Vehicle track online correction method based on thunder-vision fusion

Abstract

The invention discloses a vehicle track online correction method based on thunder-vision fusion. The method comprises the following steps of original track interval positioning, longitudinal track smoothing processing, transverse track smoothing processing, vehicle data breakpoint detection, vehicle abnormal lane changing detection and vehicle coordinate correction value calculation. According to the invention, on-line detection is carried out on a'round trip 'mode of a laneid sequence, a geometric fitting formula of a lane center line and a coordinate smooth correction model are coupled, and identification and restoration of an abnormal lane changing track are rapidly completed; according to the method, off-line training is not needed, the calculation amount is low, the method can be directly deployed on an edge calculation unit, instant correction of abnormal lane changing data and accurate restoration of lane-level coordinates are achieved, and therefore the integrity of thunder-vision fusion data and the stability of upper-layer application are comprehensively improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of road traffic, in particular to a vehicle trajectory online correction method based on radar and vision fusion. BACKGROUND

[0002] With the rapid development of intelligent transportation system to lane-level fine control, radar and vision fusion system can continuously output high-frequency vehicle trajectory coordinates and corresponding lane numbers in all-weather and complex lighting conditions due to the cooperative perception advantages of millimeter wave radar and high-definition camera, and has become a key infrastructure in the field of current traffic information collection and state perception.

[0003] However, in actual operation, affected by various factors such as physical obstruction, device vibration, electromagnetic interference, communication delay or packet loss, multi-target interference, etc., the vehicle trajectory data output by the radar and vision fusion system often has two typical problems: first, the random absence of coordinate sequence, usually caused by short-time obstruction or instantaneous communication interruption, resulting in discontinuity of the trajectory in the time dimension; second, abnormal lane jumping, which is manifested as a non-physical mutation of lane number at adjacent time, which is usually caused by sensor noise, coordinate drift or multi-target tracking error, and seriously affects the judgment of the real driving behavior of the vehicle by lane-level applications.

[0004] At present, the mainstream methods for vehicle trajectory repair can be roughly divided into three categories: traditional methods based on interpolation, filtering methods based on state estimation and prediction methods based on deep learning. Traditional interpolation methods such as linear interpolation and spline interpolation assume that the vehicle moves at a constant speed or uniform acceleration in the missing interval, and the calculation process is simple, but it is difficult to adapt to the frequent lane changing, acceleration and deceleration and other nonlinear behaviors in real traffic, and the repaired trajectory is easy to deviate from the actual lane line. The filtering-based method introduces a vehicle kinematic model for state prediction and update, which can handle dynamic changes to some extent, but its performance is highly dependent on the prior statistical characteristics of process noise and observation noise, and its robustness to non-Gaussian distributed outliers and sudden changes is poor, and it is prone to lag or drift in strong nonlinear stages such as vehicle lane changing and sudden braking. The deep learning-based method can capture complex temporal dependence and behavior patterns through learning from a large amount of labeled data, and has high prediction accuracy, but this kind of method generally has the problems of complex model, large amount of calculation, dependence on offline training and a large amount of labeled data, etc., and it is difficult to realize low-delay real-time deployment on resource-constrained edge computing nodes.

[0005] The existing method generally ignores the lane geometry constraints on the vehicle trajectory. The movement of the vehicle on the road is not random movement in free space, and its lateral position is strictly limited by the lane width, and the longitudinal movement is also guided by the lane line direction. Ignoring the lane geometry characteristics of trajectory repair, even if the continuity of the time dimension is guaranteed, the repaired coordinate points may deviate from the true lane center line, and then cause "secondary error" to the downstream tasks such as lane-level traffic statistics, vehicle trajectory clustering and abnormal event identification.

[0006] Therefore, there is an urgent need in the art for a trajectory online correction scheme that can balance real-time, accuracy and lane constraints. The scheme should have the following characteristics: first, it can be deployed in an edge computing unit in a lightweight manner without complex offline training or a large number of parameter tuning; second, it can simultaneously handle the problems of breakpoint loss and abnormal lane change, and ensure the consistency and rationality of the trajectory in the time and space dimensions; third, it deeply integrates the lane center line geometry model to ensure that the correction result conforms to the actual driving rules of the vehicle on the structured road. SUMMARY

[0007] To solve the above problems, the present application provides a vehicle trajectory online correction method based on radar and vision fusion. After arranging radar and vision integrated detectors along the road to collect high-frequency trajectory coordinates of all vehicles in real time, the method sequentially undergoes six steps of original trajectory interval positioning, longitudinal trajectory smoothing processing, lateral trajectory smoothing processing, vehicle data breakpoint detection, vehicle abnormal lane change detection and vehicle coordinate correction value calculation to correct the vehicle abnormal lane change.

[0008] To solve the problems in the background art and achieve the above technical purposes, the present application is implemented by the following technical solutions:

[0009] A vehicle trajectory online correction method based on radar and vision fusion, comprising the following steps: original trajectory interval positioning, longitudinal trajectory smoothing processing, lateral trajectory smoothing processing, detecting vehicle data breakpoint, detecting vehicle abnormal lane change and calculating vehicle coordinate correction value.

[0010] Preferably, the original trajectory interval positioning comprises: time stamp alignment, coordinate system unification and data validity verification of the original vehicle trajectory data, eliminating invalid data with longitude and latitude exceeding the preset threshold range, and retaining valid data.

[0011] Preferably, the longitudinal trajectory smoothing processing comprises: using a m-second sliding time window to respectively perform arithmetic average processing on the longitude and latitude coordinates of the valid data, and only calculating the mean value of the valid values when there is data loss in the window to generate smoothed correction coordinates.

[0012] Preferably, the specific calculation method in the longitudinal trajectory smoothing processing step is:

[0013] Longitude coordinate correction formula: , wherein, is the original longitude coordinate value, is the corrected longitude coordinate value, and n is the number of effective original longitude coordinates in the window;

[0014] Latitude coordinate correction formula: , wherein, is the original latitude coordinate value, is the corrected latitude coordinate value, and n is the number of effective original latitude coordinates in the window.

[0015] Preferably, the transverse trajectory smoothing processing comprises:

[0016] Based on the measured data, taking L2 lane as the reference lane, the least square method is used to establish the center line fitting equation of L1, L2 and L3 lanes,

[0017] L1 lane (lane_id=0):

[0018] L2 lane (lane_id=1):

[0019] L3 lane (lane_id=2):

[0020] In the formula, a1, b1, c1 are the coefficients of the center line fitting equation of L1 lane; a2, b2, c2 are the coefficients of the center line fitting equation of L2 lane; a3, b3, c3 are the coefficients of the center line fitting equation of L3 lane;

[0021] The lane_id field is used to determine the lane in which the vehicle is located, and the latitude coordinate is substituted into the fitting equation of the corresponding lane to calculate the corrected longitude coordinate of the lane.

[0022] Preferably, the detection of vehicle data breakpoints comprises the following steps: using a m-second sliding time window to detect the null value of the smoothed coordinate sequence, marking the breakpoints caused by shielding or communication interruption, extracting the longitude and latitude of 1-3 effective historical points before the breakpoint as the repair basis, and if the number of continuous effective data before the breakpoint is less than 2, then the default motion parameters are used.

[0023] Preferably, the detection of vehicle abnormal lane changing comprises the following steps:

[0024] A three-frame analysis window is used to calculate the lane change strength index A:

[0025]

[0026] In the formula, A represents the lane change value, This represents the lane number detected by the vehicle at the current time t. This indicates the lane number detected by the vehicle at the next time t+1. This indicates the lane number detected by the vehicle at time t+2;

[0027] When lane_id(t) = lane_id(t+2) ≠ lane_id(t+1), it is determined to be an abnormal lane change.

[0028] Preferably, the calculation of the vehicle coordinate correction value includes:

[0029] For coordinate data breakpoints, the coordinates are predicted and corrected by combining differential displacement calculation with lane fitting formula based on the number of valid historical points before the breakpoint.

[0030] For abnormal lane changes, the lane_id at times t, t+1, and t+2 is reset to the value at time t, and the coordinates are recalculated using the corresponding lane fitting formula.

[0031] Preferably, the correction method for coordinate data breakpoints includes:

[0032] When there are 3 valid historical points t1, t2, and t3 before the breakpoint, calculate the displacement. , , ,pass Calculate the latitude coordinates and combine them with the corresponding lane fitting formula to obtain the longitude coordinates lng4;

[0033] When there are two valid historical points t1 and t2 before the breakpoint, calculate the displacement. ,pass Calculate the latitude coordinates and combine them with the corresponding lane fitting formula to obtain the longitude coordinates lng3;

[0034] When there is only one valid historical point t1 before the breakpoint, a fixed displacement is used. Compensation is then applied, and the corresponding lane fitting formula is used to predict the coordinates at subsequent time points.

[0035] Preferably, during the correction process for abnormal lane changes, the reset lane_id is used to redetermine the lane where the vehicle is located, and then the corresponding lane fitting formula is selected to recalculate the longitude correction coordinates at each time during the abnormal period. The latitude coordinates are kept as smoothed values ​​or finely adjusted according to lane constraints.

[0036] The beneficial effects of this invention are as follows: This invention quickly identifies and repairs abnormal lane change trajectories by online detection of the "round-trip jump" pattern of the lane_id sequence, coupled with the lane centerline geometric fitting formula and the coordinate smoothing correction model; this method does not require offline training, has low computational cost, and can be directly deployed on edge computing units to achieve real-time correction of abnormal lane change data and accurate restoration of lane-level coordinates, thereby comprehensively improving the integrity of radar-visual fusion data and the stability of upper-layer applications. Attached Figure Description

[0037] Figure 1 This is a flowchart of the method of the present invention.

[0038] Figure 2 This is a schematic diagram of lane centerline fitting.

[0039] Figure 3 This is a schematic diagram of breakpoint identification and repair. Detailed Implementation

[0040] To enable those skilled in the art to better understand the technical solutions of the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments in this specification. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0041] This invention provides an online vehicle trajectory correction method based on radar-visual fusion. By deploying integrated radar-visual detectors along the roadside to collect high-frequency trajectory coordinates of all vehicles in real time, the method sequentially performs six steps: original trajectory interval positioning, longitudinal trajectory smoothing, lateral trajectory smoothing, vehicle data breakpoint detection, abnormal lane change detection, and vehicle coordinate correction value calculation. This achieves correction of abnormal lane changes. (See [link to relevant documentation]). Figure 1 The specific technical solution is as follows:

[0042] (I) Step 1. Localization of the original trajectory interval

[0043] The raw vehicle trajectory data output by the radar-visual fusion system is first subjected to standardized preprocessing, which includes timestamp alignment, coordinate system unification, and data validity verification.

[0044] Based on the actual geographic information of the target road segment, the system pre-sets a reasonable threshold range for latitude and longitude data. If the longitude or latitude in the data exceeds the preset range, it is considered invalid data and is removed; if both longitude and latitude are within the preset range, it is determined to be valid data and enters the longitudinal trajectory smoothing process.

[0045] (II) Step 2. Smoothing of longitudinal trajectory

[0046] The longitudinal trajectory smoothing process employs a sliding time window method, using m seconds as the sliding time window (m can be adjusted according to road traffic density and equipment sampling frequency, ranging from 0.1 to 1 second). The longitude (Lng) and latitude (Lat) coordinates collected by the radar equipment are arithmetically averaged to generate smoothed correction coordinates within that second. If data is missing within the window, only the valid values ​​are averaged to effectively suppress trajectory noise caused by high-frequency signal jitter and partial data loss. The specific steps are as follows:

[0047] Step 1: Calculate the smoothed longitude coordinates. The longitude coordinates (Lng) measured by the radar camera are smoothed in m-second increments. A correction coordinate is generated every m seconds; the correction coordinate is the average longitude coordinate within that second. The calculation method is as follows:

[0048] (1)

[0049] (2)

[0050] In the formula: The original longitude coordinates are directly measured by the radar-visual fusion equipment at the corresponding moment. To smooth the original longitude coordinates longitudinally, at the corresponding time t n The output is the corrected final longitude coordinate value. In actual processing, the average value is calculated based on the number of valid coordinate points actually received within the window, ensuring that the smoothing process can still proceed normally even when some data points are missing.

[0051] Step 2: Set the start and end points of the sliding time window as t1→t n Within m seconds, n raw trajectory data points are received. The arithmetic mean of the latitudes of all valid raw points is taken, and missing values ​​are not counted. Finally, the result is... , The latitude coordinates, used as the smoothed correction coordinates for that second, are calculated as follows:

[0052] (3)

[0053] (4)

[0054] In the formula: The original latitude coordinates are directly measured by the radar-visual fusion equipment at the corresponding moment. To perform vertical smoothing on the original latitude coordinate values, at the corresponding time t n The output is the corrected final latitude coordinate value.

[0055] (III) Step 3. Lateral trajectory smoothing

[0056] Step 1: In the lane centerline fitting stage, based on the measured data, the centerline equations of the three lanes (L1, L2, L3) are established using the least squares method. See also Figure 2 Using lane L2 as the reference lane, the longitude-latitude fitting relationship of lane centerline is obtained by fitting the measured centerline coordinates of lane L2 using the least squares method. Then, the fitting formulas of the centerlines of lanes L1, L2, and L3 are obtained, and the fitting error is controlled within 0.1 meters.

[0057] Step 2: Determine the range of the longitude coordinates (Lng) using the lane_id field in the radar detection data to identify the lane where the point is located, and then select the correction formula corresponding to the lane type:

[0058] If lane_id=0, the new longitude coordinates are calculated using the fitting equation for lane L1, as shown in the following formula:

[0059] (5)

[0060] If lane_id=1, the new longitude coordinates are calculated using the fitting equation for the L2 lane, as shown in the following formula:

[0061] (6)

[0062] If lane_id=2, the new longitude coordinates are calculated using the fitting equation for the L3 lane, as shown in the following formula:

[0063] (7)

[0064] In the formula: a1, b1, c1 are the coefficients of the L1 lane centerline fitting equation; a2, b2, c2 are the coefficients of the L2 lane centerline fitting equation; a3, b3, c3 are the coefficients of the L3 lane centerline fitting equation.

[0065] Step 3: Substitute the latitude coordinates (Lat) into the fitting formula of the corresponding lane centerlines (L1, L2, L3) to calculate the lane-smoothing corrected longitude coordinates (Lng).

[0066] (iv) Step 4. Detect vehicle data breakpoints

[0067] After completing lane smoothing, the system conducts an in-depth analysis of the completeness and logical consistency of the current vehicle trajectory data. By constructing a two-level progressive detection mechanism, it sequentially identifies physical missing data and abnormal jumps in logic, thereby comprehensively diagnosing typical problems in the trajectory data and providing clear targets for subsequent accurate corrections. The specific detection steps are as follows:

[0068] Step 1: Perform null value detection on the preprocessed coordinate sequence to identify breakpoints caused by data loss due to occlusion or communication interruption. A sliding time window of m seconds is used to scan the continuous coordinate sequence. If an invalid value appears within the window, it is marked as a breakpoint to avoid false positives caused by fixed thresholds.

[0069] Step 2: If a coordinate breakpoint exists, mark it as a breakpoint when the coordinates are confirmed to be empty at the current time, and extract the latitude, longitude, and lane ID (lane_id) of the 1-3 valid historical points before the breakpoint as the basis for repair. If there are fewer than 2 valid points of continuous data before the breakpoint, use the default motion parameters to fix the displacement. As compensation.

[0070] (v) Step 5. Detect abnormal lane changes by vehicles

[0071] Step 1: With complete coordinate data, the system further performs a logical consistency analysis on the lane number (lane_id) sequence accompanying the coordinates. When a vehicle travels on a structured road, its lane change process should have spatial continuity and smooth motion; the difference in lane numbers between adjacent moments is typically only 0 (keeping in lane) or 1 (changing to an adjacent lane). To identify non-physical lane jumps caused by noise interference, target mismatch, or data jitter, the system uses a three-frame analysis window to calculate the lane change intensity index A, calculated using the following formula:

[0072] (8)

[0073] In the formula: A is the lane change value, which is used to quantify the magnitude of the lane number change of vehicles at adjacent time points; The lane number detected by the vehicle at time t; The lane number detected by the vehicle at time t+1; This is the lane number detected by the vehicle at time t+2.

[0074] Based on the detection results in step 4, determine the type of correction required at the current moment.

[0075] Step 2: Based on the physical continuity of vehicle lane-changing behavior, that is, during a normal lane-changing process, the directions of two adjacent lane changes should remain consistent or remain stationary. When three consecutive lane number changes satisfy lane_id(t) = lane_id(t+2) ≠ lane_id(t+1), that is... This indicates that the vehicle changes lanes in one direction and then reverses back in a very short time. This non-monotonic change violates the actual motion law of the vehicle and does not meet the physical characteristic that lane changing requires continuous lateral displacement. Therefore, it is judged as an abnormal lane change.

[0076] (vi) Step 6. Calculate the vehicle coordinate correction value

[0077] Based on the above detection results, for the two types of problems, namely "coordinate data breakpoints" and "abnormal lane change jumps," a differentiated correction algorithm that deeply integrates lane geometric constraints and vehicle kinematics is adopted to achieve online real-time repair and reconstruction of trajectory data. The specific steps are as follows:

[0078] (1) Correction of coordinate data breakpoints

[0079] Step 1: See Figure 3 Determine if the vehicle coordinates at the current time t are null. If so, proceed to the breakpoint correction process; otherwise, return to the abnormal lane change detection step.

[0080] Step 2: If the vehicle coordinates at time t4 are empty, predict the coordinates at time t4 (lng4, lat4) based on the coordinates t1(lng1, lat1), t2(lng2, lat2), and t3(lng3, lat3) of the previous three times (t1, t2, t3), and calculate the displacements d1, d2, and d3 using the following formulas:

[0081] (9)

[0082] (10)

[0083] (11)

[0084] In the formula: d1 is the vehicle displacement from time t1 to time t2; d2 is the vehicle displacement from time t2 to time t3; d3 is the distance between the vehicle breakpoint coordinates from time t3 to time t4.

[0085] Step 3: Calculate the latitude coordinate lat4 at time t4. Determine the lane the vehicle is in using the lane_id field, and calculate the longitude coordinate lng4 according to the fitting formula for the corresponding lane. Finally, obtain the coordinates (lng4, lat4) at time t4. n (lng n lat n The calculation method is the same; the calculation method is as follows:

[0086] (12)

[0087] (13)

[0088] In the formula: The formula is the lane fit formula for the lanes where the first 3 coordinate points are located. If the first 3 coordinate points are not in the same lane, the fit formula for the lane where the nearest point is located will be used.

[0089] Step 4: If the vehicle coordinates at time t3 are empty, and there are only two valid coordinates before this empty value, t1(lng1, lat1) and t2(lng2, lat2), then predict the coordinates at time t3 (lng3, lat3) based on these two coordinates, and calculate the displacement d:

[0090] (14)

[0091] Step 5: Calculate the coordinates (lng3, lat3) at time t3 based on the displacement d. Determine the lane the vehicle is in using the lane_id field, and use the corresponding lane fitting formula to calculate the coordinates. The specific formula is as follows:

[0092] (15)

[0093] (16)

[0094] In the formula: The formula is the lane fit formula for the lanes where the first two coordinate points are located. If the first two coordinate points are not in the same lane, the formula for the lane where the nearest point is located will be used.

[0095] Step 6: If there is only one valid coordinate before the null value time (time t1 (lng1, lat1)), then take a fixed displacement. As compensation, the coordinates at subsequent time points are predicted (taking time t2 as an example, the coordinates are (lng2, lat2)). The lane number of the vehicle is determined by the lane_id field, and the corresponding lane fitting formula is used for calculation. The specific formula is as follows:

[0096] (17)

[0097] (18)

[0098] In the formula: The formula for fitting the lane direction of the lane where the previous coordinate point is located; The value of the fixed displacement is related to the degree of data loss and the preset road segment parameters. It needs to be set in conjunction with the statistical characteristics of the historical trajectory data of the road segment, and the value range is 0.5-5 meters.

[0099] (2) Correction for abnormal lane change jumps

[0100] Step 7: When an abnormal lane change is determined (i.e., lane_id(t) = lane_id(t+2) ≠ lane_id(t+1)), the lane number of the abnormal period is reset, and the lane_id values ​​corresponding to times t, t+1, and t+2 are all reset to the lane_id value at time t. Then, the lane where the point is located is re-determined based on the reset lane_id, and the correction formula (5)-(7) of the corresponding lane is selected to recalculate the coordinate values ​​at each time.

[0101] Step 8: For vehicles that do not exhibit abnormal lane-changing behavior (i.e., lane change value A satisfies normal lane-changing logic) The lane_id values ​​corresponding to times t, t+1, and t+2 retain their original data values ​​at their respective times. Then, return to Step 2-3 of the lateral trajectory smoothing process to redetermine the lanes at each point and complete the coordinate smoothing correction.

[0102] This invention addresses the issues of missing breakpoints and abnormal lane change jumps in vehicle trajectory data collected by radar-visual fusion systems. It provides a lightweight, online-running trajectory correction method with the following advantages:

[0103] Balancing real-time performance with ease of deployment: This invention requires no offline training process, has low computational load, can be directly deployed on edge computing units, and does not require complex parameter tuning. It can meet the real-time data processing needs of all-weather, high-concurrency scenarios and effectively adapt to resource-constrained edge computing environments.

[0104] Comprehensive correction: This method can simultaneously handle two typical problems: missing trajectory breakpoints and abnormal lane change jumps. By constructing a two-level progressive detection mechanism, it accurately identifies data anomalies, and then uses a differentiated correction algorithm to achieve consistency restoration of the trajectory in the time and space dimensions, ensuring the integrity and rationality of the trajectory data.

[0105] High correction accuracy: Deeply integrates the lane centerline geometric model and incorporates lane geometric constraints into the coordinate correction process. This ensures that the corrected trajectory strictly follows the actual driving patterns of vehicles on structured roads, avoiding the "secondary error" of trajectory deviation from the lane and significantly improving the accuracy and usability of trajectory data.

[0106] Significant application value: By correcting abnormal trajectory data in real time, it is possible to restore the actual driving path of vehicles, eliminate logical jumps in lane_id sequences, effectively reduce the risk of false alarms and policy failures, and provide reliable data support for upper-layer applications such as high-precision traffic flow statistics, proactive safety warnings and vehicle-road cooperative control, thus comprehensively improving the application value of radar-visual fusion data and the stability of upper-layer management and control services.

[0107] In summary, this invention generates a repair trajectory that conforms to vehicle motion patterns and strictly follows lane space constraints by coupling calculations of differential trend estimation, lane change logic judgment, and lane centerline geometric fitting, combined with lane attribute verification and motion continuity verification. This effectively solves the problems of poor adaptability, reliance on prior conditions, and neglect of lane constraints in existing methods, and provides high-quality data support for lane-level traffic management.

[0108] Regarding the embodiments of the present invention, it should also be noted that, without conflict, the embodiments of the present invention and the features in the embodiments can be combined with each other to obtain new embodiments.

[0109] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. The scope of protection of the present invention should be determined by the scope of the claims. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for online vehicle trajectory correction based on radar-visual fusion, characterized in that, Includes the following steps: The system includes original trajectory interval positioning, longitudinal trajectory smoothing, lateral trajectory smoothing, detection of vehicle data breakpoints, detection of abnormal vehicle lane changes, and calculation of vehicle coordinate correction values.

2. The method according to claim 1, characterized in that, The original trajectory interval positioning includes: performing timestamp alignment, coordinate system 1 verification, and data validity check on the original vehicle trajectory data, removing invalid data whose latitude and longitude exceed the preset threshold range, and retaining valid data.

3. The method according to claim 2, characterized in that, The longitudinal trajectory smoothing process includes: using an m-second sliding time window, performing arithmetic mean processing on the longitude and latitude coordinates of the valid data respectively; when there is missing data within the window, only the mean is calculated on the valid values ​​to generate smoothing correction coordinates.

4. The method according to claim 3, characterized in that, The specific calculation method in the longitudinal trajectory smoothing process is as follows: Longitude coordinate correction formula: In the formula, These are the original longitude coordinates. The corrected longitude coordinates are the values, and n is the number of valid original longitude coordinates within the window. Latitude coordinate correction formula: In the formula, These are the original latitude coordinates. The corrected latitude coordinates are denoted as n, where n is the number of valid original latitude coordinates within the window.

5. The method according to claim 1, characterized in that, The lateral trajectory smoothing process includes: Based on measured data, using lane L2 as the baseline lane, the centerline fitting equations for lanes L1, L2, and L3 are established using the least squares method: L1 lane (lane_id=0): L2 lane (lane_id=1): L3 lane (lane_id=2): In the formula, a1, b1, and c1 are the coefficients of the L1 lane centerline fitting equation; a2, b2, and c2 are the coefficients of the L2 lane centerline fitting equation; and a3, b3, and c3 are the coefficients of the L3 lane centerline fitting equation. The lane number of the vehicle is determined by the lane_id field. The latitude coordinates are then substituted into the fitting equation of the corresponding lane to calculate the corrected longitude coordinates for lane smoothing.

6. The method according to claim 1, characterized in that, The detection of vehicle data breakpoints includes the following steps: using an m-second sliding time window to perform null value detection on the smoothed coordinate sequence, marking breakpoints caused by occlusion or communication interruption, extracting the latitude and longitude and lane_id of 1-3 valid historical points before the breakpoint as the basis for repair, and if there are fewer than 2 consecutive valid data points before the breakpoint, then the default motion parameters are enabled.

7. The method according to claim 1, characterized in that, The detection of abnormal lane changes by the vehicle includes the following steps: The lane change intensity index A is calculated using a three-frame analysis window. In the formula, A represents the lane change value. This represents the lane number detected by the vehicle at the current time t. This indicates the lane number detected by the vehicle at the next time t+1. This indicates the lane number detected by the vehicle at time t+2; When lane_id(t) = lane_id(t+2) ≠ lane_id(t+1), it is determined to be an abnormal lane change.

8. The method according to claim 1, characterized in that, The calculation of the vehicle coordinate correction value includes: For coordinate data breakpoints, the coordinates are predicted and corrected by combining differential displacement calculation with lane fitting formula based on the number of valid historical points before the breakpoint. For abnormal lane changes, the lane_id at times t, t+1, and t+2 is reset to the value at time t, and the coordinates are recalculated using the corresponding lane fitting formula.

9. The method according to claim 8, characterized in that, Methods for correcting coordinate data breakpoints include: When there are 3 valid historical points t1, t2, and t3 before the breakpoint, calculate the displacement. , , ,pass Calculate the latitude coordinates and combine them with the corresponding lane fitting formula to obtain the longitude coordinates lng4; When there are two valid historical points t1 and t2 before the breakpoint, calculate the displacement. ,pass Calculate the latitude coordinates and combine them with the corresponding lane fitting formula to obtain the longitude coordinates lng3; When there is only one valid historical point t1 before the breakpoint, a fixed displacement is used. Compensation is then applied, and the corresponding lane fitting formula is used to predict the coordinates at subsequent time points.

10. The method according to claim 8, characterized in that, During the correction process for abnormal lane changes, the reset lane_id is used to redetermine the lane where the vehicle is located, and then select the corresponding lane fitting formula to recalculate the longitude correction coordinates at each time during the abnormal period. The latitude coordinates are kept as smoothed values ​​or finely adjusted according to lane constraints.

Images