A lane keeping anomaly detection method and device in a decentralized vehicle group environment

Abstract

The application discloses a lane keeping anomaly detection method and device in a decentralized vehicle group environment, mainly aiming at the problems that the existing single-vehicle intelligent lane keeping scheme cannot distinguish between sensing anomalies and real driving deviations, the lane reference cognition of the decentralized vehicle group is not unified, and the safety redundancy is insufficient. The visual lane line parameters of the vehicle and neighbor vehicles are collected first, and after the coordinate system alignment, the vehicle group consensus lane line is generated through a decentralized distributed consistency filter. Then, the multi-dimensional lateral deviation is calculated to obtain the vehicle sensing deviation and neighbor control deviation. The threshold comparison is performed to determine the anomaly type and output the corresponding detection signal. Compared with the prior art, the application can accurately identify multiple types of anomalies without relying on a central node and a high-precision map, can guarantee the collaborative stability of the vehicle group, can improve the safety and robustness of the lane keeping system, and is suitable for commercial vehicle high-speed platoon scenes and the like.

Description

Technical Field

[0001] This invention belongs to the field of intelligent connected vehicle technology, specifically relating to a method and device for detecting lane keeping anomalies in a decentralized vehicle group environment. Background Technology

[0002] Lane keeping assist relies on the vehicle's own sensors and real-time communication with other vehicles to ensure the vehicle stays in the correct lane through path planning and adjustments. Sensors are the primary source of lane perception data for the vehicle, while vehicle-to-vehicle (V2V) communication is a crucial means for vehicles to share driving status and perception information. This decentralized vehicle platooning model allows each vehicle to act as an independent decision-maker, collectively achieving coordinated control. The V2V system can provide necessary redundancy and information supplementation in the event of vehicle malfunctions, thereby improving the robustness and flexibility of the entire system, which is particularly valuable for applications such as high-speed commercial vehicle platooning and long-haul logistics collaborative cruising.

[0003] However, existing lane-keeping methods primarily rely on individual vehicle intelligence, where each vehicle independently assesses the deviation between sensor data and a pre-set desired trajectory, and then implements corrective measures. When environmental conditions are unfavorable, such as rain, fog, strong sunlight, or obstructions, or when sensors malfunction due to hardware damage or calibration failure, these individual vehicle intelligence methods may fail to accurately identify whether the sensor information is biased or completely erroneous. This can lead to the vehicle misjudging its current position and direction, resulting in lane departure or lateral control failure. In such cases, the consequences can range from minor driving disturbances to serious accidents such as chain collisions and traffic congestion, posing a significant risk to traffic safety.

[0004] More importantly, when a vehicle's perception deviates, the vehicle itself cannot distinguish whether the deviation stems from sensor malfunction or a genuine deviation, and can only blindly execute corrective control. In a decentralized vehicle swarm scenario, this single-vehicle decision-making lacking consensus verification leads to inconsistent perceptions of lane references among vehicles, making it difficult to guarantee the stability of coordinated vehicle swarm driving and failing to leverage the inherent advantages of multi-node information complementarity and enhanced safety redundancy within the vehicle swarm.

[0005] Therefore, to address the aforementioned technical issues, it is necessary to provide a method and device for detecting lane keeping anomalies in a decentralized vehicle group environment.

[0006] The information disclosed in this background section is intended only to enhance the understanding of the overall background of the invention and should not be construed as an admission or in any way implying that the information constitutes prior art known to those skilled in the art. Summary of the Invention

[0007] The purpose of this invention is to provide a method and apparatus for detecting lane keeping anomalies in a decentralized vehicle group environment, which can solve the technical problems mentioned in the background art.

[0008] To achieve the above objectives, a specific embodiment of the present invention provides the following technical solution:

[0009] A method for detecting lane keeping anomalies in a decentralized vehicle group environment includes the following steps:

[0010] S1. Collect the vehicle's visual lane line parameters, receive the neighbor visual lane line parameters broadcast by neighbor vehicles in the decentralized vehicle group through the V2V communication module, and convert the neighbor visual lane line parameters from the neighbor vehicle coordinate system to the vehicle's coordinate system to obtain the neighbor mapped lane line set.

[0011] S2. Input the vehicle's visual lane line parameters and the neighbor's mapped lane line set into a distributed consensus filter without a central fusion node, and generate a local consensus lane line of the vehicle group consensus through interaction with neighbor vehicle data;

[0012] S3. Calculate the first lateral deviation of the vehicle based on the local consensus lane line of the vehicle, calculate the second lateral deviation of the vehicle based on the visual lane line parameters of the vehicle, and simultaneously calculate the third lateral deviation set of each neighboring vehicle relative to the local consensus lane line of the vehicle.

[0013] S4. Calculate the absolute value of the difference between the first lateral deviation and the second lateral deviation to obtain the perceived deviation degree of the vehicle.

[0014] The neighbor control deviation set is obtained based on the third lateral deviation set;

[0015] The vehicle's perceived deviation and neighbor's control deviation sets are compared with corresponding preset thresholds to determine the anomaly type.

[0016] S5. Generate and output an anomaly detection signal that identifies the anomaly source category based on the anomaly type.

[0017] In one or more embodiments of the present invention, the vehicle visual lane line parameters are lane line polynomial coefficients obtained by semantic segmentation and lane line fitting of road images captured by vehicle-mounted cameras. The lane line polynomial coefficients are used to represent the lateral offset and heading angle deviation of the lane line relative to the vehicle coordinate system.

[0018] In one or more embodiments of the present invention, the neighbor visual lane line parameters are transformed from the neighbor vehicle coordinate system to the current vehicle coordinate system, specifically as follows:

[0019] Based on the positioning coordinates and heading angles broadcast by neighboring vehicles, and the positioning coordinates and heading angles of this vehicle, calculate the relative rotation matrix and translation vector between this vehicle and the corresponding neighboring vehicle.

[0020] The neighbor visual lane line parameters are mapped to the vehicle coordinate system to obtain the neighbor mapped lane line set.

[0021] In one or more embodiments of the present invention, generating local consensus lane lines for a vehicle through a distributed consensus filter without a centralized fusion node specifically includes the following steps:

[0022] S21. Use the vehicle's visual lane line parameters at the current moment as the initial state estimate of the distributed consensus filter;

[0023] S22. Broadcast the state estimate of the current iteration cycle of this vehicle to all neighboring vehicles, and receive the state estimates of the current iteration cycle sent by the neighboring vehicles to obtain the set of neighboring state estimates;

[0024] S23. Based on the set of neighbor state estimates and the state estimate of the vehicle in the current iteration cycle, calculate the state estimate of the vehicle in the next iteration cycle according to the preset consistency update rule.

[0025] S24. Determine whether the difference between the state estimates of this vehicle and neighboring vehicles meets the preset convergence condition. If it does, then use the state estimate of the current iteration period as the local consensus lane line output of this vehicle.

[0026] If the condition is not met, the state estimate of the next iteration cycle will be used as the new state estimate of the current iteration cycle, and the process will return to step S22.

[0027] In one or more embodiments of the present invention, the specific steps for receiving the state estimate value sent by a neighboring vehicle are as follows:

[0028] This vehicle pre-generates and maintains a neighbor communication list, which contains unique identifiers of neighboring vehicles that have established stable V2V communication links with this vehicle.

[0029] This vehicle only receives state estimates from neighboring vehicles whose unique identifiers are listed in the neighbor communication list.

[0030] In one or more embodiments of the present invention, the preset consistency update rule is as follows: based on the state estimate of the current iteration cycle of the vehicle and all state estimates in the set of neighboring state estimates, a weighted fusion calculation is performed. The weighting coefficients are dynamically adjusted according to the communication link quality between the vehicle and the corresponding neighboring vehicle, and all weighting coefficients satisfy the normalization condition. The fused value is used as the state estimate of the next iteration cycle.

[0031] In one or more embodiments of the present invention, each of the lateral deviations in the first lateral deviation, the second lateral deviation, and the third lateral deviation set is the lateral offset of the vehicle body center point relative to the lane center line of the corresponding lane line.

[0032] In one or more embodiments of the present invention, the vehicle's perceived deviation degree and the neighbor's control deviation degree set are compared with corresponding preset thresholds to determine the anomaly type, including:

[0033] If the perceived deviation of the vehicle is greater than the preset perceived deviation threshold, and all elements in the neighbor control deviation set are less than the preset control deviation threshold, then the abnormality type is determined to be a visual perception abnormality of the vehicle.

[0034] If at least one element in the neighbor control deviation set is greater than the preset control deviation threshold, and the vehicle's perceived deviation is less than the preset perceived deviation threshold, then the abnormality type is determined to be the control execution abnormality of the corresponding neighbor vehicle.

[0035] If the perceived deviation of the vehicle is greater than the preset perceived deviation threshold, and at least one element in the neighbor control deviation set is greater than the preset control deviation threshold, then the abnormality type is determined to be road geometry change.

[0036] In one or more embodiments of the present invention, the corresponding preset threshold includes a preset perception deviation threshold and a preset control deviation threshold, both of which are dynamically adjusted according to the current driving speed of the vehicle group, road curvature, and road surface adhesion coefficient.

[0037] To achieve the above objectives, a specific embodiment of the present invention provides the following technical solution:

[0038] A lane keeping anomaly detection device in a decentralized vehicle group environment includes:

[0039] The data acquisition module is used to collect the GPS coordinates, heading angle, visual lane line parameters and V2V communication signal-to-noise ratio of the vehicle, and receive the GPS coordinates, heading angle and visual lane line parameters of all one-hop neighbor vehicles through V2V communication, and synchronously transmit them to the coordinate alignment module.

[0040] The coordinate alignment module calculates the relative rotation matrix and translation vector between the vehicle and neighboring vehicles based on the received data, maps the visual lane line parameters of the neighboring vehicles to the coordinate system of the vehicle, forms a neighboring mapped lane line set, and transmits it to the distributed consensus estimation module.

[0041] The distributed consensus estimation module uses a distributed consensus filter. It takes the vehicle's visual lane line parameters and the neighbor's mapped lane line set as inputs, dynamically allocates weighting coefficients based on the V2V communication signal-to-noise ratio, and iteratively updates the lane line state estimate until the convergence condition is met, thus generating the vehicle's local consensus lane line.

[0042] The lateral deviation calculation module calculates the multi-dimensional lateral deviation of the vehicle and each of its one-hop neighboring vehicles, forms a deviation vector set, and transmits it to the anomaly determination module.

[0043] The anomaly determination module presets a perception deviation threshold and a control deviation threshold. The threshold is dynamically adjusted according to the vehicle speed of the vehicle group. The anomaly type is determined by comparing the deviation with the threshold.

[0044] The signal output module outputs and broadcasts the corresponding abnormal signal based on the judgment result; if there is no abnormality, the detection process continues to be executed.

[0045] Compared with the prior art, the beneficial effects of the present invention are: to identify vehicle sensor information deviation, to distinguish whether the deviation is caused by sensor malfunction or actual vehicle deviation, to avoid blind correction, to solve the problems of vehicle position and direction misjudgment, lane departure, and lateral control failure caused by perception deviation, and to reduce traffic safety risks.

[0046] Meanwhile, by generating consensus lane references for vehicle groups through a decentralized distributed solution, the problem of inconsistent lane reference recognition among vehicles in decentralized vehicle group scenarios is solved, ensuring the stability of vehicle group collaboration and leveraging the advantages of multi-node information complementarity and security redundancy. It does not rely on central nodes, roadside equipment or high-precision maps, and can complete collaborative detection solely through vehicle-to-vehicle communication, significantly improving system robustness and vehicle group driving safety, and is suitable for core scenarios such as high-speed commercial vehicle platooning and collaborative cruise of trunk logistics. Attached Figure Description

[0047] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0048] Figure 1 This is a flowchart of a lane keeping anomaly detection method in a decentralized vehicle group environment according to an embodiment of the present invention.

[0049] Figure 2 This is a flowchart illustrating the data acquisition and coordinate system alignment process of a lane keeping anomaly detection method in a decentralized vehicle group environment according to an embodiment of the present invention.

[0050] Figure 3 This is a flowchart of the distributed consensus initialization and iterative update process of a lane keeping anomaly detection method in a decentralized vehicle group environment according to an embodiment of the present invention.

[0051] Figure 4 This is a flowchart illustrating the distributed consensus convergence verification and boundary handling process of a lane keeping anomaly detection method in a decentralized vehicle group environment according to an embodiment of the present invention.

[0052] Figure 5 This is a flowchart illustrating the lateral deviation calculation and anti-false judgment filtering process of a lane keeping anomaly detection method in a decentralized vehicle group environment according to an embodiment of the present invention.

[0053] Figure 6 This is a flowchart illustrating the core determination of anomaly types in a lane keeping anomaly detection method in a decentralized vehicle group environment according to an embodiment of the present invention.

[0054] Figure 7 This is a flowchart of the anomaly handling and signal output process in a lane keeping anomaly detection method in a decentralized vehicle group environment according to an embodiment of the present invention. Detailed Implementation

[0055] To enable those skilled in the art to better understand the technical solutions in this disclosure, the technical solutions in the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this disclosure, and not all embodiments. Based on the embodiments in this disclosure, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this disclosure.

[0056] Example 1:

[0057] like Figures 1-7 As shown in one embodiment of the present invention, a lane keeping anomaly detection method in a decentralized vehicle group environment is described. This method triggers data acquisition every 500ms to 50ms after the vehicle activates its autonomous driving mode and lane keeping function is activated. The data acquisition frequency can be dynamically adjusted according to the vehicle speed. The specific steps of the lane keeping anomaly detection method in a decentralized vehicle group environment include:

[0058] S1. Collect the vehicle's visual lane line parameters. The vehicle's visual lane line parameters are fitted using a quadratic polynomial to characterize the geometric shape of the lane line in the vehicle's coordinate system. The mathematical model is as follows:

[0059]

[0060] In the formula, This represents the lateral offset of the lane line at the longitudinal distance x in the vehicle's coordinate system. This represents the longitudinal distance in the vehicle's coordinate system, with the center point of the vehicle body as the origin and the direction the vehicle is facing as the positive x-axis. It is the quadratic term coefficient of the lane line in the vehicle's vision, which is expressed as the rate of change of the lane line curvature and is used to describe the curvature of the curve; It is the linear coefficient of the lane line as seen by the vehicle, which represents the deviation of the lane line from the vehicle's heading angle. This is the constant term coefficient of the visual lane line of this vehicle, which represents the initial lateral offset of the lane line relative to the center point of the vehicle body. A positive number indicates that the lane line is on the left side of the vehicle, and a negative number indicates that it is on the right side.

[0061] in, , The angle between the lane line and the direction in which the vehicle is facing. The lateral offset rate of a lane line is mathematically equal to the ratio of the lateral change in the lane line to the longitudinal distance. In simpler terms... This can be equivalent to the lateral offset rate of the lane line, mathematically equal to the ratio of the lateral change of the lane line to the longitudinal travel distance. It visually represents the lateral offset of the lane line relative to the vehicle by meters for every meter the vehicle travels forward. This conversion transforms the angular heading deviation into linear coefficients that can be directly used in the lane line polynomial, achieving a one-to-one correspondence between the lane line geometric features and the fitting coefficients.

[0062] The above mathematical model quantifies the core features of lane lines. , , These correspond to the key geometric properties of the lane lines, providing a clear data carrier for subsequent calculations. Furthermore, the mathematical model ensures consistency in vehicle group data interaction; both the vehicle and its neighbors use the same form of quadratic polynomial fitting for lane lines, guaranteeing a unified format for lane line parameters across multiple vehicles.

[0063] After the visual lane line parameters of this vehicle are fitted with a quadratic polynomial, the visual lane line parameters of neighboring vehicles in the decentralized vehicle group are received through the V2V communication module, and the visual lane line parameters of neighboring vehicles are transformed from the coordinate system of neighboring vehicles to the coordinate system of this vehicle.

[0064] Specifically, both the neighbor vehicle coordinate system and the vehicle's coordinate system are right-handed rectangular coordinate systems, with the origin at the center point of each vehicle body. The x-axis represents the vehicle's front orientation, the y-axis represents the vehicle's lateral orientation, and the z-axis is perpendicular to the ground. The core of the coordinate transformation is to map the neighbor's visual lane line parameters from the neighbor vehicle coordinate system to the vehicle's coordinate system through a relative rotation matrix and a translation vector, thereby eliminating the position and heading differences between the two vehicles.

[0065] First, calculate the relative rotation matrix based on the GPS coordinates and heading angles of this vehicle and neighboring vehicles. Translation vector :

[0066] Let the GPS coordinates of this vehicle be... The data is collected in real time by the vehicle's onboard GPS / BeiDou positioning module. The geodetic coordinate system, with due north as the x-axis and due east as the y-axis, is a fixed absolute coordinate system used to represent the vehicle's absolute position on the real road; the heading angle is... The coordinates, collected by the vehicle's inertial measurement unit, are expressed as the angle between the vehicle's front and the true north direction of the geodetic coordinate system, with clockwise being positive. This angle is used to achieve the projection transformation from absolute position difference to relative position and is the core transformation coefficient of the formula. The GPS coordinates of neighboring vehicles are... The coordinates are broadcast from neighboring vehicles via V2V communication to this vehicle, and are sourced from the same GPS coordinates as this vehicle, ensuring the accuracy of the position difference calculation. The heading angle is... The relative heading angle between the two vehicles is .

[0067] Relative rotation matrix The core function is to rotate the lateral and longitudinal coordinate components of the neighboring vehicle coordinate system around the origin by the relative heading angle. This eliminates the coordinate system heading deviation between the vehicle and neighboring vehicles caused by their different frontal orientations, achieving heading alignment between the two vehicle coordinate systems. Relative rotation matrix Represented as:

[0068]

[0069] In the formula, the first column and Used to process the longitudinal coordinate components in the neighboring vehicle coordinate system, the second column. and It is used to process the horizontal coordinate component. Through the combined operation of "cosine multiplied by vertical, negative sine multiplied by horizontal" and "sine multiplied by vertical, cosine multiplied by horizontal", the rotation transformation of the vertical and horizontal coordinates can be completed simultaneously, ensuring the rotation synchronization of the coordinate components.

[0070] This rotation matrix is ​​constructed based on the principle of rotation transformation in a two-dimensional Cartesian coordinate system. It is used to eliminate the heading angle deviation between the vehicle and neighboring vehicles, uniformly transforming the coordinate components in the neighboring vehicle's coordinate system to the heading reference in the vehicle's coordinate system. The first column of the matrix corresponds to the rotation transformation weight of the longitudinal coordinate in the neighboring coordinate system, and the second column corresponds to the rotation transformation weight of the lateral coordinate in the neighboring coordinate system. Through rotation transformation, heading alignment of the longitudinal and lateral coordinates can be completed simultaneously, avoiding false offsets in lane line parameters caused by differences in the heading of the two vehicles. This rotation matrix can correct heading deviations solely through mathematical calculations, requiring no additional hardware support. It has low computational load, high real-time performance, and is fully adaptable to the real-time coordinate transformation requirements in high-speed platooning scenarios, serving as the core foundation for accurate mapping of neighboring lane line parameters.

[0071] In decentralized vehicle platooning scenarios, there are often slight differences in the heading of neighboring vehicles compared to the vehicle itself. Without rotation transformation, the lane line parameters perceived by neighboring vehicles will show a false shift due to heading deviation. The rotation matrix can accurately correct heading deviation through pure mathematical operations, requiring no additional hardware support. It has low computational load, high real-time performance, and is fully adaptable to the coordinate transformation requirements in high-speed vehicle platooning scenarios. It is also the foundation for accurately mapping neighboring lane line parameters to the vehicle's coordinate system.

[0072] Position difference translation vector This is the core formula for the two-dimensional position projection transformation from the geodetic coordinate system to the vehicle body coordinate system, along with the relative rotation matrix. To form a complete coordinate transformation system, the position difference translation vector Convert the absolute GPS position difference between the two vehicles into relative longitudinal and lateral distances centered on your own vehicle. , , forming a translation vector ,in:

[0073]

[0074] In the formula, , It is expressed as the absolute difference in north and east directions between this vehicle and the neighboring vehicle in the geodetic coordinate system, which is used to reflect the positional distance between the two vehicles on the real road; This indicates the longitudinal distance of a neighboring vehicle relative to the current vehicle, perfectly matching the x-axis of the current vehicle's coordinate system: a positive value indicates that the neighboring vehicle is in front of the current vehicle, and a negative value indicates that the neighboring vehicle is behind the current vehicle. The absolute value of the value represents the distance between the two vehicles in the front-to-back direction. The lateral distance of a neighboring vehicle relative to this vehicle is perfectly matched with the y-axis of this vehicle's coordinate system: a positive value indicates that the neighboring vehicle is to the left of this vehicle, and a negative value indicates that it is to the right of this vehicle. The absolute value of the value represents the distance between the two vehicles in the left-right direction.

[0075] It is the difference in the absolute northward position of the two vehicles multiplied by the cosine of the heading angle of the vehicle itself, i.e. The projection component of the northward position difference along the longitudinal axis of the vehicle is obtained. It is the difference in the absolute eastward position of the two vehicles multiplied by the sine of the vehicle's heading angle, i.e. The projection component of the eastward position difference onto the longitudinal axis of this vehicle is obtained. The sum of the two projection components is the longitudinal distance of the neighboring vehicle relative to this vehicle. The function of this formula is to convert the absolute position difference between the two vehicles in the geodetic coordinate system into the relative distance in the front-rear direction with the vehicle body as the reference, thereby realizing the projection conversion from absolute position to relative position.

[0076] It is the projection of the absolute position difference to the lateral distance of the vehicle, and The calculation logic is similar, but a negative sign is added to adjust the direction. It is the projection component of the northward position difference onto the lateral axis of this vehicle. It is the projected component of the eastward position difference onto the lateral axis of this vehicle; the sum of the two is the lateral distance of the neighboring vehicle relative to this vehicle. The formula combines the absolute distances of the two vehicles in the north-south and east-west directions into a relative distance in the left-right direction.

[0077] In calculation and In the formula, the trigonometric functions are used for direction calibration. The vehicle's heading angle determines the angle between the vehicle's relative coordinate system and the geodetic coordinate system. and The proportional allocation ensures that the absolute position difference can be accurately projected onto the longitudinal and lateral axes of the vehicle, avoiding positional deviations caused by inconsistencies between the vehicle's driving direction and the geodetic coordinate system.

[0078] In calculation and In the formula, the GPS coordinates of the neighbor are obtained through V2V communication. Combined with the vehicle's positioning and heading data, the dynamic relative position can be calculated in real time without relying on the central node, roadside equipment or high-precision map. The amount of calculation is small and can meet the update needs in high-speed scenarios.

[0079] Secondly, the neighbor's visual lane line parameters in the neighbor's vehicle coordinate system are converted to the current vehicle's visual lane line parameters using a quadratic polynomial of the same form, with coefficients respectively. , , Through the above rotation matrix Translation vector Mapping to the vehicle's coordinate system yields the mapped lane lines of neighboring vehicles. After simplification, its polynomial coefficients The core formula is as follows:

[0080]

[0081] In the formula, , , These are the coefficients of the quadratic, linear, and constant terms of the neighboring visual lane lines, respectively, and their meanings are consistent with the parameters of the vehicle's visual lane lines. This is represented as the lateral offset of the neighboring mapped lane line in the vehicle's coordinate system; , , These are the coefficients of the quadratic, linear, and constant terms of the lane lines mapped from the neighboring vehicle, respectively, and their meanings are consistent with the lane line parameters of the vehicle's vision.

[0082] Specifically, This formula represents the rate of change of lane curvature. It uses the cosine of the relative heading angle to rotate and correct the curvature coefficient perceived by neighboring vehicles, eliminating the curvature perception projection deviation caused by the difference in the orientation of the two vehicles, and ensuring that the curvature of the mapped lane line is consistent with the actual curvature of the road.

[0083] This formula represents the heading angle deviation of the lane lines. It is based on the rotation transformation rules of the two-dimensional coordinate system and performs rotation correction on the heading deviation coefficients of adjacent lane lines. The rotational projection term for the neighbor's heading deviation; The inherent heading compensation term caused by the rotation of the coordinate system is combined with the coordinate system to achieve the coordinate system alignment of the lane line heading trend, ensuring that the mapped heading characteristics match the reference of the vehicle's coordinate system.

[0084] The constant term coefficient represents the initial lateral offset of the lane line. Based on the rotation transformation correction, this formula superimposes the translation compensation term of the relative position of the two vehicles to eliminate the positional difference between the two vehicles and ensure that the initial lateral offset of the mapped lane line is completely consistent with the representation of the actual position of the road in the vehicle coordinate system.

[0085] The formula mainly addresses the issue of inconsistent coordinate systems among multiple vehicles in a decentralized vehicle group. Since each vehicle's coordinate system is based on its own body, there are differences in heading and position. The lane line parameters of neighboring vehicles cannot be directly used for calculation of this vehicle. It uses the rotation matrix and translation vector mentioned earlier to unify the lane line parameters of neighboring vehicles into the coordinate system of this vehicle, ensuring accurate parameters and adapting to the dynamic changes of the vehicle group.

[0086] Finally, collect the lane line parameters of all one-hop neighbor vehicles to form a neighbor map lane line set. In the formula, k is the number of neighboring vehicles in a hop, and k≥1.

[0087] Once the neighboring lane line set has been aligned with the vehicle's coordinate system, providing a foundation for multi-vehicle data interaction and collaborative analysis, the next step is to proceed to step S2 in order to achieve unified recognition of lane lines among vehicle groups.

[0088] S2. Input the vehicle's visual lane line parameters and the neighboring mapped lane line sets into a distributed consensus filter without a central fusion node. Generate a local consensus lane line for the vehicle group through data interaction with neighboring vehicles. Specifically, the vehicle broadcasts its lane line state estimate for the current iteration cycle to all one-hop neighboring vehicles, and simultaneously receives the state estimates for the current iteration cycle from all neighboring vehicles. After receiving the vehicle's state estimate, neighboring vehicles synchronously feed back their own state estimates to the vehicle, forming a two-way interaction. Vehicles that have not established a stable V2V communication link do not participate in the iterative interaction.

[0089] The state variables of a distributed consensus filter are defined as lane line polynomial coefficient vectors. Where i represents the vehicle number and t represents the iteration period. These are the coefficients of the quadratic, linear, and constant terms of the lane line for the i-th vehicle in the t-th iteration.

[0090] In summary, the visual lane line parameters of this vehicle are lane line polynomial coefficients obtained by semantic segmentation and lane line fitting of road images captured by the on-board camera. The lane line polynomial coefficients include at least constant term coefficients and first-order term coefficients, which respectively represent the lateral offset and heading angle deviation of the lane line relative to the vehicle's coordinate system.

[0091] In S2, a local consensus lane line for this vehicle is generated through a distributed consensus filter without a centralized fusion node, specifically including the following steps:

[0092] S21. Use the vehicle's current visual lane line parameters as the initial state estimate of the distributed consensus filter, i.e. Neighboring vehicles simultaneously use their own visual lane line parameters as initial state estimates. ;

[0093] S22. Estimate the state of the vehicle in the current iteration cycle. Broadcast to all neighboring vehicles and receive state estimates for the current iteration period from neighboring vehicles. The set of neighbor state estimates is obtained. ;

[0094] In summary, the specific steps for transforming the neighbor's visual lane line parameters from the neighbor's vehicle coordinate system to the vehicle's coordinate system are as follows: based on the positioning coordinates and heading angles broadcast by the neighbor's vehicle and the positioning coordinates and heading angles of the vehicle, calculate the relative rotation matrix and translation vector between the vehicle and the corresponding neighbor vehicle; map the neighbor's visual lane line parameters to the vehicle's coordinate system to obtain the neighbor's mapped lane line set.

[0095] This vehicle pre-generates and maintains a neighbor communication list, and the dynamic maintenance rules for the list throughout its entire lifecycle are as follows:

[0096] The vehicle's unique identifier will only be added to the neighbor communication list if the signal-to-noise ratio of the V2V communication link between the neighbor vehicle and this vehicle is greater than or equal to the preset communication threshold (initial value -95dBm) for three consecutive detection cycles, and if valid positioning coordinates, heading angle and visual lane line parameters are continuously broadcast.

[0097] If a neighboring vehicle already in the list does not receive its state estimate for two consecutive iterations, or if the communication signal-to-noise ratio is lower than the preset communication threshold for five consecutive detection cycles, it is determined that the communication is interrupted or the vehicle has driven out of the communication range. It will be permanently removed from the neighbor communication list and its broadcast data will no longer be received until the admission rules are met again.

[0098] If a neighbor vehicle's state estimate is not received occasionally during a single iteration cycle, it will only be temporarily removed from the set of neighbor state estimates for this iteration, without triggering permanent removal from the list, thus avoiding a decrease in consensus accuracy caused by occasional data packet loss.

[0099] It is worth noting that after each iteration cycle begins, the state broadcast and reception are completed first, and then the consistency update calculation is performed. If the state estimate of a neighbor vehicle is not received, the temporary removal rule mentioned above is executed, and the vehicle will not participate in this iteration. If the vehicle triggers temporary removal continuously, the permanent removal rule is executed simultaneously, and the neighbor communication list is updated.

[0100] S23. Based on the neighbor state estimate set and the current iteration cycle state estimate of the vehicle, calculate the next iteration cycle of the vehicle according to the preset consistency update rule. State estimate The specific mathematical model is as follows:

[0101]

[0102] This formula is the core update rule for consensus estimation of lane lines in a vehicle group using a distributed consensus filter. Its core significance lies in the fact that it eliminates the need for a central fusion node. It achieves iterative convergence of lane line parameters for multiple vehicles by weighted fusion of the lane line state estimates of the current vehicle and its one-hop neighbor vehicles. It ensures the accuracy of consensus estimation through dynamically allocated weighting coefficients and can adapt to dynamic topology changes in the vehicle group. Furthermore, it connects the state initialization and neighbor state interaction mentioned earlier, and the output state estimate for the next cycle is used for convergence judgment, providing a core benchmark for subsequent steps.

[0103] The weighting coefficients satisfy the normalization condition:

[0104]

[0105] Weighting coefficient dynamic adjustment rules: The weighting coefficients are allocated based on the quality of the V2V communication link between the vehicle and its neighboring vehicles, following the normalization principle; the better the communication link quality, the larger the weighting coefficient of the corresponding vehicle, that is, the higher the reliability of the state estimates of the vehicle itself and its neighboring vehicles, the greater the weight assigned.

[0106] In vehicle corps scenarios without a central node, the V2V communication link between a vehicle and its neighbors is susceptible to factors such as high-speed driving and environmental interference, leading to differences in communication quality and consequently affecting the reliability of lane state estimates transmitted by neighboring vehicles. Simultaneously, the normalization principle is set to avoid distortion in the consistency update calculation caused by the sum of weighting coefficients not being 1, ensuring that the iterative process conforms to mathematical logic and guaranteeing the convergence of the consensus estimate. Furthermore, through differentiated weight allocation, the impact of a single vehicle's communication anomaly on the overall vehicle corps consensus estimate is effectively avoided. Even if some neighboring vehicles have poor communication quality, their interference can be reduced by setting low weights, ensuring the stable operation of the anomaly detection scheme and meeting the requirements of decentralized vehicle corps to resist single points of failure.

[0107] In summary, the preset consistency update rule is as follows: Based on the current state estimate of the vehicle in the current iteration cycle and all state estimates in the neighbor state estimate set, a weighted fusion calculation is performed. The weighting coefficients are dynamically adjusted according to the communication link quality between the vehicle and the corresponding neighbor vehicle, and all weighting coefficients meet the normalization condition. The fused value is then used as the state estimate for the next iteration cycle.

[0108] The consistency update rule adopts a weighted fusion strategy. The weighting coefficients of the vehicle and its neighboring vehicles are dynamically allocated according to the quality of the V2V communication link. The better the communication quality, the higher the corresponding weight. The sum of all weights is 1. Consistent convergence of lane line state estimation of the vehicle group is achieved through iterative weighted fusion.

[0109] In S22, the specific steps for receiving state estimates sent by neighboring vehicles are as follows: This vehicle pre-generates and maintains a neighbor communication list, which contains unique identifiers of neighboring vehicles that have established stable V2V communication links with this vehicle; This vehicle only receives state estimates from neighboring vehicles corresponding to the unique identifiers in the neighbor communication list.

[0110] After S23 is completed, proceed to S24 to determine whether the difference between the state estimates of the current vehicle and its neighboring vehicles meets the preset convergence condition. That is, the maximum deviation between the state estimate of the current vehicle in the current iteration cycle and the state estimates of all one-hop neighboring vehicles in the current iteration cycle must be less than the preset convergence threshold. The degree of deviation between the state estimates of the two vehicles is measured by the L2 norm. The smaller this distance, the more consistent the two vehicles' perception of the lane lines.

[0111] It is worth noting that the preset convergence threshold is in meters (m). In this embodiment, the preset value is 0.02, which is used to determine whether the lane line state estimates of each vehicle in the vehicle group have reached a consensus. It can be dynamically adjusted according to the actual driving scenario. In high-speed driving scenarios, the threshold can be appropriately reduced to improve consensus accuracy. The two-vector Euclidean distance is used to calculate the degree of deviation between the state estimates of the current vehicle and each neighboring vehicle. The unit is meters (m). The smaller the distance, the more consistent the two vehicles' perception of the geometric shape and position of the lane line, and the smaller the deviation.

[0112] If the above convergence conditions are met, the estimated state of the vehicle in the current iteration period will be used as the local consensus lane line of the vehicle. The consensus lane line adopts a quadratic polynomial form consistent with the visual lane line of the vehicle. The coefficients of its quadratic, linear, and constant terms are completely consistent with the corresponding coefficients of the estimated state of the vehicle in the current iteration period, and the meaning of the coefficients is also consistent with the parameters of the visual lane line of the vehicle.

[0113] If the convergence condition is not met, the estimated state value of the vehicle in the next iteration cycle will be used as the new current state estimate. The process will return to step S22 to re-execute the neighbor state interaction and consistency update calculation until the convergence condition is met. If the convergence condition is not met even after the number of iterations reaches the preset maximum value, a communication anomaly alarm will be triggered. The vehicle's visual lane line will be used as a temporary driving reference to ensure the safety of the vehicle group.

[0114] For the boundary scenario with no neighboring vehicles, i.e., k=0, where this vehicle has not established a stable V2V communication link with any other vehicle, the following processing logic is executed:

[0115] Skip the distributed consensus lane line generation process in step S2 and directly use the vehicle's visual lane line parameters as the only lane reference; when performing the deviation calculation in steps S3-S4, the first lateral deviation and the second lateral deviation have the same value, the vehicle's perceived deviation is always 0, and the vehicle's visual perception abnormality judgment is not triggered.

[0116] Temporarily disable the neighbor vehicle control anomaly detection function, and only retain the lane keeping lateral deviation monitoring function of this vehicle. When the lateral deviation of this vehicle relative to its own visual lane line exceeds the preset single vehicle safety threshold, trigger the single vehicle lane departure audible and visual alarm.

[0117] Continuously monitor V2V communication broadcasts. When a neighboring vehicle that meets the access rules is detected, immediately resume the complete vehicle group collaboration anomaly detection process and re-execute steps S1 to S5.

[0118] S3. Calculate the first lateral deviation of the vehicle based on the local consensus lane line, and calculate the second lateral deviation of the vehicle based on the vehicle's visual lane line parameters. Simultaneously, calculate the third lateral deviation set of each neighboring vehicle relative to the local consensus lane line. That is, using the center point of the vehicle body as the reference, calculate the lateral deviations of the vehicle relative to the consensus lane line, the vehicle relative to its own visual lane line, and neighboring vehicles relative to the consensus lane line, and create a deviation vector set to provide data support for future anomaly tracing.

[0119] Among them, each of the lateral deviations in the first lateral deviation, second lateral deviation, and third lateral deviation sets is the lateral offset of the vehicle body center point relative to the lane center line of the corresponding lane line.

[0120] Specifically, the coordinates of the center point of the vehicle body in the vehicle coordinate system are defined as follows: ,in, =0 is the longitudinal reference point. =0, which is the lateral reference point; the coordinates of the center point of the neighboring vehicle in this vehicle coordinate system are: The relative translation vector from step S1 It is confirmed that, among them, = , = .

[0121] First lateral deviation The lateral deviation of this vehicle relative to the lane markings of the local consensus area will affect the longitudinal coordinate of the vehicle's center point. Substituting 0 into the local consensus lane line polynomial of this vehicle, we calculate the lateral offset of the consensus lane line at that position, and its relationship to the lateral coordinates of the vehicle's center point. The difference that equals 0 is the first lateral deviation:

[0122]

[0123] In the formula, the first lateral deviation This indicates the actual degree of deviation of the vehicle from the lane line agreed upon by the group of vehicles; >0 indicates that the vehicle is to the left of the center line of the consensus lane. <0 indicates that it is on the right. =0 indicates that the vehicle is traveling entirely along the center line of the consensus lane.

[0124] First lateral deviation Consistent with the calculation logic of the first lateral deviation, the first lateral deviation This indicates the degree to which the vehicle deviates from its perceived lane line; the meaning of the value is the same as... Consistent, with the only benchmark being its own visual lane lines.

[0125] Third lateral deviation set : The lateral deviation of each neighboring vehicle relative to the locally agreed lane line of the vehicle itself. In the calculation, the longitudinal coordinate of the center point of the j-th neighboring vehicle is used. Substituting the local consensus lane line polynomial into the equation, the lateral offset of the consensus lane line at that location is calculated, and compared with the lateral coordinates of the center point of the neighboring vehicle. The difference is the third lateral deviation of the j-th neighboring vehicle:

[0126]

[0127] In the formula, Let be the third lateral deviation of the j-th neighboring vehicle, used to represent the actual degree of deviation of that neighboring vehicle from the locally agreed lane line relative to the vehicle itself. The meaning of the value and Consistent. It contains lateral deviation data for all one-hop neighboring vehicles, which is used to subsequently determine the control status of neighboring vehicles.

[0128] Finally, create the deviation vector set. , as input data for S4.

[0129] S4. Calculate the absolute value of the difference between the first lateral deviation and the second lateral deviation to obtain the perceived deviation degree of the vehicle. The vehicle perception deviation is used to represent the degree of deviation between the vehicle's own visual perception and the consensus of the vehicle group. The larger the difference, the higher the probability of a perception anomaly. The vehicle perception deviation quantifies the degree of deviation between the vehicle's own perception and the consensus of the vehicle group, determining whether the vehicle's visual perception is abnormal; and it controls the deviation set through neighboring vehicles. The absolute value of each element in the third lateral deviation set is used, i.e. This indicates the degree to which each neighboring vehicle deviates from the consensus lane line of the vehicle group. The larger the absolute value, the higher the probability of abnormal control of the neighboring vehicle.

[0130] This embodiment presets two core thresholds to determine abnormal perception of the vehicle itself and abnormal control of neighboring vehicles, respectively. The initial preset perception deviation threshold is 0.15m, and the initial preset control deviation threshold is 0.2m. These two thresholds are not fixed values ​​and can be dynamically adjusted based on actual scenario parameters such as the current vehicle speed, road curvature, and road surface adhesion coefficient. The adjustment principle is that the higher the vehicle speed, the larger the threshold setting, thus adapting to the deviation tolerance in high-speed driving scenarios and ensuring the rationality of the judgment logic.

[0131] The current speed of the vehicle group is collected by the vehicle's IMU module. In actual calculation, the average speed of all vehicles in the group is taken to ensure the rationality of the threshold adjustment and to adapt to the overall driving status of the vehicle group.

[0132] Simply put, the main purpose of step S4 is to compare the vehicle's perceived deviation degree and the neighbor's control deviation degree set with the corresponding preset thresholds to determine the anomaly type.

[0133] Among them, the vehicle's perceived deviation and neighbor's control deviation sets are compared with corresponding preset thresholds to determine the anomaly types, including:

[0134] The vehicle's perception deviation exceeds the preset perception deviation threshold, and the control deviation of all neighboring vehicles does not exceed the preset control deviation threshold. This means that the vehicle's perception deviates significantly from the consensus of the vehicle group, but none of the neighboring vehicles have shown any deviation. This indicates that the root cause of the deviation is an abnormality in the vehicle's vision sensor, rather than abnormal vehicle group driving or changes in road geometry.

[0135] There exists at least one neighboring vehicle whose control deviation exceeds a preset control deviation threshold, while the vehicle's perceived deviation does not exceed a preset perception deviation threshold. This means that the vehicle's perception is consistent with the consensus of the vehicle group, but one or more neighboring vehicles are seriously deviating from their lanes. This indicates that the root cause of the deviation is an abnormal execution of lane keeping control by that neighboring vehicle, such as a control system failure or a person's illegal lane change.

[0136] The vehicle's perception deviation exceeds the preset perception deviation threshold, and simultaneously, at least one neighboring vehicle's control deviation exceeds the preset control deviation threshold. The significant deviation between the vehicle's perception and the group's consensus, coupled with the synchronous deviation of neighboring vehicles, indicates that the root cause of the deviation is not a problem with the vehicle itself, but rather a real change in road geometry, leading to a shift in the group's overall perception of lane lines.

[0137] For boundary scenarios involving anomaly detection, the following refined rules and false positive avoidance logic are added:

[0138] The rules for preventing false judgments of neighbor vehicle control anomalies are as follows: If the control deviation of a single neighbor vehicle exceeds the preset control deviation threshold, it must be continuously monitored for two or more cycles before it is finally determined that the control execution of the neighbor vehicle is abnormal. This avoids momentary false judgments caused by the neighbor vehicle temporarily avoiding or making minor directional adjustments. If the neighbor vehicle is finally determined to be abnormal, its unique identification is recorded simultaneously. If it is determined to be abnormal for three consecutive monitoring cycles, it is added to the communication isolation list and permanently removed from the neighbor communication list, and will no longer participate in the subsequent consensus generation process.

[0139] Priority judgment rule for multiple vehicles simultaneously abnormal: If the control deviation of neighboring vehicles is concentrated, and more than 50% of the neighboring vehicles have control deviation greater than the preset control deviation threshold, regardless of whether the vehicle's perceived deviation exceeds the standard, it will be prioritized as a road geometry change scenario rather than multiple vehicles simultaneously abnormal, so as to avoid batch misjudgment caused by objective road changes such as sudden changes in road alignment and construction sections.

[0140] The rules for distinguishing between normal driving behavior and abnormal behavior are as follows: If the turn signal status and lateral acceleration data broadcast by a neighboring vehicle meet the preset lane change characteristics, such as the left / right turn signal being on and the lateral acceleration being greater than 0.2m / s² for more than 1 second, even if its control deviation exceeds the threshold, it will not be judged as an abnormal control execution. It will be temporarily removed from the neighboring state estimation value set. After the lane change is completed and the lateral deviation returns to within the threshold, it will resume participating in consensus calculation, thus avoiding false isolation and false alarms caused by normal driving behavior.

[0141] After S4 completes the anomaly type determination, S5 is immediately triggered to generate and output an anomaly detection signal that identifies the anomaly source category based on the anomaly type. If no anomaly is determined, i.e., the vehicle's perceived deviation does not exceed the preset perceived deviation threshold, and the control deviation of all one-hop neighbor vehicles does not exceed the preset control deviation threshold, then no anomaly signal is output, and only the anomaly detection process continues to be executed; if an anomaly is determined, the corresponding anomaly detection signal is immediately output and simultaneously broadcast to all one-hop neighbor vehicles.

[0142] The anomaly detection signal uses digital encoding, which corresponds one-to-one with the anomaly type. The specific output logic is as follows:

[0143] The first anomaly detection signal corresponds to an abnormal visual perception of the vehicle: it is output to the lane keeping actuator of the vehicle, triggering the vehicle to switch to a cooperative leading mode based on the local consensus lane line of the vehicle as the control reference. That is, the vehicle no longer relies on its own visual sensor data, but uses the lane line of consensus of the group of vehicles as the reference for path planning and correction; at the same time, it broadcasts to neighboring vehicles, informing them that the vehicle has a perception abnormality and is currently driving in the consensus lane line.

[0144] The second anomaly detection signal corresponds to an abnormal control execution of a neighboring vehicle: it is output to the vehicle's lane-keeping actuator, triggering the vehicle to add the corresponding abnormal neighboring vehicle to the communication isolation list, suspend receiving status data and lane line parameters from that neighboring vehicle, and simultaneously switch to an autonomous isolation mode based on the vehicle's visual lane line as the control reference to prevent the vehicle from veering off course following the abnormal neighboring vehicle; at the same time, it broadcasts to other neighboring vehicles, informing them of a neighboring vehicle's control abnormality and requesting them to give way.

[0145] The third anomaly detection signal corresponds to road geometry changes: it is output to the vehicle's lane keeping actuator, triggering the vehicle to keep the lane line as the control reference, but reducing the lane keeping control gain to achieve smooth passage and avoid excessive correction caused by road changes; at the same time, it is broadcast to all neighboring vehicles to inform them of the road geometry changes and suggest reducing the control gain and driving smoothly.

[0146] In this embodiment, the output duration of the anomaly detection signal is 1 second. Furthermore, differentiated recovery judgment rules are set for different anomaly types to ensure the stability of mode switching.

[0147] If the perceived deviation of the vehicle is less than the preset perceived deviation threshold for three consecutive detection cycles, it is determined that the vehicle's visual perception has returned to normal. The first abnormal detection signal output is immediately stopped, and the system switches from the collaborative-dominated mode back to the normal vehicle group collaborative detection mode, restoring the weight allocation of the vehicle's visual data in consensus generation.

[0148] If a neighboring vehicle added to the communication isolation list has its lateral deviation less than the preset control deviation threshold for five consecutive detection cycles, and the communication link quality meets the list admission rules, it can be re-added to the neighbor communication list, resume participation in consensus generation and anomaly detection processes, and have its communication isolation lifted.

[0149] When the perceived deviation of this vehicle and the control deviation of all neighboring vehicles return to within the corresponding threshold and continue for two consecutive detection cycles, it is determined that the road alignment has been restored to a stable state. The third abnormal detection signal output is stopped, the lane keeping control gain is restored to the normal value, and the recovery information is broadcast to the group of vehicles simultaneously.

[0150] If the abnormal state completely disappears within one detection cycle after a single abnormality is triggered, the output of the abnormal signal will immediately stop and the normal detection and control mode will be seamlessly restored. If the abnormal state lasts for more than 1 second, the corresponding abnormal signal will be continuously output and the vehicle's audible and visual alarm will be triggered to remind the driver to pay attention until the corresponding recovery conditions are met.

[0151] In this embodiment, the beneficial effects of the lane keeping anomaly detection method in a decentralized vehicle group environment include at least the following:

[0152] It can effectively identify deviations in information acquired by vehicle sensors, distinguish whether the deviation is due to sensor malfunction or actual vehicle deviation, prevent the vehicle from blindly executing corrective control, and solve the problems of vehicle misjudging its current position and direction, deviating from the lane, or experiencing lateral control failure caused by perception deviations, thereby reducing traffic safety risks such as driving disorder, chain collisions, and road congestion.

[0153] By using a distributed scheme with no centralized fusion nodes, a lane reference based on vehicle consensus is generated, which solves the problem of inconsistent perception of lane reference among vehicles in decentralized vehicle group scenarios. This ensures the stability of vehicle group collaborative driving and fully leverages the inherent advantages of multi-node information complementarity and enhanced safety redundancy in vehicle groups.

[0154] It can accurately determine three types of anomalies: visual perception anomalies of the vehicle itself, control execution anomalies of neighboring vehicles, and road geometric changes. It can generate and output anomaly detection signals that identify the anomaly source category based on the anomaly type, thereby achieving accurate anomaly identification and targeted handling, and further improving the robustness and flexibility of the lane keeping system.

[0155] Without relying on central nodes, roadside equipment, or high-precision maps, it can complete multi-vehicle data interaction and collaborative anomaly detection through vehicle-to-vehicle communication. It is especially suitable for application scenarios such as high-speed platooning of commercial vehicles and collaborative cruise of trunk logistics, which greatly improves the safety of vehicle group driving.

[0156] Example 2:

[0157] One embodiment of the present invention provides a lane keeping anomaly detection device in a decentralized vehicle group environment, which is mainly applied to the vehicle in the decentralized vehicle group and effectively solves the problems of misjudgment, missed judgment and poor adaptability of existing devices.

[0158] The lane keeping anomaly detection device mainly includes a data acquisition module, a coordinate alignment module, a distributed consensus estimation module, a lateral deviation calculation module, an anomaly determination module, and a signal output module. Among them, the data acquisition module, as the basic data input unit of the lane keeping anomaly detection device, undertakes the dual functions of acquiring data from the vehicle itself and receiving data from neighboring vehicles.

[0159] The data acquisition module uses the vehicle's GPS positioning module, IMU inertial measurement unit, onboard vision sensor, and V2V communication module to collect the vehicle's core data in real time. GPS coordinates determine the vehicle's real-time position, heading angle indicates its direction of travel, and visual lane line parameters are obtained by processing lane lines captured by the onboard camera using algorithms. These parameters reflect the vehicle's visual perception of the current lane. The V2V communication signal-to-noise ratio is used to evaluate the quality of the communication link between the vehicle and neighboring vehicles. Simultaneously, the data acquisition module receives GPS coordinates, heading angles, and visual lane line parameters from all one-hop neighboring vehicles via the V2V communication link. After initial verification, the collected and received data is synchronously transmitted to the coordinate alignment module to ensure the accuracy and reliability of subsequent data processing.

[0160] The coordinate alignment module connects to the data acquisition module to address the issue of inconsistent coordinate systems among multiple vehicles in a decentralized vehicle group, providing a unified benchmark for collaborative data analysis. After receiving the GPS coordinates and heading angles of the vehicle and all one-hop neighboring vehicles from the data acquisition module, the coordinate alignment module first calculates the relative positional relationship between the two vehicles based on their GPS coordinates. Then, it calculates a relative rotation matrix using the heading angles of both vehicles to correct heading deviations. Simultaneously, it calculates a relative translation vector to correct positional deviations. Subsequently, using the calculated relative rotation matrix and relative translation vector, it transforms the visual lane line parameters of each neighboring vehicle, mapping them to the vehicle's coordinate system. This ensures that the lane line parameters of the vehicle and all neighboring vehicles are in the same coordinate system. Finally, it integrates all transformed neighboring vehicle lane line parameters to form a neighbor-mapped lane line set. After verification, this set is synchronously transmitted to the distributed consensus estimation module.

[0161] The distributed consensus estimation module employs a distributed consensus filter, eliminating the need for a central fusion node and fully adapting to the architecture characteristics of decentralized vehicle groups. The module uses the vehicle's visual lane line parameters and the neighbor-mapped lane line sets transmitted by the coordinate alignment module as core inputs. It combines this with the V2V communication signal-to-noise ratio transmitted by the data acquisition module to dynamically allocate weighting coefficients. Higher communication link quality results in larger weighting coefficients for the corresponding vehicle, reflecting the reliability differences in lane line state estimates between the vehicle and its neighbors. Furthermore, all weighting coefficients strictly adhere to the normalization principle, meaning the sum of the vehicle's weighting coefficient and the weighting coefficients of all neighboring vehicles is 1, avoiding distortion in weighted calculations.

[0162] Subsequently, the distributed consensus estimation module iteratively updates the lane line state estimates of the vehicle and all one-hop neighbor vehicles according to the preset consensus update rules. It completes one iteration every 50ms until the convergence condition is met, and finally generates the local consensus lane line of the vehicle. This consensus lane line can represent the unified understanding of the current lane by the vehicle group, providing a core benchmark for subsequent deviation calculation and anomaly judgment, and is synchronously transmitted to the lateral deviation calculation module.

[0163] The lateral deviation calculation module is connected to both the data acquisition module and the distributed consensus estimation module. It is used to calculate multi-dimensional lateral deviations, providing data support for anomaly detection. Based on the vehicle's own coordinate system, the vehicle's visual lane line parameters transmitted by the data acquisition module, the vehicle's local consensus lane lines transmitted by the distributed consensus estimation module, and the position information of neighboring vehicles in the vehicle's own coordinate system, the lateral deviation calculation module calculates three types of core lateral deviations:

[0164] First, the lateral deviation of this vehicle relative to the local consensus lane line;

[0165] Second, the lateral deviation of this vehicle relative to its own visual lane line;

[0166] Third, the lateral deviation of each neighboring vehicle relative to the local consensus lane line of this vehicle.

[0167] Subsequently, all calculated lateral deviations are integrated to form a multi-dimensional deviation vector set, which clearly presents the deviation between the vehicle and the vehicle group consensus, and between neighboring vehicles and the vehicle group consensus. After data verification, the data is transmitted to the anomaly detection module.

[0168] The anomaly detection module connects to the lateral deviation calculation module and is responsible for accurately determining the type of anomaly. The anomaly detection module pre-sets two core thresholds: a perception deviation threshold and a control deviation threshold. The initial value for the perception deviation threshold is 0.15m, and the initial value for the control deviation threshold is 0.2m. These thresholds are not fixed but dynamically adjusted based on the current speed of the vehicles in the group. The adjustment principle is that the higher the speed, the larger the threshold setting, thus adapting to the deviation tolerance in high-speed driving scenarios and ensuring the rationality and adaptability of the judgment logic. Subsequently, based on the multi-dimensional deviation vector set transmitted by the lateral deviation calculation module, the anomaly detection module calculates the vehicle's perception deviation degree and the control deviation degrees of each neighboring vehicle. By comparing the perception deviation degree with the perception deviation threshold and the control deviation degree with the control deviation threshold, it accurately determines three types of anomalies: visual perception anomalies of the vehicle itself, control execution anomalies of neighboring vehicles, and road geometric changes, providing a basis for subsequent signal output.

[0169] The signal output module connects to the anomaly detection module and is responsible for outputting anomaly signals and continuously advancing the detection process. After the anomaly detection module determines the anomaly type, the signal output module immediately triggers the corresponding signal output process: if an anomaly is detected, it outputs the corresponding anomaly detection signal based on the anomaly type and simultaneously broadcasts it to all one-hop neighbor vehicles via the V2V communication link, alerting neighbor vehicles to the anomaly and facilitating coordinated response within the vehicle group; if no anomaly is detected, it does not output an anomaly signal but continuously triggers the detection process, synchronizing data every 50ms and repeatedly executing the workflow of each module to ensure real-time monitoring of the vehicle group's lane-keeping status and guarantee the safety and stability of the vehicle group's high-speed driving.

[0170] It will be apparent to those skilled in the art that this disclosure is not limited to the details of the exemplary embodiments described above, and that this disclosure can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of this disclosure is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within this disclosure. No reference numerals in the claims should be construed as limiting the scope of the claims.

[0171] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A method for detecting lane keeping anomalies in a decentralized vehicle group environment, characterized in that, Includes the following steps: S1. Collect the vehicle's visual lane line parameters, receive the neighbor visual lane line parameters broadcast by neighbor vehicles in the decentralized vehicle group through the V2V communication module, and convert the neighbor visual lane line parameters from the neighbor vehicle coordinate system to the vehicle's coordinate system to obtain the neighbor mapped lane line set. S2. Input the vehicle's visual lane line parameters and the neighbor's mapped lane line set into a distributed consensus filter without a central fusion node, and generate a local consensus lane line of the vehicle group consensus through interaction with neighbor vehicle data; S3. Calculate the first lateral deviation of the vehicle based on the local consensus lane line of the vehicle, calculate the second lateral deviation of the vehicle based on the visual lane line parameters of the vehicle, and simultaneously calculate the third lateral deviation set of each neighboring vehicle relative to the local consensus lane line of the vehicle. S4. Calculate the absolute value of the difference between the first lateral deviation and the second lateral deviation to obtain the perceived deviation degree of the vehicle. The neighbor control deviation set is obtained based on the third lateral deviation set; The vehicle's perceived deviation and neighbor's control deviation sets are compared with corresponding preset thresholds to determine the anomaly type. S5. Generate and output an anomaly detection signal that identifies the anomaly source category based on the anomaly type.

2. The lane keeping anomaly detection method in a decentralized vehicle group environment according to claim 1, characterized in that, The vehicle's visual lane line parameters are lane line polynomial coefficients obtained by semantic segmentation and lane line fitting of road images captured by the vehicle-mounted camera. The lane line polynomial coefficients are used to represent the lateral offset and heading angle deviation of the lane line relative to the vehicle's coordinate system.

3. The lane keeping anomaly detection method in a decentralized vehicle group environment according to claim 1, characterized in that, The neighbor visual lane line parameters are transformed from the neighbor vehicle coordinate system to the current vehicle coordinate system, specifically as follows: Based on the positioning coordinates and heading angles broadcast by neighboring vehicles, and the positioning coordinates and heading angles of this vehicle, calculate the relative rotation matrix and translation vector between this vehicle and the corresponding neighboring vehicle. The neighbor visual lane line parameters are mapped to the vehicle coordinate system to obtain the neighbor mapped lane line set.

4. The lane keeping anomaly detection method in a decentralized vehicle group environment according to claim 1, characterized in that, The process of generating local consensus lane lines for this vehicle through a distributed consensus filter without a centralized fusion node includes the following steps: S21. Use the vehicle's visual lane line parameters at the current moment as the initial state estimate of the distributed consensus filter; S22. Broadcast the state estimate of the current iteration cycle of this vehicle to all neighboring vehicles, and receive the state estimates of the current iteration cycle sent by the neighboring vehicles to obtain the set of neighboring state estimates; S23. Based on the set of neighbor state estimates and the state estimate of the vehicle in the current iteration cycle, calculate the state estimate of the vehicle in the next iteration cycle according to the preset consistency update rule. S24. Determine whether the difference between the state estimates of this vehicle and neighboring vehicles meets the preset convergence condition. If it does, then use the state estimate of the current iteration period as the local consensus lane line output of this vehicle. If the condition is not met, the state estimate of the next iteration cycle will be used as the new state estimate of the current iteration cycle, and the process will return to step S22.

5. The lane keeping anomaly detection method in a decentralized vehicle group environment according to claim 4, characterized in that, The specific steps for receiving state estimates sent by neighboring vehicles are as follows: This vehicle pre-generates and maintains a neighbor communication list, which contains unique identifiers of neighboring vehicles that have established stable V2V communication links with this vehicle. This vehicle only receives state estimates from neighboring vehicles whose unique identifiers are listed in the neighbor communication list.

6. The lane keeping anomaly detection method in a decentralized vehicle group environment according to claim 4, characterized in that, The preset consistency update rule is as follows: based on the state estimate of the current iteration cycle of the vehicle and all state estimates in the set of neighboring state estimates, a weighted fusion calculation is performed. The weighting coefficients are dynamically adjusted according to the communication link quality between the vehicle and the corresponding neighboring vehicles, and all weighting coefficients meet the normalization condition. The fused value is then used as the state estimate of the next iteration cycle.

7. The lane keeping anomaly detection method in a decentralized vehicle group environment according to claim 1, characterized in that, The lateral deviations in the first, second, and third lateral deviation sets are all lateral offsets of the vehicle body center point relative to the lane center line of the corresponding lane line.

8. The lane keeping anomaly detection method in a decentralized vehicle group environment according to claim 1, characterized in that, The vehicle's perceived deviation and neighbor's control deviation sets are compared with corresponding preset thresholds to determine the anomaly types, including: If the perceived deviation of the vehicle is greater than the preset perceived deviation threshold, and all elements in the neighbor control deviation set are less than the preset control deviation threshold, then the abnormality type is determined to be a visual perception abnormality of the vehicle. If at least one element in the neighbor control deviation set is greater than the preset control deviation threshold, and the vehicle's perceived deviation is less than the preset perceived deviation threshold, then the abnormality type is determined to be the control execution abnormality of the corresponding neighbor vehicle. If the perceived deviation of the vehicle is greater than the preset perceived deviation threshold, and at least one element in the neighbor control deviation set is greater than the preset control deviation threshold, then the abnormality type is determined to be road geometry change.

9. The lane keeping anomaly detection method in a decentralized vehicle group environment according to claim 1, characterized in that, The corresponding preset thresholds include a preset perception deviation threshold and a preset control deviation threshold. Both the preset perception deviation threshold and the preset control deviation threshold are dynamically adjusted based on the current vehicle speed, road curvature, and road surface adhesion coefficient.

10. A lane keeping anomaly detection device in a decentralized vehicle group environment, characterized in that, include: The data acquisition module is used to collect the GPS coordinates, heading angle, visual lane line parameters and V2V communication signal-to-noise ratio of the vehicle, and receive the GPS coordinates, heading angle and visual lane line parameters of all one-hop neighbor vehicles through V2V communication, and synchronously transmit them to the coordinate alignment module. The coordinate alignment module calculates the relative rotation matrix and translation vector between the vehicle and neighboring vehicles based on the received data, maps the visual lane line parameters of the neighboring vehicles to the coordinate system of the vehicle, forms a neighboring mapped lane line set, and transmits it to the distributed consensus estimation module. The distributed consensus estimation module uses a distributed consensus filter. It takes the vehicle's visual lane line parameters and the neighbor's mapped lane line set as inputs, dynamically allocates weighting coefficients based on the V2V communication signal-to-noise ratio, and iteratively updates the lane line state estimate until the convergence condition is met, thus generating the vehicle's local consensus lane line. The lateral deviation calculation module calculates the multi-dimensional lateral deviation of the vehicle and each of its one-hop neighboring vehicles, forms a deviation vector set, and transmits it to the anomaly determination module. The anomaly determination module presets a perception deviation threshold and a control deviation threshold. The threshold is dynamically adjusted according to the vehicle speed of the vehicle group. The anomaly type is determined by comparing the deviation with the threshold. The signal output module outputs and broadcasts the corresponding abnormal signal based on the judgment result; if there is no abnormality, the detection process continues to be executed.

Images