction of the starting and ending points of the lane-changing trajectory, and to ensure that the starting point of the lane-changing trajectory is absolutely not located in the rear part of the lane-changing trajectory, and the ending point is not located in the front part of the lane-changing trajectory.

[0138] Step 4-3, keep the lane-changing trajectory complete: filter out the vehicle trajectory whose duration before or after lane-changing is less than the set time keeping threshold, and ensure the completeness of the obtained lane-changing trajectory.

[0139] The calculation formula of the above set time keeping threshold ΔT is preferably:

[0140]

[0141] In the formula, represents the number of trajectory shooting frame numbers.

[0142] In the present embodiment, the duration before and after the trajectory lane-changing both exceeds the set time keeping threshold 3s (i.e. the frame number exceeds 30), and thus the lane-changing trajectory is complete.

[0143] Step 4-4, determine the starting point and ending point of the lane-changing

[0144] After obtaining the above effective lane-changing trajectory, the starting point and ending point of the lane-changing behavior in the lane-changing trajectory are determined to obtain the lane-changing trajectory part that can truly represent the driving characteristics. Normally, the start of the lane-changing behavior should be the moment when the vehicle changes from straight driving to lane-changing turning. Ideally, this is also the moment when the angle between the vehicle body and the center line of the lane starts to increase from 0. However, in the actual trajectory, the vehicle has many unconscious slight turns, and the angle between the vehicle body and the center line of the lane cannot always remain 0 degrees. Therefore, a threshold angle α is needed to represent the lane-changing awareness of the driver. Moreover, from a physical point of view, the vehicle body should be completely located on one side of the lane line before the lane-changing behavior and cannot touch the lane line. Therefore, it is also necessary to filter the lane-changing trajectory in which the vehicle body touches the lane line at the start of the lane-changing.

[0145] A, starting point: starting point i q located in the data set , the starting point is the last trajectory frame number i at which the vehicle body is about to contact the lane line * or the last trajectory frame number i at which the angle between the vehicle body and the lane line is less than the set angle threshold α ′ .

[0146] The discriminant of the above lane-changing starting point i q is:

[0147]

[0148] Among them:

[0149] ​

[0150] and

[0151]

[0152]

[0153] and

[0154] wherein, is a set of trajectory frame numbers of the vehicle numbered n when driving straight.

[0155] is an angle between the vehicle body and the lane line of the vehicle numbered n at the i1th trajectory shooting frame.

[0156] is a set of trajectory frame numbers of the vehicle numbered n in the trajectory in which the vehicle body does not contact the lane line.

[0157] is a lateral coordinate of the vehicle numbered n at the i2th trajectory shooting frame.

[0158] X MidLin据 is a lateral coordinate of the lane line of the lane change crossing.

[0159] W n is a vehicle width of the vehicle numbered n.

[0160] The calculation formula of the above is preferably:

[0161] A. When the lane changing direction is from left to right, it is specifically:

[0162]

[0163] B. When the lane changing direction is from left to right, it is specifically:

[0164]

[0165] wherein, and are a lateral coordinate and a longitudinal coordinate of the vehicle numbered n at the i1th trajectory shooting frame, respectively.

[0166] and are a lateral coordinate and a longitudinal coordinate of the vehicle numbered n at the i1+1th trajectory shooting frame, respectively.

[0167] B. End point: end point i zLocated in the dataset The termination point is the minimum trajectory frame number i when the vehicle body approaches the lane line ** Or the minimum trajectory frame number i" when the angle between the vehicle body and the lane line is less than the set angle threshold α.

[0168] The discriminant of the above termination point i z Is:

[0169]

[0170] Where:

[0171]

[0172] And

[0173]

[0174]

[0175] And

[0176] In the formula, Is the angle between the vehicle body and the lane line at the i3th trajectory shooting frame of the vehicle numbered n.

[0177] Is the lateral coordinate of the vehicle numbered n at the i4th trajectory shooting frame.

[0178] The most commonly used signs of lane change start and end are: start-when the vehicle tire or some surface crosses the lane boundary, end-when the rear tire or rear surface of the vehicle crosses the lane boundary. This purely physically based lane change trajectory extraction can only extract the feature values in the middle of the lane change process, and cannot obtain the information of the entire lane change trajectory, which makes the human-like lane change behavior simulation and research of intelligent vehicles not complete enough. Rule four of the method is to regard the start sign as the intention of the driver to start turning to the adjacent lane when changing lanes, and the end sign is when the driver stably positions in the target lane, and the slope of the trajectory edge is used as an identification index of the lane change intention and trend. The extracted trajectory data will be very complete, and the behavior characteristics of the driver adjusting the vehicle when changing lanes and homing the vehicle when ending the lane change can be obtained.

[0179] The selection of the lane changing threshold in the present application needs to be judged according to the average vehicle speed of the road section where the data set is collected. In a low-speed environment, the probability of meaningless shaking of the driving trajectory with a large angle will greatly increase. At this time, the threshold can be set to be higher. The example in this paper is the data in the NGSIM data set, which is a trajectory measured in a medium-high speed environment. By continuously adjusting the threshold in the program and observing the trajectory extraction results, a small lane changing threshold of 1 degree is set. Figure 5 It can be seen that the threshold set to this value can effectively filter the meaningless shaking part before and after the starting point and the ending point of the lane changing trajectory, and the ideal lane changing trajectory conforming to the "S" shape is obtained.

[0180] Therefore, the above-mentioned included angle threshold α needs to be set according to the average vehicle speed of the road section where the original trajectory data is collected, and the included angle threshold α is proportional to the average vehicle speed of the collected road section. In this embodiment, when in a medium-high speed environment (such as above 16 km / h), the included angle threshold α is 1 degree.

[0181] In this embodiment, the trajectory data of vehicle No. 54 in the first time period of the i-80 highway data set is used to introduce the related calculations near the starting point and the ending point in detail.

[0182] 550. Therefore, the angle calculated from the data around it is taken as an example:

[0183]

[0184]

[0185]

[0186]

[0187] Through calculation, it can be seen that Less than 1 degree, indicating that the state of the vehicle at this time can be regarded as straight driving; but in the next frame Greater than 1 degree, indicating that the vehicle at this time has a tendency to turn, and through script recognition, frame 478 is the last frame of the straight driving state of the vehicle before lane changing, so the starting point of the trajectory in this example is frame number 478. Greater than 1 degree, indicating that the state of the vehicle at this time can be regarded as lane changing; but in the next frame Greater than 1 degree, indicating that the vehicle at this time begins to return to straight driving, and through script recognition, frame 550 is the first frame of the straight driving state of the vehicle before lane changing, so the starting point of the trajectory in this example is frame number 550.

[0188] Step 4-5, lane length discrimination is performed on the trajectory data that has been cut by the starting and ending points, and the lane changing lateral displacement The trajectory is less than the minimum set lane change displacement ω.

[0189] Steps 4-6: Filter out the trajectories where the lead car or following car number changes before the vehicle crosses the midpoint of the lane-changing vehicle, and the trajectories where the lead car or following car number changes after the vehicle crosses the midpoint of the lane-changing vehicle.

[0190] In this embodiment, rules 5 (i.e., steps 4-5) and 6 (i.e., steps 4-6) are used to judge and identify the extracted trajectory exhibiting lane-changing behavior. The script determines that the example trajectory meets the requirements. Finally, the extracted trajectory is visualized as follows: Figure 5 As shown, the extracted trajectory exhibits strong lane-changing behavior characteristics both in the global trajectory and in its neighboring trajectories, and its shape also conforms to the expectations of common lane-changing trajectories. All of these are sufficient to prove the reliability of this set of heuristic rules.

[0191] The system filters out trajectories where the lead car or following car number changes before the vehicle crosses the midpoint of the lane-changing vehicle, as well as trajectories where the lead car or following car number changes after the vehicle crosses the midpoint of the lane-changing vehicle. Additionally, if there is a data frame with the lead car or following car number equal to 0 in the extracted lane-changing trajectory, the entire data frame is filtered out.

[0192] During the lane-changing process, the data of the five vehicles related to the lane-changing behavior are available and valid, which enables most lane-changing behavior characteristic studies to be fully analyzed. For example, the vehicle collision risk TTC analysis in step 6 requires complete data of the five vehicles.

[0193] Step 5, Trajectory Data Expansion: Trajectory data expansion includes primary expansion and secondary expansion, specifically as follows:

[0194] A. First expansion: For the lane change trajectory data of vehicle number n obtained in step 4-1, data matching is performed on the keywords of the preceding vehicle number and the following vehicle number in the current track before the lane change, so that the lane change trajectory data includes the trajectory shooting frame data of the vehicle, the lead vehicle, and the following vehicle of vehicle number n in the current frame.

[0195] B. Secondary expansion: After the first expansion is completed, data matching is performed on the keywords of the preceding vehicle number and the following vehicle number in the target track after the lane change, so that the lane change trajectory data includes the trajectory shooting frame data of the vehicle with vehicle number n, the lead vehicle before the lane change, the following vehicle before the lane change, the lead vehicle after the lane change, and the following vehicle after the lane change in the current frame.

[0196] Step 6: Calculate the higher-order data collision time (TTC). The specific calculation formula is as follows:

[0197]

[0198]

[0199]

[0200]

[0201] In the formula,

[0202] and These represent the collision times of vehicle number n before and after changing lanes with the lead vehicle;

[0203] and These represent the collision times between vehicle n and the following vehicle before and after the lane change;

[0204] and These represent the longitudinal coordinates of the lead vehicle before and after the lane change of vehicle number n in frame i, respectively.

[0205] and These represent the longitudinal coordinates of the vehicle before and after lane change in frame i, respectively;

[0206] This represents the vertical coordinate of vehicle number n in frame i.

[0207] and These represent the longitudinal speeds of the lead car before and after the lane change of vehicle number n in frame i, respectively.

[0208] and These represent the longitudinal speeds of the vehicle numbered n before and after the lane change in frame i, respectively.

[0209] This represents the longitudinal speed of vehicle number n in frame i.

[0210] Step 7: Calculate the Standardized Fuel Rate (NFR). The specific calculation formula is as follows:

[0211] NFR = 1.71·VSP 0.42

[0212] in:

[0213]

[0214] In the formula, and These represent the lateral and longitudinal velocities of vehicle number n in frame i, respectively.

[0215] and These represent the lateral and longitudinal accelerations of vehicle number n in frame i, respectively.

[0216] In this example, the five-car identifier in the classic lane-changing model is as follows: Figure 2 As shown in the example data, a trajectory data point with vehicle number 54 and frame number 520 shows that the vehicle's speed is 5.881283 m / s and its acceleration is 1.40888 m / s². 2 This example demonstrates how to calculate NFR:

[0217]

[0218] NFR = 1.71·VSP 0.42 =1.71 × 9.8927 0.42 =4.48875

[0219] The preferred embodiments of the present invention have been described in detail above. However, the present invention is not limited to the specific details in the above embodiments. Within the scope of the technical concept of the present invention, various equivalent transformations can be made to the technical solutions of the present invention, and these equivalent transformations all fall within the protection scope of the present invention.

Claims

1. A heuristic rule based vehicle lane change trajectory identification method, characterized in that: Comprising the following steps: Step 1, data segmentation record judgment: sort the original trajectory data by global time as the key word, judge whether the original trajectory data separates the data observed in different time periods for separate recording; when there is only one shooting frame serial number for the same global time in the original trajectory data, it is judged that the original trajectory data separates the data observed in different time periods for separate recording, and directly enters step 3; otherwise, it is judged that the original trajectory data does not separate the data observed in different time periods for separate recording, and enters step 2; Step 2, data segmentation record: when the global time corresponds to a shooting frame serial number, the original trajectory data is divided into a time period of observation data for separate recording, so that the same global time has only one shooting frame serial number; Step 3, data smoothing: first, sort each time period of observation data for separate recording by vehicle number n as the main key word and shooting frame serial number i as the secondary key word; then, the position data of each time period of observation data after sorting is smoothed to obtain the smoothed observation data of vehicle number n in each time period; Step 4, lane change trajectory identification: based on heuristic rules, the smoothed observation data of each vehicle number n in step 3 is subjected to lane change trajectory identification, which comprises the following steps: Step 4-1, lane change trajectory judgment: the change of lane number column values in a group of smoothed observation data is used to preliminarily judge whether the current trajectory is a lane change trajectory; when the number of lane number changes in a group of smoothed observation data exceeds once, it is regarded as a lane change, and the trajectories without lane change and the trajectories with more than 2 lane changes are filtered out to obtain the lane change trajectory data with one lane change; Step 4-2, determining the lane-changing midpoint: for the lane-changing trajectory data obtained in step 4-1, the lane line position distribution of the corresponding data set collection road section is determined to determine the lane-changing midpoint; the trajectory frame sequence number set of the vehicle numbered n before crossing the lane-changing midpoint is called The trajectory frame sequence number set of the vehicle numbered n after crossing the lane-changing midpoint is called Step 4-3, maintaining the integrity of the lane change trajectory: the vehicle trajectory with a duration of the trajectory data before or after the lane change less than the set time maintaining threshold is filtered out to ensure the integrity of the obtained lane change trajectory; Step 4-4, determining the starting point and the ending point of the lane change, specifically: A, starting point: starting point i q Located in the dataset The starting point is the last trajectory frame number i when the vehicle body is about to contact the lane line * Or the last trajectory frame number i' when the angle between the vehicle body and the lane line is less than the set angle threshold α B, end point: end point i z Located in the dataset In which the end point is the minimum trajectory frame sequence number i of the vehicle body approaching the lane line ** Or the minimum trajectory frame sequence number i" of the vehicle body and the lane line angle less than the set angle threshold α. 2.The heuristic rule based vehicle trajectory identification method of changing lane according to claim 1, characterized in that: Step 4 - In step 4-4, the lane change start point i q The discriminant is: Wherein: and and In the formula, is a set of track frame numbers when the nth vehicle is driving straight; is the angle between the body of the numbered n vehicle and the lane line at the i1th track shot frame; is a set of sequence numbers of trajectory frames in which the vehicle body does not contact the lane line in the n-th vehicle trajectory; is the lateral coordinate of the vehicle numbered n at the i2trajectory shot frame; X MidLine is the lateral coordinate of the lane line of the lane change; W n The vehicle width of the n-th vehicle. 3.The heuristic rule based vehicle trajectory identification method of changing lane according to claim 2, characterized in that: Step 4-4 in, The calculation formula is: A, when the lane change direction is from left to right, specifically: B, when the lane change direction is from left to right, specifically: wherein and are the lateral and longitudinal coordinates of the n-th vehicle at the i-th trajectory shot frame, respectively. and and are the lateral and longitudinal coordinates of the n-th vehicle at the (i+1)-th frame of the i-th trajectory, respectively.

4. The heuristic rule-based vehicle trajectory identification method of changing lanes according to claim 3, characterized in that: Step 4-4, termination point i z The discriminant is: Wherein: and and In the formula, is the angle between the vehicle body and the lane line at the i3th trajectory shooting frame of the n-th vehicle; is the lateral coordinate of the vehicle numbered n at the i4th trajectory shot frame. 5.The heuristic rule based vehicle trajectory identification method for lane change according to claim 1, wherein: In step 4, the included angle threshold α needs to be set according to the average speed of the original trajectory data collection section, and the included angle threshold α is proportional to the average speed of the collection section.

6. The heuristic rule-based vehicle trajectory identification method of changing lanes according to claim 5, characterized in that: When the medium-high speed environment is in the medium-high speed environment, the included angle threshold α is 1 degree. 7.The heuristic rule based vehicle trajectory identification method for lane change according to claim 1, wherein: In step 4-3, the calculation formula of the set time maintaining threshold ΔT is: In the formula, is represented as the number of trajectory shot frame numbers. 8.The heuristic rule based vehicle trajectory identification method for lane change according to claim 1, wherein: Further comprising step 5, trajectory data expansion: trajectory data expansion includes primary expansion and secondary expansion, specifically: A, primary expansion: for the lane change trajectory data of vehicle number n obtained in step 4-1, data matching is performed on the key words of the front and rear vehicle numbers in the current lane before the lane change, so that the lane change trajectory data contains the current frame trajectory shooting frame data of the vehicle number n, the lead vehicle and the following vehicle; B, secondary expansion: after the primary expansion is completed, data matching is performed on the key words of the front and rear vehicle numbers in the target lane after the lane change, so that the lane change trajectory data contains the current frame trajectory shooting frame data of the vehicle number n, the lead vehicle before the lane change, the following vehicle before the lane change, the lead vehicle after the lane change and the following vehicle after the lane change. 9.The heuristic rule based vehicle trajectory identification method for lane change according to claim 5, wherein: Also included are steps 6, calculating the high-order data time-to-collision TTC, taking it as the safety evaluation index of traffic conditions, because it stipulates the remaining time before the vehicle at the current time collides with its leader, and the specific calculation formula is: In the formula, and tn and tn+1 represent the time of collision with the leading vehicle before and after the vehicle n changes lane, respectively; and tn and tn+1 represent the time to collision before and after the vehicle n changes lane, respectively, with the following vehicle; and Xi(n) and X2(n) represent the longitudinal coordinates of the leading vehicle before and after the vehicle n changes lane at frame i, respectively; and Xi(n) and Xf(n) represent the longitudinal coordinates of the vehicle numbered n before and after changing lanes, respectively, at frame i. represents the longitudinal coordinate of the vehicle with number n at the frame with index i; and vi, n, i and v, n, i represent the longitudinal speed of the vehicle n before and after the lane change, respectively, at frame i; and vi, n, i and vi, n, i represent the longitudinal speed of the vehicle preceding the change of lane and of the following vehicle in the frame with index i, respectively, numbered n; denotes the longitudinal velocity of the vehicle with the number n at the frame with the sequence number i.

10. The heuristic rule-based vehicle trajectory identification method of changing lanes according to claim 9, characterized in that: Also included are steps 7, calculating the normalized fuel consumption rate NFR, and the specific calculation formula is: NFR = 1.71 · VSP 0.42 Wherein: wherein and vi,n(i) and vx,n(i) denote the lateral and longitudinal velocity of the vehicle numbered n at the frame with index i, respectively; and Xi and Xn represent the lateral and longitudinal acceleration of the vehicle numbered n at frame i.

Images