Vehicle control method and electronic device

Abstract

The application discloses a vehicle control method and an electronic device. The application relates to the technical field of automatic driving and artificial intelligence, and the method comprises the following steps: evaluating the perception credibility of a tracking target of a vehicle to obtain a credibility level of the tracking target; in the case that the credibility level is a first level, predicting a motion state vector at a current moment by using current perception data, or in the case that the credibility level is a second level, predicting the motion state vector at the current moment by using historical trajectory data; generating a trajectory sequence of the tracking target at multiple moments after the current moment by using a recursive prediction method based on the motion state vector at the current moment and the credibility level; and controlling the vehicle to run based on the trajectory sequence of the tracking target at the multiple moments after the current moment. The application solves the technical problem of low safety of vehicle control in the prior art.

Description

Technical Field

[0001] This application relates to the fields of autonomous driving technology and artificial intelligence technology, and more specifically, to a vehicle control method and electronic device. Background Technology

[0002] In autonomous driving scenarios in mines, trajectory prediction is a crucial step in autonomous driving control. Through trajectory prediction, changes in the surrounding traffic situation can be perceived in advance, improving the rationality, foresight, and safety of the vehicle's driving decisions. However, in complex actual working conditions, targets often experience problems such as instantaneous changes in heading angle, missing tracking data, and unstable motion states due to perceived noise, obstruction, low-speed movement, or brief periods of stillness. This causes the trajectory prediction to deviate significantly from the actual trajectory, increasing the risk of collision and resulting in a lower level of vehicle control safety.

[0003] There is currently no effective solution to the above problems. Summary of the Invention

[0004] This application provides a vehicle control method and electronic device to at least solve the technical problem of low safety level in vehicle control in related technologies.

[0005] According to one aspect of the embodiments of this application, a vehicle control method is provided, comprising: evaluating the perception credibility of a target being tracked by the vehicle to obtain a credibility level of the target; predicting a motion state vector at the current moment using current perception data when the credibility level is a first level, or predicting a motion state vector at the current moment using historical trajectory data when the credibility level is a second level, wherein the first level is lower than the second level; generating a trajectory sequence of the target at multiple moments after the current moment by recursive prediction based on the motion state vector at the current moment and the credibility level; and controlling the vehicle to drive based on the trajectory sequence of the target at multiple moments after the current moment.

[0006] Furthermore, the motion state vector at the current moment includes at least one of the following: position component, velocity component, heading angle component, and angular velocity component; predicting the motion state vector at the current moment using the current sensing data includes: determining the motion state vector at the current moment based on the bounding box information corresponding to the tracked target in the current sensing data; wherein, when the motion state at the current moment includes a heading angle component, determining the motion state vector at the current moment based on the bounding box information corresponding to the tracked target in the current sensing data includes: determining the number of vertices of the geometry corresponding to the tracked target based on the current sensing data; if the number of vertices is greater than or equal to a preset number, determining the heading angle component based on the bounding box information; if the number of vertices is less than the preset number, determining the heading angle component based on the velocity direction of the tracked target.

[0007] Furthermore, the motion state vector at the current moment is predicted using historical trajectory data, including: determining the target trajectory point closest to the current moment from the historical trajectory data based on the confidence level of historical trajectory points in the historical trajectory data, wherein the confidence level of the target trajectory point is less than a preset threshold, and the confidence level of the historical trajectory points is used to characterize the accuracy of the historical sensing data corresponding to the historical trajectory points; extracting trajectory points located after the target trajectory point from the historical trajectory data to obtain a target trajectory point sequence, wherein the length of the target trajectory point sequence is greater than or equal to a preset length, and the confidence level of the trajectory points in the target trajectory point sequence is greater than or equal to a preset threshold; determining the motion parameters of the tracked target based on the target trajectory point sequence; and determining the motion state vector at the current moment based on the motion parameters and / or the current sensing data.

[0008] Further, based on the target trajectory point sequence, the motion parameters of the tracked target are determined, including: smoothing the angle values ​​in the target trajectory point sequence based on a sliding window of a preset length to obtain a smoothed trajectory point sequence; and determining the motion parameters based on the smoothed trajectory point sequence. Preferably, smoothing the angle values ​​in the target trajectory point sequence based on a sliding window of a preset length to obtain a smoothed trajectory point sequence includes: controlling the sliding window to slide from the first trajectory point in the target trajectory point sequence; determining multiple angle values ​​within the sliding window during each slide of the sliding window; normalizing the multiple angle values ​​to obtain multiple normalized angle values; determining the average of the multiple normalized angle values ​​to obtain the smoothed angle value within the sliding window; and so on until all angle values ​​in the target trajectory point sequence have been processed.

[0009] Further, the motion parameters include at least one of the following: velocity parameters, angular velocity parameters, and angle parameters; based on the target trajectory point sequence, the motion parameters of the tracked target are determined, including one or a combination of the following: the velocity parameter is determined based on the total displacement between adjacent trajectory points in the target trajectory point sequence and the total time corresponding to the target trajectory point sequence; the angular velocity parameter is determined based on the total change in smoothed angle values ​​in the target trajectory point sequence and the total time; the angle parameter is determined based on the average angle of the smoothed angle values ​​in the target trajectory point sequence; wherein, the angular velocity parameter is within a first preset angular velocity range.

[0010] Furthermore, based on the motion state vector at the current moment and the confidence level, a trajectory sequence of the tracked target at multiple moments after the current moment is generated through recursive prediction. This includes: determining the target turning pattern at the current moment based on the historical angular velocity data of the tracked target; determining the target covariance matrix based on the confidence level; and generating a trajectory sequence at multiple moments by recursively predicting the target turning pattern, the target covariance matrix, and the motion state vector at the current moment. Preferably, when the confidence level is at the first level, the target covariance matrix is ​​a preset matrix, and when the confidence level is at the second level, the target covariance matrix is ​​the historical covariance matrix.

[0011] Further, determining the target turning mode at the current moment based on the historical angular velocity data of the tracked target includes: acquiring a preset number of historical angular velocity values ​​closest to the current moment from the historical trajectory data; determining the target turning mode based on the preset number of historical angular velocity values ​​and the angular velocity value of the tracked target at the current moment from the current perception data; preferably, determining the target turning mode based on the preset number of historical angular velocity values ​​and the angular velocity value of the tracked target at the current moment from the current perception data includes: performing abrupt change detection on the target's historical angular velocity values ​​from the preset number of historical acceleration values ​​to obtain a preset number of first angular velocity values, wherein the target's historical angular velocity values ​​are the historical angular velocity values ​​closest to the current moment from the preset number of historical angular velocity data; performing outlier processing on the preset number of first angular velocity values ​​to obtain a preset number of second angular velocity values; and determining the target turning mode based on the preset number of second angular velocity values.

[0012] Further, based on a preset number of second angular velocity values, the target steering mode is determined, including: if the preset number of second angular velocity values ​​and the current angular velocity value meet the preset mode judgment conditions, the target steering mode is determined to be a sharp steering mode; if the preset number of second angular velocity values ​​and the current angular velocity value do not meet the preset mode judgment conditions, the target steering mode is determined to be a normal steering mode; preferably, the preset mode judgment conditions include: all preset number of historical angular velocity values ​​are less than the preset angular velocity; the preset number of historical angular velocity values ​​do not meet the monotonically increasing condition; and the current angular velocity value is within the range of the second preset angular velocity.

[0013] Furthermore, the credibility of the vehicle's tracking target is evaluated to obtain the credibility level of the tracking target, including: acquiring tracking data of the tracking target; based on the tracking data, determining at least one perception evaluation indicator of the tracking target, wherein the at least one perception evaluation indicator includes at least one of the following: the duration of the historical trajectory data of the tracking target, the tracking time of the tracking target, the time the tracking target is in motion, the speed of the tracking target at the current moment, and the confidence level of the current perception data; if any perception evaluation indicator meets the preset evaluation conditions corresponding to the perception evaluation indicator, the credibility level is determined to be the first level; if none of the at least one perception evaluation indicator meets the preset evaluation conditions corresponding to the at least one perception evaluation indicator, the credibility level is determined to be the second level.

[0014] According to another aspect of the embodiments of this application, a vehicle control device is also provided, comprising: an evaluation module for evaluating the perceived credibility of a target being tracked by the vehicle to obtain a credibility level of the target being tracked; a prediction module for predicting the motion state vector at the current moment using current perception data when the credibility level is a first level, or predicting the motion state vector at the current moment using historical trajectory data when the credibility level is a second level, wherein the first level is less than the second level; a generation module for generating a trajectory sequence of the target being tracked at multiple moments after the current moment based on the motion state vector at the current moment and the credibility level through recursive prediction; and a control module for controlling the vehicle to drive based on the trajectory sequence of the target being tracked at multiple moments after the current moment.

[0015] According to another aspect of the embodiments of this application, an electronic device is also provided, including: a memory storing an executable program; and a processor for running the program, wherein the program executes the methods in various embodiments of this application when it runs.

[0016] According to another aspect of the embodiments of this application, a computer-readable storage medium is also provided, the computer-readable storage medium including a stored executable program, wherein, when the executable program is running, it controls the device where the computer-readable storage medium is located to perform the methods of various embodiments of this application.

[0017] According to another aspect of the embodiments of this application, a computer program product is also provided, including a computer program that, when executed by a processor, implements the methods of various embodiments of this application.

[0018] According to another aspect of the embodiments of this application, a computer program product is also provided, including a non-volatile computer-readable storage medium storing a computer program that, when executed by a processor, implements the methods in various embodiments of this application.

[0019] According to another aspect of the embodiments of this application, a computer program is also provided, which, when executed by a processor, implements the methods of the various embodiments of this application.

[0020] In this application embodiment, a vehicle control method is proposed. The method first evaluates the perception credibility of the target being tracked by the vehicle to obtain the credibility level of the target being tracked. Then, if the credibility level is the first level, the motion state vector at the current moment is predicted using the current perception data, or if the credibility level is the second level, the motion state vector at the current moment is predicted using historical trajectory data. Based on the motion state vector at the current moment and the credibility level, the trajectory sequence of the target at multiple moments after the current moment is generated by recursive prediction. Finally, the vehicle is controlled to drive based on the trajectory sequence of the target at multiple moments after the current moment. This application adopts a dynamic switching state estimation strategy based on the perception reliability of the vehicle-tracking target. By disabling reliance on historical trajectory data and prioritizing the use of current perception data for state assignment in the first level with a lower confidence level, and using historical trajectory data for trajectory prediction in the second level with a higher confidence level, this approach avoids the erroneous propagation of unstable or distorted historical motion patterns to the prediction process. This achieves the technical objective of ensuring physical rationality and temporal continuity in state estimation under different perception quality conditions. It also enables the predicted trajectory to remain continuous, smooth, and non-intrusive to the vehicle's planned path even in unstable scenarios such as sudden changes in heading, low speed, and transitions between static and dynamic states. This improves the accuracy of trajectory prediction, thereby enhancing the safety of vehicle control and solving the technical problem of low safety in vehicle control in related technologies. Attached Figure Description

[0021] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0022] Figure 1 This is a flowchart of a vehicle control method according to an embodiment of this application;

[0023] Figure 2 This is a flowchart of a first-level judgment according to an embodiment of this application;

[0024] Figure 3 This is a flowchart of a first-level trajectory sequence calculation according to an embodiment of this application;

[0025] Figure 4 This is a flowchart of a trajectory rationality verification method according to an embodiment of this application;

[0026] Figure 5(a) is a diagram showing the predicted trajectory effect before improvement in the same scene instance according to an embodiment of this application;

[0027] Figure 5(b) is an improved prediction trajectory diagram of the same scene instance according to an embodiment of this application;

[0028] Figure 6 This is a schematic diagram of a vehicle control device according to an embodiment of this application. Detailed Implementation

[0029] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0030] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0031] According to an embodiment of this application, an embodiment of a vehicle control method is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0032] Figure 1 This is a flowchart of a vehicle control method according to an embodiment of this application, such as... Figure 1 As shown, the method includes the following steps:

[0033] Step S102: Evaluate the perceived credibility of the vehicle's tracking target to obtain the credibility level of the tracking target.

[0034] The aforementioned vehicle can refer to the main platform performing perception, tracking, and trajectory prediction functions. Vehicle types may include, but are not limited to, engineering work vehicles, low-speed / stationary start-up vehicles, heavy trucks, small freight / passenger vehicles, and unmanned / cooperative vehicles, etc. The specific vehicle needs to be determined based on the actual control objectives. The aforementioned vehicle can serve as the control subject of the control method proposed in this application, thereby improving vehicle safety in different environments by implementing the proposed control method.

[0035] The aforementioned tracking targets can refer to dynamic targets perceived in the vehicle's environment, that is, external dynamic obstacles continuously tracked by the perception system. Tracking targets can include, but are not limited to, other vehicles, pedestrians, non-motorized vehicles, construction equipment, animals, etc., and the specific tracking targets need to be determined based on actual tracking requirements. The tracking targets are the objects of trajectory prediction. The accuracy of the tracking target's state directly affects the safety and comfort of subsequent decision-making and planning modules. In this application, by conducting a credibility assessment on all tracking targets, classification processing is achieved, thereby distinguishing between "stable and reliable" and "unstable and unreliable" targets, avoiding the introduction of noise or low-quality data into the prediction model, and thus improving the robustness of the overall trajectory prediction.

[0036] The aforementioned perception reliability refers to a quantitative evaluation index that measures whether the perception system's state estimation of the tracked target is reliable and trustworthy at the current moment. By quantitatively evaluating the perception reliability, it is possible to dynamically decide: whether to enable unscented Kalman filter (UKF) state recursion using historical trajectory data (for high-reliability targets); whether to downgrade to bounding box direct induction mode (for low-reliability targets); and whether to suppress angular velocity updates or force zeroing (to prevent jitter amplification). This avoids trajectory prediction distortion caused by perception jitter, data loss, or low-speed disturbances, and improves the physical rationality and security of the prediction results.

[0037] The aforementioned confidence level can refer to a binary classification label for each tracked target based on the perceived confidence level assessment results, used to guide the subsequent UKF prediction processing strategy. Confidence levels may include, but are not limited to, High Confidence Level (HCL) and Low Confidence Level (LCL), with the specific confidence level determined based on the actual perceived confidence level assessment results. Confidence levels can be used as a basis for implementing the differentiated UKF processing strategy in this application.

[0038] In one optional embodiment, the perceived state of all tracked targets of the vehicle is first evaluated in batches, and the credibility level of each target is determined based on preset credibility judgment conditions. This evaluation process does not rely on the specific motion characteristics of the target or the prediction result, but directly evaluates based on the original perceived state information of the tracked target to obtain the credibility level of the tracked target, thereby completing the classification and screening of target reliability in the pre-prediction stage.

[0039] For example, assessing the perceived reliability of a vehicle's tracking target to obtain its reliability level can include: sequentially determining whether the number of historical trajectory points of the tracking target is less than 0.5 seconds, whether the system tracking time is less than 1 second, whether the motion duration is less than 1.5 seconds, whether the current speed is less than 5 km / h, or whether the confidence score output by the perception module is less than 0.6. If any condition is met, it is determined to be a low reliability level; otherwise, it is determined to be a high reliability level. This distinguishes between stable and reliable targets and unstable and easily disturbed targets, providing a classification basis for subsequent UKF prediction strategies.

[0040] Step S104: If the confidence level is the first level, the motion state vector at the current moment is predicted using the current perception data; or if the confidence level is the second level, the motion state vector at the current moment is predicted using historical trajectory data. The first level is less than the second level.

[0041] The first level mentioned above can refer to the state category that is judged as the Low Confidence Level (LCL) in the confidence level classification. That is, the perceived confidence level of the target does not meet the stable prediction conditions, and its state estimation depends on the current instantaneous perception information and does not use historical trajectory data.

[0042] The aforementioned current perception data refers to instantaneous target observation information output by the perception system (such as LiDAR, camera, or fusion module) at the current moment, without historical state association. Current perception data may include, but is not limited to, complete or incomplete bounding box information. Complete bounding box information includes: the target is a rectangle (number of polygon points = 4), allowing calculation of its heading angle; incomplete bounding box information includes: the target is a non-rectangular polygon (number of points ≠ 4), retaining only the center point coordinates and velocity direction, used for downgraded heading estimation. The specific current perception data needs to be determined based on the actual situation. At Level 1 (LCL), current perception data can be used as the sole reliable input to directly construct the motion state vector of the tracked target, replacing UKF recursion based on historical trajectories, achieving "model-free" or "weak model" state estimation, and effectively avoiding prediction divergence caused by data missingness or jitter.

[0043] The aforementioned motion state vector refers to the set of state variables in the UKF filter used to characterize the dynamic behavior of the target. The motion state vector may include, but is not limited to: the target's lateral position (x) in the global coordinate system, the target's longitudinal position (y) in the global coordinate system, the target's linear velocity (v), the target's heading angle (θ), and the target's angular velocity (ω), etc. The specific motion state vector needs to be determined according to actual requirements. The motion state vector can be used as the core input of the UKF prediction module, determining the trajectory prediction result at the next moment. The accuracy of the motion state vector directly affects the security of the downstream decision-making and planning module.

[0044] The aforementioned High Confidence Level (HCL) refers to a state category classified as having high confidence in the confidence level classification. This means the tracked target possesses sufficient, continuous, and stable historical motion data, and the target's state estimation can safely rely on historical trajectory information. The HCL can be used as a trigger condition for the standard UKF prediction strategy. The HCL indicates that the target's motion characteristics are stable and perception is reliable, allowing the use of historical trajectory data for prediction, thereby improving prediction accuracy and trajectory continuity, and meeting the high-precision trajectory prediction requirements of autonomous driving.

[0045] The aforementioned historical trajectory data can refer to multiple high-confidence state points of the target over a certain number of past moments, aligned with a time series. At Level 2 (HCL), historical trajectory data is used to calculate the target's motion state vector, providing accurate and smooth initial states and covariance priors for UKF, thus improving the rationality and physical consistency of the predicted trajectory.

[0046] In one optional embodiment, if the tracked target's confidence level is determined to be Level 1, the motion state vector is directly inferred from the current sensing data without incorporating any historical trajectory information, ensuring the timeliness and stability of state estimation under low confidence conditions. However, if the tracked target's confidence level is Level 2, then historical trajectory data is used to predict the motion state vector. Long-term observation information enhances the continuity and accuracy of state estimation. Both methods utilize drastically different data sources to perform state inference based on the difference in confidence levels, thus achieving differentiated processing for targets with different confidence levels.

[0047] Step S106: Based on the motion state vector and confidence level at the current moment, generate a trajectory sequence of the target at multiple moments after the current moment through recursive prediction.

[0048] The aforementioned recursive prediction refers to the process of using the current motion state vector as the initial state, combined with a preset time step, and iteratively calculating the target's state estimate at multiple discrete future times according to the unscented Kalman filter (UKF) state transition model. This process dynamically selects a prediction strategy based on the target's confidence level: high-confidence targets inherit the historical covariance matrix, while low-confidence targets use a covariance matrix initialized to zero.

[0049] Recursive prediction can include, but is not limited to, standard recursive modes and degraded recursive modes. The specific recursive prediction mode needs to be determined based on the confidence level. Recursive prediction is the core computational mechanism for trajectory prediction. It utilizes the nonlinear state propagation capability of the UKF (United Kingdom Functions) to probabilistically extrapolate the future position and attitude of the target while preserving physical motion constraints. In this application, the key improvement in recursive prediction lies in dynamically adjusting the covariance initialization method and turning mode judgment logic according to the confidence level. This avoids trajectory jitter caused by noise for low-confidence targets while ensuring the prediction accuracy for high-confidence targets, achieving an adaptive stability-accuracy balance.

[0050] The aforementioned trajectory sequence can refer to the set of future discrete prediction points arranged in chronological order, output by the UKF at each time step during the recursive prediction process. The trajectory sequence is the direct output of the UKF trajectory prediction and a key input for realizing the "prediction-decision-control" closed loop. This application, through a confidence level-driven recursive prediction mechanism, ensures that the trajectory sequence possesses high accuracy and smoothness under high confidence targets, and stability and physical plausibility (no sideways movement, no abrupt curvature) under low confidence targets, thereby significantly improving the safety and robustness of the autonomous driving system in complex and unstable environments.

[0051] In one optional embodiment, based on the current motion state vector and confidence level, a recursive prediction mechanism using unscented Kalman filtering (UKF) is employed to generate a trajectory sequence of the tracked target at multiple future time points. The motion state vector includes dynamic parameters such as the target's position, velocity, heading, and angular velocity, serving as the initial input for prediction. The confidence level determines the degree of reliance on historical information during prediction; high confidence allows full utilization of historical covariance and motion trends, while low confidence suppresses the propagation of historical information, ensuring that the prediction process is based solely on the finite state at the current moment. Recursive prediction involves calculating the target's motion state frame-by-frame in a time-step manner, with the output of each frame serving as the input for the next, forming a continuous trajectory sequence. This process does not rely on external corrections or resets, but autonomously evolves based solely on the current state and confidence level, thereby achieving continuous and temporal inference of the target's future path without introducing additional observation data.

[0052] Step S108: Control the vehicle's movement based on the trajectory sequence of the tracking target at multiple time points after the current time.

[0053] In one optional embodiment, the potential collision risk between the vehicle and the tracked target is assessed by analyzing the spatial relationship between the trajectory sequence of the tracked target at multiple time points after the current time and the vehicle's planned trajectory. Based on this, the vehicle's speed, acceleration, or steering commands are dynamically adjusted to complete vehicle control. For example, deceleration or lane change is triggered when the trajectory sequence at multiple time points approaches the vehicle's driving domain, or the current driving state is maintained when the trajectory sequence at multiple time points stably moves away from the vehicle's driving domain. This achieves real-time, safe, and collaborative control of the autonomous vehicle's driving behavior, ensuring smooth obstacle avoidance and efficient passage in complex dynamic environments.

[0054] In this application embodiment, a vehicle control method is proposed. The method first evaluates the perception credibility of the target being tracked by the vehicle to obtain the credibility level of the target being tracked. Then, if the credibility level is the first level, the motion state vector at the current moment is predicted using the current perception data, or if the credibility level is the second level, the motion state vector at the current moment is predicted using historical trajectory data. Based on the motion state vector at the current moment and the credibility level, the trajectory sequence of the target at multiple moments after the current moment is generated by recursive prediction. Finally, the vehicle is controlled to drive based on the trajectory sequence of the target at multiple moments after the current moment. This application adopts a dynamic switching state estimation strategy based on the perception reliability of the vehicle-tracking target. By disabling reliance on historical trajectory data and prioritizing the use of current perception data for state assignment in the first level with a lower confidence level, and using historical trajectory data for trajectory prediction in the second level with a higher confidence level, this approach avoids the erroneous propagation of unstable or distorted historical motion patterns to the prediction process. This achieves the technical objective of ensuring physical rationality and temporal continuity in state estimation under different perception quality conditions. It also enables the predicted trajectory to remain continuous, smooth, and non-intrusive to the vehicle's planned path even in unstable scenarios such as sudden changes in heading, low speed, and transitions between static and dynamic states. This improves the accuracy of trajectory prediction, thereby enhancing the safety of vehicle control and solving the technical problem of low safety in vehicle control in related technologies.

[0055] Optionally, the motion state vector at the current moment includes at least one of the following: position component, velocity component, heading angle component, and angular velocity component; predicting the motion state vector at the current moment using the current sensing data includes: determining the motion state vector at the current moment based on the bounding box information corresponding to the tracked target in the current sensing data; wherein, when the motion state at the current moment includes a heading angle component, determining the motion state vector at the current moment based on the bounding box information corresponding to the tracked target in the current sensing data includes: determining the number of vertices of the geometry corresponding to the tracked target based on the current sensing data; if the number of vertices is greater than or equal to a preset number, determining the heading angle component based on the bounding box information; if the number of vertices is less than the preset number, determining the heading angle component based on the velocity direction of the tracked target.

[0056] The aforementioned position components refer to the two scalar values ​​in the motion state vector that represent the target's two-dimensional planar position in the global coordinate system: the horizontal coordinate (x) and the vertical coordinate (y), expressed in meters (m). The position components can be directly determined from the bounding box information of the tracked target output by the current perception system (i.e., the coordinates of the bounding box's center point). The position components can be used as a fundamental component of the motion state vector, providing the UKF with a spatial reference point at the current moment and serving as the starting position for trajectory prediction. At low confidence levels, this component directly uses the bounding box center coordinates to avoid positional deviations caused by historical trajectory errors or drift.

[0057] The aforementioned velocity component can refer to the scalar value v representing the magnitude of the target's linear velocity in the motion state vector, measured in meters per second (m / s) or kilometers per hour (km / h). At low confidence levels, it is set to a preset minimum value to suppress abnormally low-speed estimations caused by perceived noise. The velocity component can be used to control the dynamic range of trajectory extrapolation. For low-confidence targets, setting a minimum velocity threshold prevents the target from being misjudged as "about to stop" due to momentary stillness or low-speed jitter, thereby avoiding premature convergence or stagnation of the predicted trajectory.

[0058] The aforementioned heading angle component refers to the two-dimensional angle value θ in the motion state vector, representing the target's orientation, measured in degrees (°), used to describe the target's orientation attitude in the global coordinate system. The heading angle component can be used to determine the trajectory's turning direction and curvature. This application dynamically acquires this component using either the bounding box orientation or the velocity direction, ensuring reasonable and stable heading estimates in both scenarios with complete and missing perception information, and avoiding trajectory abrupt changes caused by heading jitter.

[0059] The aforementioned angular velocity component refers to the scalar value ω in the motion state vector, representing the rate of change of the target's heading angle, measured in degrees per second (° / s), reflecting the target's rotational dynamics. The angular velocity component can be used to influence rotational dynamics modeling in UKF predictions. At low confidence levels, this component is forced to 0 to avoid false rotation predictions caused by insufficient data or perceived noise, ensuring that the trajectory does not exhibit "sideways" or non-physical rotation.

[0060] The bounding box information mentioned above refers to the target's circumscribed geometric bounding box data output by the perception module, typically a 2D rectangle (Box2D). Bounding box information may include, but is not limited to, the center point coordinates (x, y), width, height, rotation angle, or the coordinates of the four vertices. The specific bounding box information needs to be determined based on the actual tracked target. Bounding box information can be used for localization and attitude estimation. Bounding box information is the core carrier of current perception data, serving as the only reliable input source for position and heading at low confidence levels. The completeness of the bounding box information (number of vertices) can be used to determine how the heading angle component is obtained, and is a key basis for implementing the "perception degradation strategy."

[0061] The number of vertices in the aforementioned geometric shape can refer to the number of vertices in the polygon formed by the contour point set of the tracked target. The number of vertices in the geometric shape can be used as a criterion for determining whether the bounding box is a standard rectangle (i.e., whether it has a clear orientation). If the number of vertices is greater than or equal to a preset number, the tracked target is considered to be a rectangle, and its orientation can be safely calculated; otherwise, it is considered a non-standard shape, and the heading extraction based on the bounding box is abandoned, and the velocity direction is used instead.

[0062] The aforementioned preset quantity can refer to a threshold constant used to determine the geometric integrity of the bounding box. The preset quantity can be used as the sole hard standard for determining the integrity of the bounding box. The preset quantity is the key criterion of the "heading acquisition strategy branch" in this application, realizing the binary logic of "using the bounding box when there is one, and using the speed when there is no bounding box", ensuring the robustness of heading estimation in perception degradation scenarios.

[0063] The aforementioned velocity direction can refer to the velocity vector (v) of the target at the current moment. x v y The direction of velocity is indicated by the two-dimensional plane pointed to by . The direction of velocity can be determined by the arctangent function atan2(v). y v x The calculated angle value serves as an alternative estimate for the heading angle. When the bounding box information is unreliable, the velocity direction provides a physically reasonable alternative heading angle, avoiding random jumps in the heading angle due to noise or occlusion, and ensuring that the trajectory remains linearly smooth and extrapolated even under low confidence levels, thus conforming to motion continuity constraints.

[0064] In one optional embodiment, the motion state vector is first subdivided into components including position, velocity, heading angle, and angular velocity. Then, based on the bounding box information corresponding to the tracked target in the current sensing data, the motion state vector at the current moment is determined. Where the motion state at the current moment includes a heading angle component, the heading angle component is determined based on the bounding box information, and a mechanism for judging the number of vertices of the tracked target's geometry is introduced: when the number of vertices of the tracked target's geometry reaches or exceeds a preset threshold, the heading angle is preferentially calculated directly using the bounding box information to ensure high accuracy when the sensing information is complete; when the number of vertices is less than the preset threshold, a robust estimation method based on the target's velocity direction is automatically switched to avoid heading jitter caused by bounding box deformation, incompleteness, or noise interference. This achieves adaptive and reliable acquisition of the heading angle component. Combined with a confidence grading strategy for historical trajectories and current sensing, the stability and rationality of trajectory prediction for low-confidence targets in scenarios involving dynamic-to-static transitions, occlusion, or sensor jitter are improved, effectively solving the problem of abnormal trajectory fluctuations caused by inaccurate heading estimation in existing technologies.

[0065] For example, for obstacles that are classified as Level 1 (low confidence targets), a downgraded processing strategy is used to update the UKF state vector in order to ensure the stability and physical rationality of the state estimation in the absence of reliable historical trajectory data.

[0066] The specific update rules are as follows:

[0067] Position component update: The position coordinates (x, y) in the UKF state are directly set to the center point coordinates of the target bounding box (Box2D) at the current moment, without relying on historical trajectory calculation;

[0068] Velocity component update: Set the velocity component in the UKF state to a preset minimum value (e.g., 0.1 m / s) to prevent prediction drift or misjudgment of being stationary due to excessively low velocity;

[0069] Heading component update: A priority strategy is used to determine the heading angle θ. Specifically, this includes a priority strategy: if the target's polygon count is equal to 4 (i.e., it has a complete bounding box), then the bounding box's movement orientation angle is used as the heading angle θ; a demotion strategy: if the polygon count is not equal to 4, then the heading angle is calculated based on the current velocity vector, i.e.: ,in, , The x and y components of the current velocity in the global coordinate system.

[0070] Angular velocity component update: The angular velocity ω in the UKF state is forcibly set to 0.0° / s, which explicitly indicates that the target does not have reliable rotational motion capability, thus avoiding the divergence of rotation prediction due to noise or brief jitter.

[0071] The values ​​in the above steps are for illustrative purposes only. The specific values ​​need to be determined based on the actual situation, and no limit is set here.

[0072] Optionally, the motion state vector at the current moment is predicted using historical trajectory data, including: determining the target trajectory point closest to the current moment from the historical trajectory data based on the confidence level of historical trajectory points in the historical trajectory data, wherein the confidence level of the target trajectory point is less than a preset threshold, and the confidence level of the historical trajectory points is used to characterize the accuracy of the historical sensing data corresponding to the historical trajectory points; extracting trajectory points located after the target trajectory point from the historical trajectory data to obtain a target trajectory point sequence, wherein the length of the target trajectory point sequence is greater than or equal to a preset length, and the confidence level of the trajectory points in the target trajectory point sequence is greater than or equal to a preset threshold; determining the motion parameters of the tracked target based on the target trajectory point sequence; and determining the motion state vector at the current moment based on the motion parameters and / or the current sensing data.

[0073] The aforementioned historical trajectory points refer to single state sampling points of the tracked target recorded by the tracking system in the past time series, containing position (x, y), heading angle (θ), timestamp, and confidence information. Historical trajectory points can originate from UKF or the output historical trajectory data of associated tracking modules, etc., and the specific source needs to be determined based on the actual storage system. Historical trajectory points can be used as the original data source for motion state updates, and are used to identify stable and reliable motion patterns. This application ensures that subsequent motion parameter calculations are based solely on reliable observations by screening high-confidence trajectory points and eliminating interference from low-quality data.

[0074] The aforementioned confidence level can refer to the scalar value assigned by the perception and tracking system to each historical trajectory point, used to characterize the reliability or accuracy of the historical perception data (such as LiDAR, camera fusion output) corresponding to that point, with a value range of [0, 1]. The confidence level can be generated by the confidence level evaluation module inside the system.

[0075] In one alternative embodiment, the confidence level can be calculated by weighting at least one or more of the following factors, where the confidence level at any point is reduced if any factor satisfies the "unreliable" condition:

[0076] (1) Tracking duration factor: If the cumulative tracking time of the tracked target from the first detection to the current moment is less than 1 second, the confidence level is reduced by 0.2 to 0.4. The shorter the tracking time, the greater the reduction. (2) Motion time factor: If the cumulative motion time of the tracked target in a non-stationary state (speed > 0.5 km / h) is less than 1.5 seconds, the confidence level is reduced by 0.2 to exclude targets that have just started or are momentarily jittery. (3) Historical trajectory point quantity factor: If the total number of historical valid trajectory points of the tracked target is less than 3 (i.e., insufficient sampling within 0.5 seconds), the confidence level is set to below 0.3 to ensure that the data samples have statistical significance. (4) Current speed factor: If the current instantaneous speed of the tracked target is less than 5 km / h (low speed threshold), and the speed change rate within 3 consecutive frames is less than 0.1 m / s², it is judged as "suspected stationary but not stable", and the confidence level is reduced. (5) Perception source confidence factor: If the perception input of the current frame comes from a single sensor (such as only lidar without visual fusion) or the original confidence score output by the perception module is less than 0.5, the confidence of the historical trajectory point is multiplied by 0.7 to reflect the uncertainty brought about by the single perception mode; (6) Tracking association stability factor: If the target experiences tracking ID switching, ID jitter with neighboring targets, or data association failure in more than 2 consecutive frames, the confidence of the trajectory point is reduced by 0.3, indicating that it belongs to "drift target" or "false detection backtracking"; (7) Attitude change factor: If the absolute difference of the heading angle between a historical trajectory point and the previous frame exceeds 15° and the speed does not change synchronously (non-sharp turn scenario), it is judged as perception jitter or mismatch, and the confidence of the point is directly set to 0.1.

[0077] Each of the above factors is calculated independently for its weighting. The final confidence level is calculated as [initial confidence level 0.9] × Π(1 - weighting factor). The result should be no less than 0.05 and no more than 0.95 to prevent extreme values. The above values ​​are for illustrative purposes only, and specific values ​​can be adjusted according to actual circumstances. No limitation is set here.

[0078] In particular, the confidence level of historical trajectory points is strongly correlated with the "overall confidence level of the target" (i.e., LCL / HCL defined in step S102) at the corresponding time: if the tracked target is currently judged to be of low confidence level (LCL), the confidence level of all its historical trajectory points will be automatically reduced by at least 0.2 to reflect the system's distrust of the overall state of the target; conversely, if the tracked target is of high confidence level (HCL), the confidence level of its historical trajectory points will be retained first and only affected by the objective factors such as (3), (6), and (7) mentioned above.

[0079] The aforementioned confidence level is the core criterion for selecting valid historical trajectory data. The confidence level in this application may include, but is not limited to, high confidence: greater than or equal to a preset threshold, indicating stable and reliable sensing data; low confidence: less than a preset threshold, indicating unreliable data (e.g., short tracking time, low speed, jitter, etc.). This application utilizes confidence level to achieve "data cleaning": only high-confidence trajectory points are retained for motion parameter calculation, avoiding trajectory drift or heading jumps caused by sensing noise, occlusion, or tracking loss.

[0080] The aforementioned target trajectory point can refer to the last historical trajectory point in the historical trajectory data that has a confidence level lower than a preset threshold, tracing back from the current moment; that is, the starting boundary of the high-confidence data interval. The target trajectory point can be used as a "data truncation anchor point" to accurately define the "effective historical trajectory interval" that can be used for motion parameter calculation. The target trajectory point can also ensure that subsequent calculations are based on stable trajectory segments free from low-confidence data contamination, improving the accuracy and stability of UKF state updates.

[0081] The aforementioned preset threshold can refer to a fixed numerical threshold used to distinguish between high / low confidence trajectory points. The preset threshold may include, but is not limited to, 0.55, 0.5, 0.65, etc., and the specific preset threshold needs to be determined based on actual needs. The preset threshold can be used to ensure that only high-reliability data is used to update the UKF state, avoiding the introduction of erroneous motion models due to short-term perceived jitter or low-speed targets.

[0082] The aforementioned target trajectory point sequence refers to a sequence of historical trajectory points extracted sequentially in chronological order after the target trajectory point, each with a confidence level greater than or equal to a preset threshold. The target trajectory point sequence is the sole input dataset for calculating motion parameters (velocity, heading, angular velocity). The target trajectory point sequence ensures that all data used are high-quality, continuous, undisturbed, and high-confidence trajectories, providing a smooth and reasonable initial state for UKF, thereby improving the physical consistency and stability of the predicted trajectory.

[0083] The aforementioned preset length refers to the required length of the target trajectory sequence. The preset length can be set to 3, but the specific preset length needs to be determined based on actual requirements. The preset length is used to ensure data processing consistency and algorithm stability.

[0084] The aforementioned motion parameters can refer to quantitative indicators of the target's motion characteristics. Motion parameters may include, but are not limited to, velocity v, angular velocity ω, and heading angle θ. In this application, the motion parameters are calculated from a smoothed sequence of historical trajectory points and used to initialize the UKF state vector, serving as the core input for high-confidence target prediction.

[0085] In one optional embodiment, when the perceived confidence level of the tracked target is level two, the motion state vector at the current moment is determined by introducing a confidence mechanism based on historical trajectory points. In this process, firstly, based on the confidence level of historical trajectory points in the historical trajectory data, the target trajectory point closest to the current moment with a confidence level below a preset threshold is identified. This point marks the turning point where the confidence level in the historical trajectory data decreases. Then, consecutive trajectory points with confidence levels not lower than the preset threshold are extracted from after the target trajectory point, forming a target trajectory point sequence with a preset length and high reliability. This effectively eliminates low-confidence interference points caused by heading jitter, sensor false detections, or static / dynamic transitions, ensuring that the historical trajectory data used to calculate motion parameters has high accuracy and strong temporal consistency. Subsequently, the motion parameters of the tracked target are more realistically derived based on the target trajectory point sequence. Finally, the motion parameters and / or current perceived data are comprehensively used to determine the motion state vector at the current moment, improving the stability and rationality of trajectory prediction under unstable perception conditions. Ultimately, this solves the problem of inaccurate motion state estimation caused by low-confidence noise mixed in historical trajectory data, achieving accurate and robust prediction of the target trajectory.

[0086] Optionally, based on the target trajectory point sequence, the motion parameters of the tracked target are determined, including: smoothing the angle values ​​in the target trajectory point sequence based on a sliding window of a preset length to obtain a smoothed trajectory point sequence; and determining the motion parameters based on the smoothed trajectory point sequence.

[0087] The preset length mentioned above can refer to a fixed value representing the number of historical trajectory points covered by the sliding window. For example, the preset length can be set to 5, meaning that the number of consecutive trajectory points participating in the angle smoothing calculation each time is 5. The above preset length is only an example; the specific preset length needs to be determined based on the time and field of view requirements of the angle change.

[0088] The aforementioned sliding window can refer to a fixed-length data window that moves forward point by point in chronological order on the target trajectory point sequence.

[0089] The aforementioned angle value can refer to the heading angle θ recorded at each trajectory point in the target trajectory point sequence, in radians (rad) or degrees (°). It represents the target's orientation in the global coordinate system, and the value range is normalized to [-π, π] (or [-180°, 180°]) to avoid abrupt jumps when the angle crosses ±π (e.g., from 179° to -179°). The angle value can be used as the core input for calculating angular velocity and average heading. Due to perception noise, multi-sensor fusion errors, or target occlusion, the original angle value often exhibits instantaneous jumps. If directly used for UKF state updates, it will cause the predicted trajectory to produce "sawtooth" or "spinning" anomalies.

[0090] The aforementioned smoothing process can refer to smoothing each angle value within the sliding window to eliminate angle jumps across boundaries and suppress high-frequency noise.

[0091] The smoothed trajectory point sequence mentioned above can refer to the new trajectory point sequence generated after performing trigonometric function normalization smoothing on each trajectory point in the original target trajectory point sequence through a sliding window.

[0092] In one optional embodiment, the angle values ​​in the target trajectory point sequence are smoothed based on a sliding window of preset length. This involves normalizing each angle value within the window to obtain a smoothed trajectory point sequence. Subsequently, based on the smoothed heading angles of each point in the sequence, the target's motion parameters, including velocity, angular velocity, and average heading, are calculated. Velocity is determined by the displacement and time difference between adjacent points in the sequence; angular velocity is obtained by dividing the cumulative change in the smoothed heading angles by the total time; and the average heading is determined by the arithmetic mean of all smoothed heading angles. This process effectively eliminates heading jumps caused by perceived noise or angle crossing ±π boundaries, ensuring physical continuity and temporal consistency in the heading change sequence. Simultaneously, it reduces the interference of original angle jitter on motion estimation, improves the stability and accuracy of motion parameters, and provides more reliable and smooth input data for subsequent UKF state updates, avoiding trajectory curvature anomalies or prediction distortions caused by sudden heading changes.

[0093] Preferably, based on a sliding window of a preset length, the angle values ​​in the target trajectory point sequence are smoothed to obtain a smoothed trajectory point sequence, including: controlling the sliding window to slide from the first trajectory point in the target trajectory point sequence; determining multiple angle values ​​within the sliding window during each slide of the sliding window; normalizing the multiple angle values ​​to obtain multiple normalized angle values; determining the average value of the multiple normalized angle values ​​to obtain the smoothed angle value within the sliding window; until all angle values ​​in the target trajectory point sequence have been processed.

[0094] The aforementioned normalization process refers to mapping the original heading angle to a uniform periodic interval [-180°, 180°] (or [-π, π]), ensuring numerical continuity of all angle values ​​and avoiding abrupt changes when the angle crosses the ±180° (or ±π) boundary. Normalization methods may include, but are not limited to, periodic modular arithmetic normalization, trigonometric function normalized averaging, and adaptive interval normalization. The specific normalization method needs to be determined based on actual requirements.

[0095] The aforementioned normalized angle values ​​can refer to the normalized heading angle corresponding to each trajectory point currently covered by the sliding window.

[0096] The aforementioned smoothed angle value can refer to a single angle value obtained by performing an arithmetic mean operation on multiple normalized angle values ​​within the sliding window, which serves as the smoothed heading estimate corresponding to the current window center point.

[0097] In one optional embodiment, the sliding window is controlled to slide forward point by point, starting from the first trajectory point in the target trajectory point sequence. During each slide, the original heading angles corresponding to the multiple trajectory points currently covered in the window are extracted, and periodic modulo normalization processing is performed on each original angle value to uniformly map them to the interval [-180°, 180°], resulting in multiple normalized angle values. Subsequently, an arithmetic mean operation is performed on these normalized angle values ​​to calculate the smoothed angle value corresponding to the center point of the current window, and this value is used to replace the original heading angle as the updated heading of the trajectory point. This process continues until all trajectory points in the sequence have been processed, generating a smoothed trajectory point sequence. The above process effectively eliminates numerical jumps in heading angle caused by periodic boundaries through normalization, ensuring the physical continuity of angle data within the sliding window. Furthermore, an arithmetic mean is performed to suppress the interference of perceived noise and instantaneous jitter on heading estimation, making the output smooth angle value closer to the actual motion trend. This improves the stability and consistency of heading data in the trajectory point sequence, laying a reliable input foundation for the accurate calculation of subsequent motion parameters (such as angular velocity and average heading).

[0098] Optionally, the motion parameters include at least one of the following: velocity parameters, angular velocity parameters, and angle parameters; based on the target trajectory point sequence, the motion parameters of the tracked target are determined, including one or a combination of the following: determining the velocity parameters based on the total displacement between adjacent trajectory points in the target trajectory point sequence and the total time corresponding to the target trajectory point sequence; determining the angular velocity parameters based on the total change in smoothed angle values ​​in the target trajectory point sequence and the total time; determining the angle parameters based on the average angle of the smoothed angle values ​​in the target trajectory point sequence; wherein the angular velocity parameters are within a first preset angular velocity range.

[0099] The aforementioned velocity parameter refers to the average linear velocity per unit time within the time interval covered by the target trajectory point sequence. It is calculated by dividing the total displacement distance between adjacent trajectory points by the total time corresponding to the target trajectory point sequence. The velocity parameter can be used to reflect the overall movement speed of the tracked target within the observation window and is a direct input to the velocity component in the UKF state vector.

[0100] The aforementioned angular velocity parameter refers to the average rotation rate per unit time within the time interval covered by the target trajectory point sequence, calculated by dividing the total change in smoothed angle values ​​by the total time corresponding to the target trajectory point sequence. The angular velocity parameter can be used to describe the overall turning speed of the tracked target within the observation window.

[0101] The aforementioned angle parameter can refer to the arithmetic mean of all smoothed angle values ​​in the target trajectory point sequence, representing the overall orientation trend of the target within the observation window.

[0102] The total displacement between adjacent trajectory points can refer to the cumulative sum of Euclidean distances between all adjacent trajectory point pairs in the target trajectory point sequence, from the first trajectory point to the last trajectory point, representing the overall movement path length of the target within the observation window.

[0103] The total time corresponding to the above target trajectory point sequence can be the difference between the timestamp of the last trajectory point and the timestamp of the first trajectory point in the target trajectory point sequence, representing the observation duration covered by the sequence.

[0104] The total change in the aforementioned smoothing angle value can refer to the difference between the last smoothing angle value and the first smoothing angle value in the target trajectory point sequence.

[0105] The average angle of the above smoothed angle values ​​can refer to the arithmetic mean of all smoothed angle values ​​in the target trajectory point sequence.

[0106] The aforementioned first preset angular velocity range refers to a reasonable physical boundary range set for the angular velocity parameter, used to limit the output value of the angular velocity and prevent unreasonable high-speed rotation caused by noise or abnormal trajectories. The first preset angular velocity range may include, but is not limited to, [-17.5° / s, 17.5° / s], [-20° / s, 20° / s], [-22.5° / s, 22.5° / s], etc., and the specific first preset angular velocity range needs to be determined according to actual needs. The first preset angular velocity range can be used to ensure that the angular velocity parameter conforms to real vehicle dynamics constraints (such as the turning limits of trucks and mining trucks), prevent UKF from diverging due to extreme angular velocity inputs, and improve the physical rationality and safety of the predicted trajectory.

[0107] In one optional embodiment, by evaluating the confidence level of the tracked target, the motion state vector at the current moment is predicted using historical trajectory data at the second level. The target trajectory point sequence that meets the confidence condition is selected from the historical trajectory data. Then, the velocity parameter is calculated based on the total displacement and total time of adjacent trajectory points in the sequence, the angular velocity parameter is calculated based on the total change and total time of the smoothed angle value, and the angle parameter is determined based on the average value of the smoothed angle value. At the same time, the calculated angular velocity parameter is constrained by a first preset angular velocity range to ensure its physical rationality. This effectively avoids abnormal fluctuations in angular velocity caused by historical trajectory noise, data mutations, or dynamic-static transitions, improves the stability and reliability of the motion state vector, and ultimately ensures the smoothness and rationality of the trajectory prediction results in unstable target tracking scenarios, solving the trajectory distortion problem caused by exceeding the angular velocity limit.

[0108] Optionally, based on the motion state vector and confidence level at the current moment, a trajectory sequence of the tracked target at multiple moments after the current moment is generated by recursive prediction, including: determining the target turning mode at the current moment based on the historical angular velocity data of the tracked target; determining the target covariance matrix based on the confidence level; and generating a trajectory sequence at multiple moments by recursively predicting the target turning mode, the target covariance matrix, and the motion state vector at the current moment.

[0109] Preferably, when the confidence level is first, the target covariance matrix is ​​a preset matrix; when the confidence level is second, the target covariance matrix is ​​the historical covariance matrix.

[0110] The aforementioned historical angular velocity data can refer to the observed or estimated values ​​of the target angular velocity stored sequentially over several frames prior to the current moment. Historical angular velocity data can be used to determine the target's turning dynamics at the current moment, serving as input for target turning pattern classification.

[0111] The aforementioned target turning mode refers to the type of target turning behavior at the current moment, determined by preset logical rules based on historical angular velocity data. Target turning modes may include, but are not limited to, normal mode (the tracked target is in a stable or low-speed turning state) and sharp turn mode (the tracked target undergoes a drastic change in angular velocity within a short period, consistent with sharp turns or obstacle avoidance maneuvers). The specific target turning mode needs to be determined based on the historical angular velocity data of the tracked target. The target turning mode can be used to determine the state transition function parameters used by the Unscented Kalman Filter (UKF) during recursive prediction, enabling differentiated modeling of different turning behaviors and improving the rationality and physical consistency of the predicted trajectory.

[0112] The aforementioned target covariance matrix can refer to the symmetric positive definite matrix in unscented Kalman filtering (UKF) that describes the uncertainty of target state estimation. The target covariance matrix can include, but is not limited to, a preset matrix, historical covariance matrices, etc. The specific target covariance matrix needs to be determined according to the confidence level. The target covariance matrix can be used for the propagation of state uncertainty in the prediction stage.

[0113] The aforementioned preset matrix can refer to a fixed covariance matrix pre-set for low-confidence targets, in which all diagonal elements are preset small variance values, and off-diagonal covariance terms are 0, indicating that the target state uncertainty is low and each state variable is independent with no historical dependence.

[0114] The aforementioned historical covariance matrix refers to the covariance matrix retained by the tracked target after the previous UKF update, recording the evolution of uncertainty in the target's state estimation during historical tracking. The historical covariance matrix can be a 5×5 covariance matrix saved after the previous UKF update. The historical covariance matrix can be used to inherit historical uncertainty information for high-confidence targets, giving UKF predictions both memory and continuity.

[0115] In one optional embodiment, the target's turning pattern at the current moment is dynamically identified based on the historical angular velocity data of the tracked target, thereby distinguishing between normal turning and sharp turning behavior. This allows the recursive prediction process to adaptively adjust the prediction logic according to the dynamic characteristics of the turning. Simultaneously, a corresponding covariance matrix is ​​selected according to the confidence level of the tracked target. That is, a preset matrix is ​​used in the first level (low confidence) to suppress trajectory jumps caused by noise interference, and a historical covariance matrix is ​​used in the second level (high confidence) to continue the uncertainty propagation trend, ensuring that the uncertainty modeling of the motion state matches the actual perception quality. Finally, the target turning pattern, the adapted covariance matrix, and the current motion state vector are collaboratively input into the recursive prediction model to achieve smooth, continuous, and physically consistent generation of trajectory sequences for multiple subsequent moments. This effectively solves the problems of trajectory curvature distortion, unsmooth transition, and prediction lag caused by heading jitter, insufficient data, or dynamic-to-static transitions. In particular, it improves the stability and rationality of trajectory prediction in high-to-low confidence switching scenarios, achieving the technical effect of timely trajectory response, realistic shape, and no unreasonable jumps.

[0116] Optionally, the target turning mode at the current moment is determined based on the historical angular velocity data of the tracked target, including: obtaining a preset number of historical angular velocity values ​​that are closest to the current moment from the historical trajectory data; and determining the target turning mode based on the preset number of historical angular velocity values ​​and the angular velocity value of the tracked target at the current moment from the current perception data.

[0117] Preferably, determining the target turning mode based on a preset number of historical angular velocity values ​​and the angular velocity value of the tracked target at the current moment in the current sensing data includes: performing abrupt change detection on the target's historical angular velocity values ​​in the preset number of historical acceleration values ​​to obtain a preset number of first angular velocity values, wherein the target's historical angular velocity values ​​are the historical angular velocity values ​​closest to the current moment in the preset number of historical angular velocity data; performing outlier processing on the preset number of first angular velocity values ​​to obtain a preset number of second angular velocity values; and determining the target turning mode based on the preset number of second angular velocity values.

[0118] The aforementioned preset number of historical angular velocity values ​​can refer to a fixed number of target angular velocity estimates stored chronologically up to the current moment. These preset number of historical angular velocity values ​​can refer to the historical angular velocity values ​​of the most recent four frames, such as y1, y2, y3, and y4, where y4 is the historical angular velocity of the frame most recent to the current moment, and y1 is the oldest frame. These preset number of historical angular velocity values ​​can be used as historical context to determine the target's turning dynamic behavior, identifying whether it is in a "slow turn," "sharp turn," or "stable" state.

[0119] In one optional embodiment, the system first acquires a preset number (which can be 4 frames) of historical angular velocity values ​​from the historical trajectory data that are closest to the current moment. Then, it combines these preset number of historical angular velocity values ​​with the real-time angular velocity value output by the current sensing system (which can be the absolute value of the current angular velocity). Through abrupt change detection and outlier suppression mechanisms, it determines the current target turning mode of the tracked target to identify whether it belongs to "normal mode" or "sharp turn mode." This process, by fusing historical angular velocity trends with current sensing data, effectively isolates sensing noise from the true turning intention, reducing misjudgments of turning modes caused by single-frame anomalies.

[0120] The aforementioned target historical angular velocity value can refer to the historical angular velocity value closest to the current moment among a preset number of historical angular velocity values. In other words, it can refer to y4, the historical angular velocity of the frame closest to the current moment.

[0121] The aforementioned mutation detection can refer to determining whether there is a discontinuous jump in angular velocity by comparing the absolute difference between the angular velocity value in the current sensing data and the most recent frame (y4) of a preset number of historical angular velocity values.

[0122] The aforementioned preset number of first angular velocity values ​​can refer to an angular velocity sequence of length 4 after mutation detection processing. If a mutation is detected, the most recent frame (y4) is replaced with a marker value (100.0° / s), while the remaining frames remain unchanged. The preset number of first angular velocity values ​​can be used as intermediate outputs after mutation detection, preserving historical trends while marking abnormal frames, providing "labeled" inputs for subsequent outlier processing.

[0123] The aforementioned outlier handling refers to the logic of setting all elements in a predetermined sequence of angular velocity values ​​to zero if a value of 100.0° / s exists. This outlier handling effectively prevents misjudgments of the mode caused by a sudden change marker (100.0° / s) triggering a sharp turn mode, as this value still participates in subsequent mode assessments. Simultaneously, it ensures that after detecting a sudden change in angular velocity, historical trajectory data is completely cleared, forcing a "no-reference state," where the current angular velocity dominates mode judgment, avoiding false trends. Furthermore, the zeroed-out 0.0° / s sequence will no longer meet the angular velocity amplitude condition for the "sharp turn mode" (requiring >4.5° / s), thus automatically reverting to the normal mode and achieving a safe degradation.

[0124] The aforementioned preset number of second angular velocity values ​​may refer to the final 4-frame angular velocity sequence after outlier processing, which is used as the final input to the target turning mode determination module.

[0125] In one optional embodiment, the most recent historical angular velocity value (i.e., the latest historical angular velocity value) from a preset number (which can be 4 frames) of historical angular velocity values ​​is compared with the currently perceived angular velocity value. If the absolute difference between the two exceeds a preset threshold, it is determined to be an angular velocity mutation, and the most recent historical angular velocity value is replaced with an abnormal marker value (100.0° / s), forming a "first angular velocity value" sequence. Subsequently, if any element in this sequence is 100.0° / s, all four values ​​are uniformly set to zero, generating a "second angular velocity value" sequence to completely eliminate mutation interference. Finally, based on the zeroed or original "second angular velocity value" sequence, combined with the current angular velocity amplitude and its trend, it is determined whether the target is in "normal mode" or "sharp turn mode". This process, through a three-level processing mechanism of "mutation detection - abnormal marker - full sequence zeroing", achieves accurate identification and active suppression of perceived noise and non-physical jumps, avoids mode misjudgment caused by single-frame jitter, and ensures that the turning mode determination relies only on stable and reliable historical motion trends.

[0126] Optionally, the target steering mode is determined based on a preset number of second angular velocity values, including: if the preset number of second angular velocity values ​​and the current angular velocity value meet the preset mode judgment conditions, the target steering mode is determined to be a sharp steering mode; if the preset number of second angular velocity values ​​and the current angular velocity value do not meet the preset mode judgment conditions, the target steering mode is determined to be a normal steering mode.

[0127] Preferably, the preset mode judgment conditions include: the preset number of historical angular velocity values ​​are all less than the preset angular velocity; the preset number of historical angular velocity values ​​do not meet the monotonically increasing condition; and the angular velocity value at the current moment is within the range of the second preset angular velocity.

[0128] The aforementioned preset mode judgment conditions refer to three parallel logical conditions used to determine whether a target is in "sharp turn mode," all of which must be met simultaneously to trigger the sharp turn mode. These preset mode judgment conditions may include, but are not limited to, a preset number of second angular velocity values ​​all being less than a preset angular velocity, a preset number of second angular velocity values ​​not satisfying a monotonically increasing condition, and the current angular velocity value being within the range of the second preset angular velocity. These preset mode judgment conditions can be used to constitute the unique triggering mechanism for "sharp turn mode," ensuring that pattern recognition has clear, reproducible, and verifiable judgment criteria.

[0129] The aforementioned sharp turn mode refers to a motion state category in which the tracked target undergoes a significant, non-smooth change in angular velocity within a short period of time, exhibiting high dynamic turning characteristics. The sharp turn mode can be used to instruct the UKF prediction engine to adopt a high-response, low-inertia motion model, avoiding excessively small curvature or overly smooth predicted trajectory due to angular velocity lag.

[0130] The aforementioned normal steering mode refers to the motion state category corresponding to a relatively smooth and stable target steering behavior that does not meet any of the judgment conditions of the "sharp steering mode". In this case, UKF prediction uses a standard, conservative motion model for trajectory extrapolation.

[0131] The aforementioned preset angular velocity can refer to the upper limit threshold used to determine whether the historical angular velocity is in the "low amplitude" range. The preset angular velocity can be used to eliminate interference from extreme rotations (such as spinning in place or sensor failure).

[0132] The aforementioned monotonically increasing condition can refer to whether the absolute value sequence of a preset number of second angular velocity values ​​shows a frame-by-frame increasing trend.

[0133] The aforementioned second preset angular velocity range may refer to whether the absolute value sequence of a preset number of second angular velocity values ​​shows a frame-by-frame increasing trend.

[0134] In one optional embodiment, it is first determined whether a preset number of second angular velocity values ​​and the current angular velocity value meet preset mode judgment conditions. That is, it is determined that the preset number of historical angular velocity values ​​are all less than the preset angular velocity, the preset number of historical angular velocity values ​​do not meet the monotonically increasing condition, and the current angular velocity value is within the range of the second preset angular velocity. This accurately identifies whether the target turning mode is in a sharp turn mode. If the conditions are met, the recursive prediction logic for sharp turn adaptation is triggered. If the conditions are not met, it is determined that the target turning mode is in a normal turning mode. This process effectively filters out misjudgments caused by noise and brief disturbances, ensuring that the high-response prediction model can be switched to in time when the target undergoes a real sharp turn, and avoiding false activation when the heading is slightly jittery or data is insufficient. This improves the smoothness, stability, and safety of trajectory prediction in dynamic-static transitions, heading jitter, and data-sparse scenarios.

[0135] Optionally, the credibility of the vehicle's tracking target is evaluated to obtain a credibility level, including: acquiring tracking data of the tracking target; based on the tracking data, determining at least one perception evaluation indicator for the tracking target, wherein the at least one perception evaluation indicator includes at least one of the following: the duration of the historical trajectory data of the tracking target, the tracking time of the tracking target, the time the tracking target is in motion, the speed of the tracking target at the current moment, and the confidence level of the current perception data; if any perception evaluation indicator meets the preset evaluation conditions corresponding to the perception evaluation indicator, the credibility level is determined to be a first level; if none of the at least one perception evaluation indicator meets the preset evaluation conditions corresponding to the at least one perception evaluation indicator, the credibility level is determined to be a second level.

[0136] The aforementioned tracking data refers to the set of time-series observations and estimates related to the target state output by the autonomous vehicle's perception and tracking system during continuous target tracking, generated by sensors (such as LiDAR and cameras) and tracking algorithms (such as multi-target tracking). Tracking data may include, but is not limited to, the target's bounding box (Box2D) center coordinates, dimensions, historical trajectory point sequences, current velocity, heading, angular velocity, tracking ID, tracking start time, most recent valid detection time, perception confidence score, and target motion state (stationary / moving) markers. Specific tracking data needs to be determined based on actual requirements. Tracking data can be used as the sole input source for confidence level determination, supporting an objective and quantitative assessment of target tracking quality; it provides multi-dimensional and measurable feature information to determine whether the target is in a "high confidence" or "low confidence" state.

[0137] The aforementioned at least one perception evaluation indicator can refer to one of five independent dimensions extracted from tracking data to quantify the stability and reliability of target tracking. If any indicator meets its corresponding preset evaluation conditions, a "(Level 1) Low Reliability" judgment will be triggered. At least one perception evaluation indicator may include, but is not limited to, the duration of historical trajectory data, tracking time, time in motion, current speed, and the confidence level of the current perception data. Specific perception evaluation indicators can be determined according to actual needs. At least one perception evaluation indicator can be used to construct a multi-dimensional health assessment system for target "tracking quality," covering the time dimension, motion dimension, and perception confidence level dimension; enabling refined classification of "unstable targets" and avoiding misjudgment based on a single indicator.

[0138] The aforementioned preset evaluation conditions can refer to thresholds or logical rules that correspond one-to-one with each of the above-mentioned perception evaluation indicators, used to determine whether the indicator reaches the "unreliable" standard. As long as any indicator meets its corresponding preset evaluation condition, the target is determined to be of low reliability (level one).

[0139] In one optional embodiment, tracking data of the target is acquired, and at least one quantifiable perception evaluation index is calculated based on this data, including the duration of historical trajectory data, total tracking time, duration of motion state, current speed, and confidence level of the perception data, to construct a multi-dimensional credibility evaluation system. When any index reaches its corresponding preset threshold condition, the credibility level is determined to be Level 1; conversely, if all indicators fail to reach the preset condition, the credibility level is determined to be Level 2, thereby achieving a clear and objective stratification of target credibility. On this basis, for Level 1, the current perception data is preferentially used for motion state vector prediction, while for Level 2, historical trajectory is used for conservative extrapolation to ensure that a reasonable trajectory can still be output when the data is unstable or missing. Finally, the recursive prediction strategy is dynamically adjusted in combination with the credibility level to effectively suppress trajectory abrupt changes and unreasonable predictions caused by heading jitter, insufficient data, or dynamic-static transitions, thereby improving the stability and robustness of trajectory prediction in complex scenarios.

[0140] In one alternative embodiment, Figure 2 This is a flowchart of a first-level judgment according to an embodiment of this application, such as... Figure 2 As shown, the low confidence level conditions (first-level determination conditions) are first identified, including: insufficient historical trajectory data, short tracking time, low-speed target, short movement time, and low tracking confidence. When the judgment process begins, all objects are traversed to determine if prediction is needed. If so, a confidence level assessment is performed. If any condition (low confidence level condition) is met, the tracking target is marked as low confidence (first level) and logged. If none of the conditions are met, the tracking target is determined to have normal confidence (second level). Finally, after traversing all objects, the assessment is complete.

[0141] The above process is completed with a millisecond-level latency within each frame perception cycle, supporting high-concurrency target processing. This ensures that the system can achieve zero-latency, high-precision, and traceable classification responses for typical unstable objects such as "newly appearing targets, low-speed operating equipment, short-term occlusion targets, and vaguely perceived obstacles" in complex mining environments. This provides a core guarantee for the conservatism and security of subsequent trajectory prediction and is the key execution mechanism for the "intelligent hierarchical prediction" of this invention.

[0142] In one alternative embodiment, Figure 3 This is a flowchart of a first-level trajectory sequence calculation according to an embodiment of this application, such as... Figure 3As shown, the process first checks if the target has BOX (bounded box) information. If it does, the BOX orientation is corrected using a flip, the state vector is updated, and the process ends. If no bounded box information is found, the system checks for sufficient data. If insufficient data is found, the check fails, and the upstream orientation result is maintained. If sufficient data is found, the displacement point sequence is calculated, and the system further checks if angular velocity can be calculated. If angular velocity cannot be calculated, the system checks for displacement data. If no displacement data is found, the check fails, and the upstream orientation result is maintained. If displacement data exists, the displacement direction is used to determine that the update velocity is not less than 1.25 m / s, the state vector is updated, and the process ends. If angular velocity can be calculated, the motion orientation is calculated, the update velocity is determined to be not less than 1.25 m / s, the state vector is updated, and the process ends.

[0143] Specifically, the first judgment node is whether the current target has complete bounding box (Box2D) information: if the target has a standard four-point bounding box (i.e., the number of polygon points is 4), the system first extracts its motion orientation angle and uses a flip correction mechanism (i.e., adaptive correction is performed based on the angle between the velocity direction and the major axis of the bounding box to avoid the orientation being reversed due to the sensor view or occlusion) to calculate the heading angle θ. Then, the θ and the bounding box center coordinates (x, y) are used as position and heading inputs, the velocity component is set to 1.25m / s (preset minimum safe velocity threshold), the angular velocity is forcibly set to zero, the UKF state vector is directly updated, and the process terminates.

[0144] If the target lacks valid bounding box information (e.g., polygon point count ≠ 4 or the bounding box is marked as invalid), the system enters a degraded path: First, the system determines if valid displacement data exists (i.e., position points at least two consecutive timestamps can be used to calculate the displacement vector). If no valid displacement data exists, it is determined as "insufficient data," and the system does not update the heading and velocity, maintaining the previous prediction result to avoid orientation jumps caused by noise. If displacement data exists, the system further determines whether angular velocity can be calculated (i.e., whether there are at least three consecutive displacement points forming a valid motion trajectory sequence). If the angular velocity cannot be calculated (e.g., only two points), the motion orientation θ = atan2(Δy, Δx) is directly calculated based on the arctangent of the displacement vector (Δx, Δy), and the velocity is forcibly updated to be no less than 1.25 m / s, while the angular velocity remains zero, thus completing the state update. If the angular velocity can be calculated (≥3 displacement points), the average motion direction θ is calculated based on the continuous sequence of displacement points, and the angular velocity ω is calculated simultaneously, but the velocity is still protected by a lower limit (v ≥ 1.25 m / s), and the angular velocity ω is still limited to 0.0° / s (because the target is at a low confidence level and rotational dynamics are not trusted). Finally, the state vector is updated and the process ends.

[0145] This process employs a four-level fault-tolerance mechanism: "boundary box priority → displacement degradation → velocity lower limit protection → angular velocity disabling." This mechanism enables conservative state estimation with zero dependence on historical trajectories, zero-trust rotation modeling, and guaranteed motion constraints, even when target perception information is severely degraded.

[0146] Figure 4 This is a flowchart of a trajectory rationality verification method according to an embodiment of this application, such as... Figure 4 As shown, the first step in the process is mode determination: First, a jump detection is performed, that is, the change in angular velocity compared to the previous frame is greater than 5. If so, the angular velocity is set to 100° / s; if not, it is determined whether the angular velocity of the past 3 frames + the current frame is monotonically increasing and the current frame angular velocity is <99° / s. If the angular velocity of the past 3 frames + the current frame angular velocity is monotonically increasing and the current frame angular velocity is <99° / s, then it is further determined that 5° / s < angular velocity < 10° / s. If the angular velocity is within the above range, it is determined to be a sudden change mode (mode=1); if the angular velocity is not within the above range, it is determined to be a normal mode (mode=0); if the angular velocity of the past 3 frames + the current frame angular velocity is not monotonically increasing and the current frame angular velocity is <99° / s, it is directly determined to be a normal mode (mode=0).

[0147] Next, the recursive step i is performed. When mode=1, it is determined to be a sharp turn trajectory; when i<10, the angular velocity increases; when i>=10, the angular velocity decays to 40% after 20 frames; when mode=0, it is determined to be a normal trajectory, and at this time, the angular velocity dynamically and rapidly decays according to its actual magnitude. The angular velocity of the next frame is then determined. The current frame state variables (position, velocity, heading, angular velocity) and the angular velocity of the next frame are input into the recursive CTRV model to obtain the next frame state variables (position, velocity, heading, angular velocity).

[0148] Specifically, the process begins with steering mode determination: first, angular velocity jump detection is performed, comparing the predicted angular velocity of the current frame. Compared to the previous frame's historical angular velocity The absolute difference, if If the angular velocity changes abruptly, it is determined to be a sudden change in angular velocity. The current angular velocity is immediately forcibly set to 100.0° / s (as an anomaly marker value) to trigger the subsequent anomaly handling mechanism.

[0149] If no jump occurs, the system enters the rapid turn mode for judgment: the system checks the historical angular velocity of the last 3 frames ( , , Does ω0 satisfy the condition of being monotonically increasing (i.e., | |<| |<| |<| |) and the absolute value of the current angular velocity is less than 99.0° / s. If this condition is met and the absolute value of the current angular velocity falls within the range of (5.0° / s, 10.0° / s), then the target is determined to be in the real sharp turn start phase, and the system activates the angular velocity decay compensation mechanism: using this angular velocity value as a reference, it is linearly decayed to 40% of its original value in the subsequent 20 consecutive frames, realizing a "fast start, slow transition" steering dynamics simulation, avoiding trajectory "tailing" or "understeering" caused by the curvature lag of UKF prediction.

[0150] If the conditions for a sharp turn are not met, the system is judged to be in normal turning mode (mode=0). In this case, the system adopts a dynamic fast decay strategy: based on the absolute value of the current angular velocity, the system decays according to a nonlinear exponential function (e.g., ω_next=ω_current×(1-k|ω_current|), where k is the empirical decay coefficient, ω_current is the current angular velocity, and ω_next is the angular velocity of the next frame), to ensure rapid convergence during high-speed turns and smooth transition during low-speed turns.

[0151] After completing pattern identification and angular velocity correction, the system inputs the current frame's state vector and the corrected angular velocity ω_next for the next frame into the Constant Turn Rate and Velocity Model (CTRV), performs UKF prediction recursion, and outputs the position, velocity, heading, and angular velocity for the next moment. This process is executed in a closed loop within each prediction step, ensuring that the predicted trajectory conforms to the laws of motion in both space and dynamics.

[0152] Figure 5(a) is a prediction trajectory effect diagram of the same scene instance before improvement according to the embodiment of this application. As shown in Figure 5(a), the grid area in the figure is the perception area of ​​the vehicle; the squares on the grid area represent the vehicle; the two consecutive squares in front of the vehicle are used to represent the bounding box information corresponding to the tracking target; the black arrow inside the bounding box information on the far right represents the heading angle component in the motion state vector of the tracking target; the black arc is used to represent the prediction trajectory, that is, the traditional UKF prediction result without the solution of this application. It can be seen that during the target turning process, the curvature of the prediction trajectory increases abnormally due to heading jitter and angular velocity sensitivity, resulting in severe bending or even "sideways" phenomenon, which poses a security risk of intruding into the vehicle's planned path.

[0153] Figure 5(b) is an improved prediction trajectory diagram of the same scene example according to an embodiment of this application. As shown in Figure 5(b), the grid area in the figure represents the perception area of ​​the vehicle; the squares on the grid area represent the vehicle; the two consecutive squares in front of the vehicle represent the bounding box information corresponding to the tracked target; the black arrow inside the bounding box information on the far right represents the heading angle component in the motion state vector of the tracked target; the black arc in the figure represents the predicted trajectory, which is the prediction result after adopting the solution of this application. As can be seen from Figure 5(b), the trajectory transitions smoothly, and the predicted trajectory is more closely aligned with the direction of the black arrow. It can be seen that the curvature change of the predicted trajectory conforms to the physical laws, closely follows the real motion trend, and has no abrupt changes or excessive lag. This figure visually verifies the effectiveness of this application in key technical means such as "heading jitter suppression", "sharp turn mode adaptation", and "low reliability degradation", which strongly supports the technical progress of this solution in terms of safety and rationality.

[0154] According to an embodiment of this application, a vehicle control device is provided. It should be noted that the device can be used to execute the above-described vehicle control method.

[0155] Figure 6 This is a schematic diagram of a vehicle control device according to an embodiment of this application, such as... Figure 6 As shown, the device includes: an evaluation module 602, a prediction module 604, a generation module 606, and a control module 608.

[0156] Evaluation module 602 is used to evaluate the perceived credibility of the target being tracked by the vehicle and obtain the credibility level of the target being tracked; prediction module 604 is used to predict the motion state vector at the current moment using current perception data when the credibility level is first level, or to predict the motion state vector at the current moment using historical trajectory data when the credibility level is second level, wherein the first level is lower than the second level; generation module 606 is used to generate the trajectory sequence of the target being tracked at multiple moments after the current moment by recursive prediction based on the motion state vector at the current moment and the credibility level; control module 608 is used to control the vehicle's movement based on the trajectory sequence of the target being tracked at multiple moments after the current moment.

[0157] Optionally, the motion state vector at the current moment includes at least one of the following: position component, velocity component, heading angle component, and angular velocity component; the prediction module is used to determine the motion state vector at the current moment based on the bounding box information corresponding to the tracked target in the current sensing data; wherein, when the motion state at the current moment includes the heading angle component, determining the motion state vector at the current moment based on the bounding box information corresponding to the tracked target in the current sensing data includes: determining the number of vertices of the geometry corresponding to the tracked target based on the current sensing data; if the number of vertices is greater than or equal to a preset number, determining the heading angle component based on the bounding box information; if the number of vertices is less than the preset number, determining the heading angle component based on the velocity direction of the tracked target.

[0158] Optionally, the prediction module is further configured to: determine the target trajectory point closest to the current moment from the historical trajectory data based on the confidence level of the historical trajectory points in the historical trajectory data, wherein the confidence level of the target trajectory point is less than a preset threshold, and the confidence level of the historical trajectory points is used to characterize the accuracy of the historical sensing data corresponding to the historical trajectory points; extract trajectory points located after the target trajectory point from the historical trajectory data to obtain a target trajectory point sequence, wherein the length of the target trajectory point sequence is greater than or equal to a preset length, and the confidence level of the trajectory points in the target trajectory point sequence is greater than or equal to a preset threshold; determine the motion parameters of the tracked target based on the target trajectory point sequence; and determine the motion state vector at the current moment based on the motion parameters and / or the current sensing data.

[0159] Optionally, the prediction module is further configured to smooth the angle values ​​in the target trajectory point sequence based on a sliding window of a preset length, to obtain a smoothed trajectory point sequence; and to determine motion parameters based on the smoothed trajectory point sequence. Preferably, smoothing the angle values ​​in the target trajectory point sequence based on a sliding window of a preset length to obtain a smoothed trajectory point sequence includes: controlling the sliding window to slide from the first trajectory point in the target trajectory point sequence; determining multiple angle values ​​within the sliding window during each slide of the sliding window; normalizing the multiple angle values ​​to obtain multiple normalized angle values; determining the average of the multiple normalized angle values ​​to obtain the smoothed angle value within the sliding window; and so on until all angle values ​​in the target trajectory point sequence have been processed.

[0160] Optionally, the motion parameters include at least one of the following: velocity parameters, angular velocity parameters, and angle parameters; the prediction module is further configured to determine the velocity parameters based on the total displacement between adjacent trajectory points in the target trajectory point sequence and the total time corresponding to the target trajectory point sequence; determine the angular velocity parameters based on the total change in smoothed angle values ​​in the target trajectory point sequence and the total time; and determine the angle parameters based on the average angle of the smoothed angle values ​​in the target trajectory point sequence; wherein the angular velocity parameters are within a first preset angular velocity range.

[0161] Optionally, the generation module is used to determine the target turning pattern at the current moment based on the historical angular velocity data of the tracked target; determine the target covariance matrix based on the confidence level; and generate a trajectory sequence at multiple moments by recursively predicting the target turning pattern, the target covariance matrix, and the motion state vector at the current moment. Preferably, when the confidence level is the first level, the target covariance matrix is ​​a preset matrix, and when the confidence level is the second level, the target covariance matrix is ​​the historical covariance matrix.

[0162] Optionally, the generation module is further configured to acquire a preset number of historical angular velocity values ​​from the historical trajectory data that are closest to the current moment; determine the target turning mode based on the preset number of historical angular velocity values ​​and the angular velocity value of the tracked target in the current sensing data at the current moment; preferably, determining the target turning mode based on the preset number of historical angular velocity values ​​and the angular velocity value of the tracked target in the current sensing data at the current moment includes: performing abrupt change detection on the target historical angular velocity values ​​from the preset number of historical acceleration values ​​to obtain a preset number of first angular velocity values, wherein the target historical angular velocity values ​​are the historical angular velocity values ​​closest to the current moment from the preset number of historical angular velocity data; performing outlier processing on the preset number of first angular velocity values ​​to obtain a preset number of second angular velocity values; and determining the target turning mode based on the preset number of second angular velocity values.

[0163] Optionally, the generation module is further configured to determine the target steering mode as a sharp steering mode if the preset number of second angular velocity values ​​and the current angular velocity value meet the preset mode judgment conditions; and to determine the target steering mode as a normal steering mode if the preset number of second angular velocity values ​​and the current angular velocity value do not meet the preset mode judgment conditions. Preferably, the preset mode judgment conditions include: the preset number of historical angular velocity values ​​are all less than the preset angular velocity; the preset number of historical angular velocity values ​​do not meet the monotonically increasing condition; and the current angular velocity value is within the range of the second preset angular velocity.

[0164] Optionally, the evaluation module is used to acquire tracking data of the tracked target; based on the tracking data, determine at least one perception evaluation indicator of the tracked target, wherein the at least one perception evaluation indicator includes at least one of the following: the duration of the historical trajectory data of the tracked target, the tracking time of the tracked target, the time the tracked target is in motion, the speed of the tracked target at the current moment, and the confidence level of the current perception data; if any perception evaluation indicator meets the preset evaluation conditions corresponding to the perception evaluation indicator, the confidence level is determined to be the first level; if none of the at least one perception evaluation indicator meets the preset evaluation conditions corresponding to the at least one perception evaluation indicator, the confidence level is determined to be the second level.

[0165] Embodiments of this application also provide an electronic device, including: a memory storing an executable program; and a processor for running the program, wherein the program executes the methods in various embodiments of this application when it runs.

[0166] Embodiments of this application also provide a computer-readable storage medium including a stored executable program, wherein, when the executable program is running, it controls the device where the computer-readable storage medium is located to perform the methods of various embodiments of this application.

[0167] Embodiments of this application also provide a computer program product, including a computer program that, when executed by a processor, implements the methods of various embodiments of this application.

[0168] Embodiments of this application also provide a computer program product, including a non-volatile computer-readable storage medium for storing a computer program that, when executed by a processor, implements the methods in various embodiments of this application.

[0169] Embodiments of this application also provide a computer program that, when executed by a processor, implements the methods described in the various embodiments of this application.

[0170] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0171] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units can be a logical functional division, and in actual implementation, there may be other division methods. For instance, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.

[0172] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0173] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0174] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, read-only memory (ROM), random access memory (RAM), portable hard drive, magnetic disk, or optical disk.

[0175] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A vehicle control method, characterized in that, include: The perceived credibility of the vehicle tracking target is evaluated to obtain the credibility level of the tracking target; When the confidence level is at the first level, the motion state vector at the current moment is predicted using the current sensing data; or when the confidence level is at the second level, the target trajectory point sequence is extracted from the historical trajectory data based on the confidence level of the historical trajectory points in the historical trajectory data, the motion parameters of the tracked target are determined based on the target trajectory point sequence, and the motion state vector at the current moment is determined based on the motion parameters, wherein the first level is less than the second level; Based on the motion state vector at the current moment and the confidence level, a trajectory sequence of the tracked target at multiple moments after the current moment is generated by recursive prediction. The vehicle is controlled to move based on the trajectory sequence of the tracked target at multiple times after the current time.

2. The vehicle control method according to claim 1, characterized in that, The motion state vector at the current moment includes at least one of the following: position component, velocity component, heading angle component, and angular velocity component; the prediction of the motion state vector at the current moment using current sensing data includes: Based on the bounding box information corresponding to the tracking target in the current perception data, determine the motion state vector at the current moment; Wherein, when the motion state at the current moment includes the heading angle component, determining the motion state vector at the current moment based on the bounding box information corresponding to the tracked target in the current sensing data includes: Based on the current perception data, determine the number of vertices of the geometry corresponding to the tracked target; If the number of vertices is greater than or equal to a preset number, the heading angle component is determined based on the bounding box information; If the number of vertices is less than the preset number, the heading angle component is determined based on the velocity direction of the tracked target.

3. The vehicle control method according to claim 1, characterized in that, The process of extracting the target trajectory point sequence from the historical trajectory data based on the confidence level of historical trajectory points includes: Based on the confidence level of the historical trajectory points in the historical trajectory data, the target trajectory point closest to the current moment is determined from the historical trajectory data, wherein the confidence level of the target trajectory point is less than a preset threshold, and the confidence level of the historical trajectory points is used to characterize the accuracy of the historical sensing data corresponding to the historical trajectory points; The target trajectory point sequence is obtained by extracting trajectory points located after the target trajectory point from the historical trajectory data, wherein the length of the target trajectory point sequence is greater than or equal to a preset length, and the confidence level of the trajectory points in the target trajectory point sequence is greater than or equal to the preset threshold.

4. The vehicle control method according to claim 1, characterized in that, Determining the motion parameters of the tracked target based on the target trajectory point sequence includes: Based on a sliding window of preset length, the angle values ​​in the target trajectory point sequence are smoothed to obtain a smoothed trajectory point sequence. The motion parameters are determined based on the smoothed trajectory point sequence. The process of smoothing the angle values ​​in the target trajectory point sequence using a sliding window of a preset length to obtain a smoothed trajectory point sequence includes: The sliding window is controlled to slide from the first trajectory point in the target trajectory point sequence. During each slide of the sliding window, multiple angle values ​​located within the sliding window are determined, and the multiple angle values ​​are normalized to obtain multiple normalized angle values. The average value of the multiple normalized angle values ​​is determined to obtain the smoothed angle value within the sliding window, until all angle values ​​in the target trajectory point sequence have been processed.

5. The vehicle control method according to claim 1, characterized in that, The motion parameters include at least one of the following: velocity parameters, angular velocity parameters, and angle parameters; determining the motion parameters of the tracked target based on the target trajectory point sequence includes one or a combination of the following: The velocity parameter is determined based on the total displacement between adjacent trajectory points in the target trajectory point sequence and the total time corresponding to the target trajectory point sequence; The angular velocity parameter is determined based on the total change in the smoothed angle values ​​in the target trajectory point sequence and the total time. The angle parameter is determined based on the average angle of the smoothed angle values ​​in the target trajectory point sequence; The angular velocity parameter is located within a first preset angular velocity range.

6. The vehicle control method according to any one of claims 1 to 5, characterized in that, The step of generating a trajectory sequence of the tracked target at multiple time points after the current time by recursive prediction based on the motion state vector at the current time and the confidence level includes: Based on the historical angular velocity data of the tracked target, the target turning mode at the current moment is determined; Based on the aforementioned confidence level, the target covariance matrix is ​​determined; The target turning pattern, the target covariance matrix, and the motion state vector at the current moment are used to generate a trajectory sequence at multiple moments through recursive prediction. Specifically, when the confidence level is at the first level, the target covariance matrix is ​​a preset matrix; when the confidence level is at the second level, the target covariance matrix is ​​a historical covariance matrix.

7. The vehicle control method according to claim 6, characterized in that, Determining the target turning pattern at the current moment based on the historical angular velocity data of the tracked target includes: Obtain a preset number of historical angular velocity values ​​from the historical trajectory data that are closest to the current moment; Based on the preset number of historical angular velocity values ​​and the angular velocity value of the tracked target in the current sensing data at the current moment, the target turning mode is determined; The step of determining the target turning mode based on the preset number of historical angular velocity values ​​and the angular velocity value of the tracked target in the current sensing data at the current moment includes: Abrupt change detection is performed on the target historical angular velocity value in the preset number of historical acceleration values ​​to obtain the preset number of first angular velocity values, wherein the target historical angular velocity value is the historical angular velocity value closest to the current moment in the preset number of historical angular velocity data; The preset number of first angular velocity values ​​are processed for outliers to obtain the preset number of second angular velocity values. The target steering mode is determined based on the preset number of second angular velocity values.

8. The vehicle control method according to claim 7, characterized in that, Determining the target steering mode based on the preset number of second angular velocity values ​​includes: If the preset number of second angular velocity values ​​and the current angular velocity value satisfy the preset mode judgment condition, the target steering mode is determined to be a sharp steering mode. If the preset number of second angular velocity values ​​and the current angular velocity value do not meet the preset mode judgment condition, the target steering mode is determined to be the normal steering mode. The preset mode judgment conditions include: The preset number of historical angular velocity values ​​are all less than the preset angular velocity; The preset number of historical angular velocity values ​​do not satisfy the monotonically increasing condition; The angular velocity value at the current moment is within the second preset angular velocity range.

9. The vehicle control method according to any one of claims 1 to 5, characterized in that, The assessment of the perceived credibility of the vehicle tracking target to obtain the credibility level of the tracking target includes: Acquire tracking data of the target; Based on the tracking data, at least one perception evaluation index of the tracking target is determined, wherein the at least one perception evaluation index includes at least one of the following: the duration of the historical trajectory data of the tracking target, the tracking time of the tracking target, the time the tracking target is in motion, the speed of the tracking target at the current moment, and the confidence level of the current perception data; If any perception evaluation indicator meets the preset evaluation conditions corresponding to that perception evaluation indicator, the credibility level is determined to be the first level. If none of the at least one perception evaluation index meets the preset evaluation conditions corresponding to the at least one perception evaluation index, the credibility level is determined to be the second level.

10. An electronic device, characterized in that, include: Memory, which stores executable programs; A processor for running the program, wherein the program, when running, performs the method according to any one of claims 1 to 9.

Images