State prediction method, vehicle, and storage medium

Abstract

The embodiment of the application provides a state prediction method, a vehicle and a storage medium, the method comprises the following steps: obtaining a disappearing moment of an edge target; determining a context parameter related to the edge target according to the disappearing moment; constructing a noise covariance matrix according to the context parameter; predicting a state prediction vector and a state covariance matrix of the edge target according to the noise covariance matrix; and determining a motion state of the edge target according to the state prediction vector and the state covariance matrix. The application solves the technical problem that the state prediction error of the target after leaving the monitoring range of the sensor is large in the prior art, and the accuracy of the intelligent driving system for dynamic environment perception is poor.

Description

Technical Field

[0001] This application relates to the field of multi-sensor target fusion technology, and more specifically, to a state prediction method, a vehicle, and a storage medium. Background Technology

[0002] In existing technologies, when a target disappears from the sensor's field of view (FOV) and a cut-out occurs, the algorithm typically stops updating data and relies solely on a pre-defined fixed motion model for linear extrapolation. Due to the lack of correction with new observations, the state estimation error during the cut-out period is excessively large, ultimately affecting the continuity and accuracy of the intelligent driving system's perception of the dynamic environment.

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

[0004] This application provides a state prediction method, a vehicle, and a storage medium to at least solve the technical problem in the prior art where the state prediction error is large after the target leaves the sensor monitoring range, resulting in poor accuracy of the intelligent driving system in perceiving the dynamic environment.

[0005] According to one aspect of the embodiments of this application, a state prediction method is provided, comprising: obtaining the vanishing time of an edge target, wherein the edge target is located at the boundary of the sensor's field of view, and the vanishing time is used to represent the time when the edge target leaves the sensor's detection range; determining context parameters related to the edge target based on the vanishing time, wherein the context parameters are used to quantify the motion state, environmental prior constraints, and geometric positional relationships of the edge target at the vanishing time; constructing a noise covariance matrix based on the context parameters, wherein the noise covariance matrix is ​​used to differentially adjust the state prediction error accumulation rate corresponding to the edge target; predicting a state prediction vector and a state covariance matrix of the edge target based on the noise covariance matrix; and determining the motion state of the edge target based on the state prediction vector and the state covariance matrix.

[0006] Furthermore, the context parameters include: motion urgency factor, environmental constraint factor, and geometric factor. The motion urgency factor represents the maneuverability of the edge target at the vanishing moment, the environmental constraint factor represents the road structure information corresponding to the edge target at the vanishing moment, and the geometric factor represents the vanishing angle of the edge target relative to the sensor's field of view boundary at the vanishing moment.

[0007] Furthermore, the state prediction method also includes: at the disappearance time, obtaining a preset feature vector of the edge target, wherein the preset feature vector is bound and stored with the track identifier corresponding to the edge target; determining the disappearance duration of the edge target based on the disappearance time and the current time; determining the attenuation coefficient based on the preset feature vector; and determining the confidence score of the preset feature vector based on the disappearance duration and the attenuation coefficient, wherein the confidence score is used to re-associate the edge target.

[0008] Furthermore, the state prediction method also includes: re-associating edge targets based on the state prediction vector, preset feature vector, and confidence score; recovering the track corresponding to the edge target in response to successful association; and updating the noise covariance matrix in response to failed association.

[0009] Furthermore, the state prediction method also includes: determining the survival confidence of the edge target's track at the current time based on the survival confidence, state covariance matrix, and disappearance duration of the track at the disappearance time, wherein the survival confidence is used to determine whether to delete the track corresponding to the edge target; and deleting the track corresponding to the edge target in response to the survival confidence being lower than a preset confidence threshold.

[0010] Furthermore, the state prediction method also includes: in response to the position of the edge target exceeding the upper limit of the cooperative detection range of all sensors, deleting the track corresponding to the edge target.

[0011] Furthermore, the state prediction method also includes: in response to the disappearance duration exceeding a preset duration threshold, deleting the track corresponding to the edge target.

[0012] Furthermore, the state prediction method also includes: acquiring the motion state information of the target in the sensor coordinate system; determining the velocity vector of the target and the relative distance between the target and the sensor field of view boundary based on the motion state information; and determining the target as an edge target in response to the relative distance being less than a preset distance threshold and the velocity vector pointing outside the sensor field of view boundary.

[0013] According to another aspect of the embodiments of this application, a state prediction device is also provided, comprising: an acquisition module, configured to acquire the disappearance time of an edge target, wherein the edge target is located at the boundary of the sensor's field of view, and the disappearance time is used to represent the time when the edge target leaves the sensor's detection range; a first determination module, configured to determine context parameters related to the edge target based on the disappearance time, wherein the context parameters are used to quantify the motion state, environmental prior constraints, and geometric positional relationships of the edge target at the disappearance time; a construction module, configured to construct a noise covariance matrix based on the context parameters, wherein the noise covariance matrix is ​​used to differentially adjust the state prediction error accumulation rate corresponding to the edge target; a prediction module, configured to predict the state prediction vector and state covariance matrix of the edge target based on the noise covariance matrix; and a second determination module, configured to determine the motion state of the edge target based on the state prediction vector and the state covariance matrix.

[0014] Furthermore, the context parameters include: motion urgency factor, environmental constraint factor, and geometric factor. The motion urgency factor represents the maneuverability of the edge target at the vanishing moment, the environmental constraint factor represents the road structure information corresponding to the edge target at the vanishing moment, and the geometric factor represents the vanishing angle of the edge target relative to the sensor's field of view boundary at the vanishing moment.

[0015] Furthermore, the state prediction device also includes: a third determining module, used to acquire a preset feature vector of the edge target at the disappearance time, wherein the preset feature vector is bound and stored with the track identifier corresponding to the edge target; determine the disappearance duration of the edge target based on the disappearance time and the current time; determine the attenuation coefficient based on the preset feature vector; and determine the confidence score of the preset feature vector based on the disappearance duration and the attenuation coefficient, wherein the confidence score is used to re-associate the edge target.

[0016] Furthermore, the state prediction device also includes: an association module, used to re-associate the edge target based on the state prediction vector, the preset feature vector and the confidence score; and to recover the track corresponding to the edge target in response to successful association.

[0017] Furthermore, the state prediction device also includes an update module for updating the noise covariance matrix in response to association failure.

[0018] Furthermore, the state prediction device also includes: a deletion module, used to determine the survival confidence of the edge target's track at the current time based on the survival confidence, state covariance matrix and disappearance duration of the track at the disappearance time, wherein the survival confidence is used to determine whether to delete the track corresponding to the edge target; in response to the survival confidence being lower than a preset confidence threshold, the track corresponding to the edge target is deleted.

[0019] Furthermore, the deletion module is also used to delete the track corresponding to the edge target in response to the edge target's position exceeding the upper limit of the cooperative detection range of all sensors.

[0020] Furthermore, the deletion module is also used to delete the track corresponding to the edge target in response to the disappearance duration exceeding a preset duration threshold.

[0021] Furthermore, the state prediction device also includes: a fourth determination module, used to acquire the motion state information of the target in the sensor coordinate system; determine the velocity vector of the target and the relative distance between the target and the sensor field of view boundary based on the motion state information; and determine the target as an edge target in response to the relative distance being less than a preset distance threshold and the velocity vector pointing outside the sensor field of view boundary.

[0022] According to another aspect of the embodiments of this application, a vehicle is also provided, including: a memory storing an executable program; and a processor for running the executable program, wherein the executable program executes the state prediction methods of various embodiments of this application when running on the processor.

[0023] 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 execute the state prediction method of various embodiments of this application.

[0024] 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 state prediction methods of various embodiments of this application.

[0025] 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, which, when executed by a processor, implements the state prediction method in various embodiments of this application.

[0026] According to another aspect of the embodiments of this application, a computer program is also provided, which, when executed by a processor, implements the state prediction methods of various embodiments of this application.

[0027] In this embodiment, the disappearance time of the edge target is first obtained, where the edge target is located at the sensor's field of view boundary, and the disappearance time represents the time when the edge target leaves the sensor's detection range. Then, context parameters related to the edge target are determined based on the disappearance time, where the context parameters are used to quantify the motion state, environmental prior constraints, and geometric positional relationship of the edge target at the disappearance time. Then, a noise covariance matrix is ​​constructed based on the context parameters, where the noise covariance matrix is ​​used to differentially adjust the state prediction error accumulation rate corresponding to the edge target. Next, the state prediction vector and state covariance matrix of the edge target are predicted based on the noise covariance matrix. Finally, the motion state of the edge target is determined based on the state prediction vector and state covariance matrix. This achieves the purpose of non-uniform growth of the uncertainty in adaptive control state estimation during the disappearance of the edge target, thereby reducing the accumulation of state prediction error of the edge target and improving the correlation accuracy when the edge target reappears. This solves the technical problem in the prior art where the large state prediction error after the target leaves the sensor's monitoring range leads to poor accuracy of dynamic environment perception in intelligent driving systems. Attached Figure Description

[0028] 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:

[0029] Figure 1 This is a flowchart of a state prediction method according to an embodiment of this application;

[0030] Figure 2 This is a schematic diagram of an optional edge target determination method according to an embodiment of this application;

[0031] Figure 3 This is a schematic diagram of the operation of an optional state holder according to an embodiment of this application;

[0032] Figure 4 This is a schematic diagram of the operation of an optional re-association module according to an embodiment of this application;

[0033] Figure 5 This is a schematic diagram of a state prediction device according to an embodiment of this application. Detailed Implementation

[0034] 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.

[0035] 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.

[0036] According to an embodiment of this application, an embodiment of a state prediction 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.

[0037] This embodiment provides a state prediction method. Figure 1 This is a flowchart of a state prediction method according to an embodiment of this application, such as... Figure 1 As shown, the process includes the following steps:

[0038] Step S11: Obtain the disappearance time of the edge target, wherein the edge target is located at the boundary of the sensor's field of view, and the disappearance time is used to indicate the time when the edge target leaves the sensor's detection range;

[0039] In this embodiment, an edge target refers to a target that is about to leave or is in the boundary region of the sensor's field of view. For example, the criteria for determining an edge target are not only that the spatial position of the edge target is less than a set distance threshold from the FOV boundary, but also that in the last frame before the edge target disappears, the velocity vector of the edge target points outside the FOV; this is not limited here.

[0040] The disappearance time (t_disappear) refers to the specific point in time when an edge target becomes completely invisible from the effective detection range of a sensor. For example, the disappearance time can be used to represent the last time an edge target was successfully observed and recorded by the sensor; however, this is not a limitation.

[0041] Obtaining the disappearance time of an edge target can be understood as continuously monitoring targets near the FOV boundary. When a target is detected whose position in the last frame before disappearing is less than a set threshold from the FOV boundary, and whose velocity vector points outside the FOV, the target is determined to be an edge target with a high probability of being cut-out. Subsequently, the system records the timestamp of the last valid observation of the target as the disappearance time.

[0042] It can be seen that by accurately identifying edge targets about to leave the sensor's field of view and recording the moment of disappearance, the Cut-Out event is effectively distinguished from random missed detection. This provides a precise time anchor and triggering condition for the subsequent activation of a dedicated state-keeping mechanism, ensuring that the system only takes over tracking when the target is temporarily invisible due to going out of bounds, thus avoiding ineffective resource waste.

[0043] Step S12: Determine the context parameters related to the edge target based on the disappearance time. The context parameters are used to quantify the motion state, environmental prior constraints, and geometric positional relationships of the edge target at the disappearance time.

[0044] In this embodiment, context parameters refer to a set of variables used to quantify the multidimensional state characteristics of an edge target at the moment of disappearance, aiming to provide initialization conditions for subsequent state prediction. For example, context parameters include, but are not limited to: a motion urgency factor reflecting motion urgency, an environmental constraint factor reflecting environmental constraints, and a geometric factor reflecting geometric relationships.

[0045] Determining the contextual parameters related to edge targets based on the vanishing moment can be understood as follows: First, a motion urgency factor is calculated based on the velocity vector before vanishing; high-speed or highly maneuverable targets will be assigned a higher uncertainty growth weight. Second, the target's environment is analyzed in conjunction with high-precision map information; if the target is moving along the lane direction, an environmental constraint factor is used to limit the excessive expansion of lateral uncertainty. Finally, a geometric factor is determined based on the vanishing angle of the target relative to the field of view boundary; targets exiting directly opposite the boundary have greater positional uncertainty than those sliding tangentially.

[0046] It can be seen that by quantifying the urgency of motion, environmental constraints, and geometric relationships, we can provide credible initial conditions for state prediction, thereby reducing prediction bias during the Cut-Out period and improving the accuracy of subsequent state prediction.

[0047] Step S13: Construct a noise covariance matrix based on context parameters, wherein the noise covariance matrix is ​​used to differentially adjust the cumulative rate of state prediction error corresponding to edge targets.

[0048] In this embodiment, the noise covariance matrix is ​​a time-varying and state-dependent dynamic matrix that replaces the traditional fixed noise covariance during the target cut-out period, and can be represented by Q_adaptive. For example, the noise covariance matrix simulates the cumulative effect of prediction error over time by differentially amplifying or maintaining the noise variance of different state components (such as longitudinal position, lateral position, velocity, etc.), ensuring that the growth of uncertainty conforms to the actual motion logic of the edge target.

[0049] The state prediction error accumulation rate refers to how quickly the state covariance matrix (P_pred) of an edge target increases over time after it leaves the sensor's field of view. For example, for targets constrained by lanes, the lateral error accumulation rate is fast while the longitudinal error accumulation rate is slow; for high-speed maneuvering targets, the overall accumulation rate is even faster, but this is not a limitation here.

[0050] Constructing a noise covariance matrix based on contextual parameters can be understood as follows: the error accumulation of high-speed or highly maneuverable targets is accelerated by the motion urgency factor; environmental constraint factors are used to suppress the growth of longitudinal error while amplifying the lateral error in structured road scenarios to reflect lane constraints; and geometric factors are combined to increase the uncertainty for targets crossing the boundary directly, thereby constructing the noise covariance matrix.

[0051] As can be seen, the noise covariance matrix, as a diagonal matrix, amplifies each state component (position, velocity, etc.) with weight and grows slowly over time, thereby simulating the accumulation of credible prediction errors and avoiding covariance distortion caused by excessive conservatism or radicalism in traditional algorithms, thus laying the foundation for subsequent high-precision re-association.

[0052] Step S14: Predict the state prediction vector and state covariance matrix of the edge target based on the noise covariance matrix;

[0053] In this embodiment, the state prediction vector refers to the estimated value of the edge target state at the current moment calculated by the system based on the state estimate value of the previous moment and the state transition model, combined with the noise covariance matrix of the adaptive process, through a recursive prediction equation after the target enters the state maintenance phase following the Cut-Out.

[0054] For example, the state prediction vector (X_pred) contains key state information of the edge target during its disappearance, such as position coordinates and velocity vectors. It reflects the system's inference of the future position of the edge target based on the motion trend and contextual constraints before the edge target disappears, when new observation data is lacking.

[0055] The state covariance matrix (P_pred) is a statistic characterizing the uncertainty of the aforementioned state prediction vector, reflecting the distribution of the deviation between the predicted and actual states. For example, the state covariance matrix is ​​propagated not only through the traditional state transition equation but also superimposed with the noise covariance matrix of the adaptive process modulated by context parameters. The diagonal elements of the state covariance matrix represent the variance of the prediction error of each state component, while the off-diagonal elements represent the covariance between state components; this is not restricted here.

[0056] Predicting the state prediction vector and state covariance matrix of an edge target based on the noise covariance matrix can be understood as follows: taking the state at the vanishing moment as a baseline, the target's position and velocity are iteratively updated through the state transition equation to obtain the state prediction vector. Simultaneously, the noise covariance matrix of the adaptive process is introduced into the covariance prediction formula to obtain the state covariance matrix.

[0057] For example, the lateral error of edge targets constrained by lanes expands rapidly while the longitudinal error remains relatively stable. The resulting state covariance matrix reflects the prediction confidence during the Cut-Out period, providing a quantitative basis for subsequent reassociation and ensuring a balance between search range and false alarm risk when edge targets reappear.

[0058] It can be seen that by introducing an adaptive noise covariance matrix, the state covariance matrix expands non-uniformly, which not only reflects the rapid increase in uncertainty of high-speed or high-risk targets, but also uses environmental constraints to suppress the error spread in the restricted direction, thereby improving the accuracy of confidence assessment of the predicted state.

[0059] Step S15: Determine the motion state of the edge target based on the state prediction vector and the state covariance matrix.

[0060] In this embodiment, determining the motion state of an edge target based on the state prediction vector and the state covariance matrix can be understood as follows: First, the state prediction vector is analyzed, and combined with the velocity vector before disappearance, the potential motion trend of the edge target outside the field of view (such as whether it continues to move straight or turns) is inferred. Simultaneously, the geometric characteristics of the state covariance matrix (such as the shape and volume of an ellipsoid) are used to quantify the range of uncertainty in the prediction.

[0061] For example, if the state covariance volume grows too rapidly or the predicted position exceeds the cooperative detection range of all sensors, it is determined that the target may have actually disappeared or exceeded the effective tracking range; conversely, if the covariance growth is controlled and the movement trend is reasonable, the Cut-Out label is retained.

[0062] It can be seen that by integrating kinematic and uncertainty information, key criteria are provided for the decay or maintenance of the survival confidence of subsequent edge targets, preventing premature termination or incorrect retention of tracks due to prediction bias.

[0063] Through the above steps, the disappearance time of the edge target is first obtained, where the edge target is located at the boundary of the sensor's field of view, and the disappearance time represents the moment when the edge target leaves the sensor's detection range. Then, the context parameters related to the edge target are determined based on the disappearance time. The context parameters are used to quantify the motion state, environmental prior constraints, and geometric positional relationship of the edge target at the disappearance time. Then, a noise covariance matrix is ​​constructed based on the context parameters. The noise covariance matrix is ​​used to differentially adjust the accumulation rate of the state prediction error corresponding to the edge target. Next, the state prediction vector and state covariance matrix of the edge target are predicted based on the noise covariance matrix. Finally, the motion state of the edge target is determined based on the state prediction vector and state covariance matrix. This achieves the goal of non-uniform growth of the uncertainty in adaptive control state estimation during the disappearance of the edge target, thereby reducing the accumulation of the state prediction error of the edge target and improving the correlation accuracy when the edge target reappears. This solves the technical problem in the prior art where the large state prediction error after the target leaves the sensor's monitoring range leads to poor accuracy of dynamic environment perception in intelligent driving systems.

[0064] Furthermore, the context parameters include: motion urgency factor, environmental constraint factor, and geometric factor. The motion urgency factor represents the maneuverability of the edge target at the vanishing moment, the environmental constraint factor represents the road structure information corresponding to the edge target at the vanishing moment, and the geometric factor represents the vanishing angle of the edge target relative to the sensor's field of view boundary at the vanishing moment.

[0065] In this embodiment, the motion urgency factor (α_motion) is a parameter that quantifies the degree of maneuverability of a target at the moment of disappearance, and is typically calculated based on the velocity magnitude and acceleration magnitude before the moment of disappearance. High-speed or highly maneuverable edge targets have greater uncertainty in their future trajectories; therefore, the motion urgency factor will be given a higher weight, leading to a greater amplification of the accumulation rate of the state prediction error for edge targets when constructing the noise covariance matrix, in order to reflect the prediction difficulty brought about by drastic motion changes.

[0066] The environmental constraint factor (β_env) is a road structure feature parameter extracted using prior information such as high-precision maps. When an edge target disappears within a structured road (e.g., within a lane), the environmental constraint factor identifies that the movement of the edge target is physically restricted by the lane boundary. The environmental constraint factor guides the noise covariance matrix to maintain a small uncertainty growth along the lane direction (longitudinal), while allowing a larger uncertainty growth in the direction perpendicular to the lane (lateral), thus conforming to the physical laws of actual vehicle movement and avoiding the blind expansion of lateral errors.

[0067] The geometric factor (γ_geo) describes the geometric relationship between the vanishing point of an edge target and the FOV boundary, such as the angle between the velocity vector of the edge target and the boundary normal. If the edge target exits directly onto the boundary, the subsequent position of the edge target is more uncertain, and the geometric factor value is higher. If the edge target slides tangentially, the subsequent position of the edge target is relatively predictable, and the geometric factor value is lower. The geometric factor is used to differentially adjust the noise amplification factor of the position component in the covariance matrix, ensuring that the shape of the uncertainty ellipsoid matches the actual possible direction of motion of the edge target.

[0068] It can be seen that by introducing motion urgency factors, environmental constraint factors, and geometric factors, accurate modeling of the uncertainty in state prediction during the Cut-Out period is achieved. This allows for the adjustment of the error accumulation rate based on the differences in target mobility, road structure, and vanishing angle, thus avoiding covariance distortion caused by traditional linear growth.

[0069] Furthermore, the state prediction method also includes the following steps:

[0070] At the moment of disappearance, a preset feature vector of the edge target is obtained, wherein the preset feature vector is bound and stored with the track identifier corresponding to the edge target;

[0071] Determine the disappearance duration of the edge target based on the disappearance time and the current time;

[0072] The attenuation coefficient is determined based on the preset feature vector;

[0073] The confidence score of the preset feature vector is determined based on the disappearance time and decay coefficient, and the confidence score is used to re-associate the edge targets.

[0074] In this embodiment, the preset feature vector (F_vec) is a robust set of features extracted from the edge target at the moment of disappearance for identification, including but not limited to physical features (such as radar effective cross-section, optical size, and class probability), motion features (such as typical speed and acceleration pattern before disappearance), and contextual features (such as associated lane identifiers and disappearance area type). For example, the preset feature vector, as an important feature of the edge target, is protected and stored during the Cut-Out period to prevent the loss of identity features due to a lack of new observation data, providing a crucial basis for subsequent re-association.

[0075] A track identifier is a unique identifier assigned to each tracked edge target to distinguish different individuals in a complex multi-target environment. For example, in the Cut-Out mechanism, the track identifier is bound to a preset feature vector and a state preservation module to ensure that even if an edge target temporarily leaves the sensor's field of view, the edge target's historical trajectory, state information, and feature vector can still be uniquely indexed and managed by the system as a whole, avoiding confusion with other newly emerging targets or residual tracks.

[0076] The disappearance duration (Δt) refers to the time interval from when the target is determined to be cut-out and the disappearance time is recorded until the current evaluation time. For example, the disappearance duration is a key variable for measuring the time an edge target remains outside the sensor's field of view. As the disappearance duration increases, the probability of the edge target reappearing generally decreases; however, this is not a limitation here.

[0077] The attenuation coefficient (λ) is a constant or function parameter set according to the target type (e.g., vehicle, pedestrian) and environmental complexity. It controls the rate at which the confidence of the preset feature vector decreases over time. For example, a larger attenuation coefficient may correspond to a more complex environment or easily confused target type; this is not a limitation here. The attenuation coefficient reflects the system's subjective assessment of the reliability of target features that have not been observed for a long time, and is used to balance the sensitivity and false alarm rate of reassociation.

[0078] The confidence score (C_feat) is a numerical indicator that quantifies the effectiveness of the pre-defined feature vector in the re-association process. It is initially set to 1.0 and decays exponentially with the duration of disappearance. For example, when re-appearing edge targets for association matching, the confidence score is used as a weight in the association calculation: the higher the confidence score, the greater the weight of feature matching and the more significant its contribution to the association result; the lower the confidence score, the smaller the weight, and the more the judgment relies on spatial location and motion consistency, thereby effectively preventing erroneous associations caused by outdated features.

[0079] At the moment of disappearance, obtaining the preset feature vector of the edge target can be understood as immediately extracting the preset feature vector of the edge target at the moment of disappearance. This includes, but is not limited to, the physical attributes of the edge target (such as radar effective cross-section, optical pixel aspect ratio, and class probability), motion characteristics (such as the speed range and acceleration mode before disappearance), and contextual characteristics (such as associated lane identifiers and disappearance area type). The extracted preset feature vector is then strongly bound to the track identifier of the edge target and stored. This ensures that even if there is no new observation data update during the disappearance of the edge target, the system can still quickly obtain the feature information of the edge target through the track identifier index, preventing identifier switching or confusion caused by information loss.

[0080] Determining the disappearance duration of an edge target based on the disappearance time and the current time can be understood as recording the timestamp of the disappearance time and calculating the difference between the current time and the disappearance time in each state maintenance cycle to determine the disappearance duration of the edge target.

[0081] Determining the attenuation coefficient based on a preset feature vector can be understood as analyzing the target type (e.g., vehicle, pedestrian, bicycle) and environmental context (e.g., intersection, straight road, complex occlusion area) within the preset feature vector to determine the attenuation coefficient. For example, for pedestrians or targets in complex intersection environments, due to the variable and easily confused motion patterns in such environments, the system assigns a larger attenuation coefficient, meaning the feature information becomes invalid more quickly; while for vehicles or targets moving at a constant speed on straight roads, a smaller attenuation coefficient is assigned. By setting differentiated attenuation coefficients, the system can flexibly respond to uncertainties in different scenarios, avoiding both over-analysis of stable target features and outdated reliance on easily confused target features, thereby optimizing the accuracy and robustness of re-association.

[0082] Determining the confidence score of the preset feature vector based on the disappearance time and decay coefficient can be understood as follows, according to the formula C_feat(t)=exp(-λ The confidence score of the preset feature vector is calculated using Δt. When the edge target reappears and multi-sensor data is correlated, this confidence score is used as a weight in the correlation calculation. For example, the confidence score decays exponentially with the disappearance time Δt, but this is not a limitation.

[0083] It can be seen that by constructing a feature memory mechanism, the problem of identity feature loss of edge features during Cut-Out is solved. Physical features, motion features, and context features are bound to track identifiers, and an adaptive attenuation coefficient based on target type and environmental complexity is introduced to dynamically calculate the confidence score of the preset feature vector. This enables intelligent weighting of feature matching when the target is re-associated, significantly improving the association accuracy and stability when the Cut-Out target reappears in multi-sensor fusion.

[0084] Furthermore, the state prediction method also includes the following steps:

[0085] The edge targets are re-associated based on the state prediction vector, the preset feature vector, and the confidence score;

[0086] Upon successful association, the track corresponding to the edge target is restored;

[0087] In response to association failure, update the noise covariance matrix.

[0088] In this embodiment, re-associating edge targets based on state prediction vectors, preset feature vectors, and confidence scores can be understood as follows: the state prediction vectors generated by the adaptive mechanism provide spatial location priors, narrowing the association search range; and the preset feature vectors with time-sensitive weights are introduced, combined with confidence scores to quantify the matching degree of target identity features. By calculating the spatial distance and feature similarity between observed data and predicted states, and assigning feature weights, a probabilistic association gate is formed. This not only considers the consistency of geometric positions but also strengthens the continuity constraint of identity features, effectively distinguishing regressed targets outside the field of view from newly emerging targets in the scene, and significantly reducing the risk of false or missed associations due to prediction errors.

[0089] In response to successful association, restoring the track corresponding to the edge target can be understood as follows: when the association degree exceeds a preset threshold, it is considered a successful association, and the target re-enters the observation range of the multi-sensor system. For example, after successful association, the target's Cut-Out label is removed, and the target is re-integrated into the standard multi-sensor fusion tracking loop. The predicted state is updated using the latest observation data at the current moment using Kalman filtering, correcting state variables such as position and velocity, and resetting the state covariance matrix to reflect the reduced uncertainty brought about by the observations. Simultaneously, the track corresponding to the target is restored to ensure that the target participates in subsequent global data association and trajectory smoothing.

[0090] In response to association failure, updating the noise covariance matrix can be understood as follows: if association still fails after multi-sensor fusion and feature matching, it indicates that the target may not have reappeared or has undergone violent maneuvers beyond the prediction range. The system determines that the track is still in a Cut-Out state at the current moment and needs to continue maintaining the state prediction. For example, the system further updates the noise covariance matrix based on the current disappearance duration and context parameters, causing the noise covariance matrix to continue to expand according to an adaptive law, reflecting the increase in prediction uncertainty over time.

[0091] As can be seen, by re-associating edge targets based on state prediction vectors, preset feature vectors, and confidence scores, the track is restored in time when the association is successful, ensuring continuous tracking. If the association fails, the noise covariance matrix is ​​dynamically adjusted to reflect the uncertainty, which significantly improves the robustness and accuracy of Cut-Out target re-association and avoids misjudgment and trajectory interruption.

[0092] Furthermore, the state prediction method also includes the following steps:

[0093] The survival confidence of the edge target's track at the current time is determined based on the survival confidence, state covariance matrix and disappearance duration of the track at the disappearance time. The survival confidence is used to determine whether to delete the track corresponding to the edge target.

[0094] In response to a survival confidence level falling below a preset confidence threshold, the tracks corresponding to edge targets are deleted.

[0095] In this embodiment, survival confidence is a probability indicator that measures the possibility of a Cut-Out target track still existing during its disappearance period. Survival confidence dynamically decays over time and as prediction uncertainty increases. Survival confidence comprehensively considers the state of the edge target before disappearance (such as the initial value of survival confidence), the determinant of the state covariance matrix (reflecting the volume of the prediction error ellipsoid), and the duration since disappearance, thereby quantifying whether the track should continue to be retained.

[0096] For example, a higher survival confidence level indicates a greater likelihood of the marginal target reappearing, and the system is more inclined to retain the track. Conversely, a lower survival confidence level indicates that the marginal target may have permanently disappeared or has been unobservable for a long time, and the system tends to delete the track corresponding to the marginal target to reduce noise interference.

[0097] A preset confidence threshold (Th_low) is used to determine the lifecycle end point of the Cut-Out track. For example, the preset confidence threshold is usually set according to the safety requirements of the application scenario, sensor performance, and historical data statistics (such as 0.1, which is not limited here), aiming to balance the continuity of tracking and the real-time performance of the system, and to ensure the stability and efficiency of the multi-target tracking system.

[0098] Determining the survival confidence of an edge target's trajectory at the current moment based on its survival confidence at the time of disappearance, its state covariance matrix, and the duration of disappearance can be understood as using the initial survival confidence at the time of the edge target's disappearance as a benchmark, combined with the determinant of the state covariance matrix and the duration of the edge target's disappearance, to calculate the survival confidence of the edge target's trajectory at the current moment.

[0099] Responding to the fact that the survival confidence is lower than the preset confidence threshold, the deletion of the track corresponding to the edge target can be understood as follows: when the calculated current survival confidence is lower than the system's preset critical threshold (such as 0.1), it is determined that the edge target has been permanently disappeared or lost for a long time and no longer has tracking value. At this time, the track is deleted to avoid invalid tracks occupying system resources and reduce the error caused by invalid tracks.

[0100] It can be seen that by dynamically evaluating the survival probability of the track and retaining only the track with a survival confidence level higher than the preset confidence threshold, the real-time performance and stability of the system are improved, and the efficiency of multi-sensor fusion is optimized.

[0101] Furthermore, the state prediction method also includes the following steps:

[0102] In response to the edge target's position exceeding the upper limit of the cooperative detection range of all sensors, the track corresponding to the edge target is deleted.

[0103] In this embodiment of the application, in response to the position of the edge target exceeding the upper limit of the cooperative detection range of all sensors, deleting the track corresponding to the edge target can be understood as follows: when the predicted position of the edge target exceeds the maximum cooperative detection range of all sensors, it is determined that the edge target has left the observable area of ​​all sensors. At this time, deleting the track corresponding to the edge target can avoid meaningless long-term prediction and resource occupation for invalid targets.

[0104] It can be seen that by determining the spatial boundary, edge targets that have left the sensor monitoring area can be removed in time, avoiding invalid predictions and waste of resources, ensuring that the system focuses on the effective area, and significantly improving the efficiency and accuracy of multi-sensor fusion.

[0105] Furthermore, the state prediction method also includes the following steps:

[0106] In response to a disappearance duration exceeding a preset duration threshold, the track corresponding to the edge target is deleted.

[0107] In this embodiment, the preset duration threshold refers to the maximum allowable time for an edge target to disappear outside the sensor's field of view. For example, the preset duration threshold is set according to the security requirements of the application scenario, the sensor refresh rate, and the target's motion characteristics, such as 5 seconds or 10 seconds; this is not limited here.

[0108] Responding to the disappearance duration exceeding a preset duration threshold, deleting the track corresponding to the edge target can be understood as deleting the track corresponding to the edge target when the disappearance duration of the edge target exceeds the preset threshold, thereby ensuring the real-time performance, stability, and resource utilization efficiency of the multi-sensor fusion system and avoiding invalid tracks from interfering with normal tracking.

[0109] It can be seen that by managing the duration of disappearance, long-term lost targets can be removed in a timely manner, preventing the accumulation of prediction errors and confusion in association, ensuring the efficiency and stability of tracking, avoiding invalid tracks from interfering with normal perception, and improving the overall fusion accuracy.

[0110] Furthermore, the state prediction method also includes the following steps:

[0111] Acquire the motion state information of the target in the sensor coordinate system;

[0112] The velocity vector of the target and the relative distance between the target and the sensor's field of view boundary are determined based on the motion state information.

[0113] When the relative distance is less than a preset distance threshold and the velocity vector points outside the sensor's field of view, the target is determined to be an edge target.

[0114] In this embodiment of the application, motion state information refers to the set of real-time kinematic parameters of the edge target in the sensor coordinate system, which typically includes, but is not limited to, the target's position coordinates, velocity vector (including magnitude and direction), and acceleration vector.

[0115] Obtaining the target's motion state information in the sensor coordinate system can be understood as receiving the raw observation values ​​(such as distance, angle, Doppler velocity, etc.) output by the sensor, mapping them to a unified sensor coordinate system through coordinate transformation, and further calculating or filtering to obtain the target's precise position, instantaneous velocity vector, and acceleration information in the sensor coordinate system, ensuring that subsequent analysis is based on accurate target dynamic characteristics.

[0116] Determining the target's velocity vector based on motion state information, and the relative distance between the target and the sensor's field of view boundary, can be understood as directly extracting or calculating the target's velocity vector based on motion state information, thus clarifying the target's speed and direction of motion. Combining the target's current position with the sensor's preset field of view geometry model, the shortest straight-line distance from the target's current position to the field of view boundary is calculated.

[0117] The determination that a target is an edge target in response to a relative distance less than a preset distance threshold and a velocity vector pointing outside the sensor's field of view boundary can be understood as follows: when the relative distance between the target and the field of view boundary is less than a preset safety threshold and the target's velocity vector direction points outside the field of view (indicating that the target is moving outward), the target is determined to be an edge target.

[0118] It can be seen that by comprehensively analyzing the target's velocity direction and boundary distance, edge targets about to leave the sensor's field of view can be accurately identified, improving the accuracy of state estimation before the edge target disappears and the success rate of re-association, thus enhancing the system's robustness.

[0119] Figure 2This is a schematic diagram of an optional edge target determination method according to an embodiment of this application, such as... Figure 2 As shown, the input includes the target's position, velocity, and acceleration information in the sensor coordinate system. Based on the sensor's precise FOV geometric model, the distance between the target and the FOV boundary and the relative motion direction are calculated. It is determined whether the target was at the FOV edge in the last frame before disappearing (e.g., the distance to the boundary is less than the threshold d_th), and if the target's velocity vector points outside the FOV, then the target is classified as an edge target and labeled with a Cut-Out tag. Simultaneously, the edge target's vanishing point state X_disappear, vanishing time t_disappear, FOV boundary information, and the observed feature sequence of the N frames before vanishing are recorded.

[0120] Figure 3 This is a schematic diagram of the operation of an optional state holder according to an embodiment of this application, such as... Figure 3 As shown, when a target is identified as an edge target, the process switches to an independent state maintainer module, which takes over the state management of that edge target, replacing the traditional simple linear extrapolation. First, the state maintainer is initialized with the survival confidence and feature confidence corresponding to the edge target. Then, contextual parameters are extracted, including: motion urgency factor, environmental constraint factor, and geometric factor.

[0121] Specifically, the motion urgency factor α_motion is based on the velocity v and acceleration a before disappearance. α_motion = min(1.0, ||v|| / v_max + ||a|| / a_max), and for high-speed, highly maneuverable targets, uncertainty should increase faster. The environmental constraint factor β_env utilizes prior map or road information. If the target disappears on a structured road and moves along the lane direction, its longitudinal uncertainty increases slowly, while its lateral uncertainty increases rapidly (constrained by lane width). If it disappears in an open area, its uncertainty increases isotropically. The environmental constraint factor can be quantified by querying high-precision maps to obtain information such as lane direction and width. The geometric factor γ_geo is based on the target's vanishing angle relative to the FOV boundary. A target disappearing directly into the boundary (normal direction) has greater uncertainty regarding its future position than one sliding tangentially.

[0122] The process noise covariance matrix Q_adaptive is dynamically adjusted using the factors mentioned above. Specifically, during the Cut-Out period, Q_adaptive(t) = F(α_motion, β_env, γ_geo). Q_base g(t). Here, Q_base = diag([σ_x², σ_y², σ_vx², σ_vy²]), which represents the standard process noise. F(·) is a diagonal matrix used to differentially amplify the noise of each state component. For example, if β_env indicates that the target is moving along the lane, the noise amplification factor for the lateral position in the F matrix is ​​much greater than that for the longitudinal position. g(t) is a function that grows slowly over time (e.g., sqrt(t)), simulating the cumulative effect of prediction error over time, but more conservative than linear growth.

[0123] During the Cut-Out phase, the state holder periodically (within the system fusion cycle) performs a prediction: X_pred(k+1) = Φ X_pred(k), P_pred(k+1)=Φ P_pred(k) Φ^T+Q_adaptive(Δt), where Φ is the state transition matrix and Δt is the total time from t_disappear. The introduction of Q_adaptive(Δt) makes the growth of P_pred justified, rather than arbitrary.

[0124] While predicting the state, the features of edge targets are protected. Specifically, when an edge target disappears, a robust set of features F_vec is extracted, which may include: physical features: average radar cross section, optical size (pixel aspect ratio), estimated target category (vehicle, person, bicycle) probability; motion features: typical speed range before disappearance, acceleration pattern (e.g., uniform speed, gradual change); contextual features: associated lane identifier, disappearance area type (intersection, straight road). Furthermore, F_vec is bound to track identifiers for storage. A feature confidence score C_feat is set, initially set to 1.0. As the cut-out time Δt increases, this confidence score decays exponentially: C_feat(t) = exp(-λ Δt), where λ is the attenuation coefficient, which is related to the target type and environmental complexity. When targets are re-associated, C_feat will be used as the weight of the feature matching term in the association calculation. The lower the confidence level, the smaller the weight of the feature matching term.

[0125] A survival confidence C_track is maintained for each Cut-Out track. Specifically, C_track is C_0 (e.g., 0.9) when it disappears. C_track decays with time and the increase of uncertainty. The decay rate is positively correlated with the determinant of P_pred (i.e., the volume of the uncertainty ellipsoid) and Δt, and C_track(t) = C_0. exp(-η det(P_pred(t)) Δt).

[0126] The Cut-Out track is deleted when any of the following conditions are met: C_track is below the threshold Th_low (e.g., 0.1); the target's predicted position has exceeded the maximum cooperative detection range of all sensors; Δt exceeds the maximum allowable disappearance time T_max (set according to the application scenario, e.g., 5 seconds).

[0127] Furthermore, if fused information from other sensors (indirect observation) indirectly supports the existence of the edge target, the decay of C_track can be slowed down, or even slightly improved.

[0128] Finally, the motion state of the edge target is determined based on the state prediction vector and the state covariance matrix, for reference by the intelligent driving system.

[0129] Figure 4 This is a schematic diagram illustrating the operation of an optional re-association module according to an embodiment of this application, such as... Figure 4 As shown, when the cut-out target re-enters the sensor's field of view, the process enters the re-association phase. At this point, instead of directly using the traditional distance threshold, a comprehensive association metric is constructed: combining spatial location matching, temporal consistency, similarity of feature continuity, and current confidence. Through a multi-hypothesis probability association gate, the system distinguishes between the re-emerging cut-out target and the truly new target. If the association is successful, the normal tracking mode of the trajectory is restored, and the state is updated using the new observations. If the association fails or the confidence is too low, the noise covariance matrix continues to be updated.

[0130] Compared with traditional multi-sensor target fusion algorithms, this application significantly reduces state prediction errors during target disappearance by employing an uncertainty non-uniform expansion method based on the target disappearance context and a mechanism for freezing and decaying the storage of partial target attribute features. Furthermore, the cumulative error of the association gate is smaller. Simultaneously, the target re-association method for multi-sensor systems constructs a probabilistic association gate with fusion time consistency constraints, trajectory backtracking smoothness checks, and multiple hypothesis management, which can significantly improve the fusion accuracy of the cut-out.

[0131] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of the relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0132] According to an embodiment of this application, a state prediction device is provided. It should be noted that the device can be used to execute the above-described state prediction method.

[0133] Figure 5 This is a schematic diagram of a state prediction device according to an embodiment of this application, such as... Figure 5 As shown, the state prediction device 500 includes: an acquisition module 501, used to acquire the disappearance time of an edge target, wherein the edge target is located at the boundary of the sensor's field of view, and the disappearance time is used to represent the time when the edge target leaves the sensor's detection range; a first determination module 502, used to determine context parameters related to the edge target based on the disappearance time, wherein the context parameters are used to quantify the motion state, environmental prior constraints, and geometric positional relationships of the edge target at the disappearance time; a construction module 503, used to construct a noise covariance matrix based on the context parameters, wherein the noise covariance matrix is ​​used to differentially adjust the state prediction error accumulation rate corresponding to the edge target; a prediction module 504, used to predict the state prediction vector and state covariance matrix of the edge target based on the noise covariance matrix; and a second determination module 505, used to determine the motion state of the edge target based on the state prediction vector and the state covariance matrix.

[0134] Furthermore, the context parameters include: motion urgency factor, environmental constraint factor, and geometric factor. The motion urgency factor represents the maneuverability of the edge target at the vanishing moment, the environmental constraint factor represents the road structure information corresponding to the edge target at the vanishing moment, and the geometric factor represents the vanishing angle of the edge target relative to the sensor's field of view boundary at the vanishing moment.

[0135] Furthermore, the state prediction device also includes: a third determining module, used to acquire a preset feature vector of the edge target at the disappearance time, wherein the preset feature vector is bound and stored with the track identifier corresponding to the edge target; determine the disappearance duration of the edge target based on the disappearance time and the current time; determine the attenuation coefficient based on the preset feature vector; and determine the confidence score of the preset feature vector based on the disappearance duration and the attenuation coefficient, wherein the confidence score is used to re-associate the edge target.

[0136] Furthermore, the state prediction device also includes: an association module, used to re-associate the edge target based on the state prediction vector, the preset feature vector and the confidence score; and to recover the track corresponding to the edge target in response to successful association.

[0137] Furthermore, the state prediction device also includes an update module for updating the noise covariance matrix in response to association failure.

[0138] Furthermore, the state prediction device also includes: a deletion module, used to determine the survival confidence of the edge target's track at the current time based on the survival confidence, state covariance matrix and disappearance duration of the track at the disappearance time, wherein the survival confidence is used to determine whether to delete the track corresponding to the edge target; in response to the survival confidence being lower than a preset confidence threshold, the track corresponding to the edge target is deleted.

[0139] Furthermore, the deletion module is also used to delete the track corresponding to the edge target in response to the edge target's position exceeding the upper limit of the cooperative detection range of all sensors.

[0140] Furthermore, the deletion module is also used to delete the track corresponding to the edge target in response to the disappearance duration exceeding a preset duration threshold.

[0141] Furthermore, the state prediction device also includes: a fourth determination module, used to acquire the motion state information of the target in the sensor coordinate system; determine the velocity vector of the target and the relative distance between the target and the sensor field of view boundary based on the motion state information; and determine the target as an edge target in response to the relative distance being less than a preset distance threshold and the velocity vector pointing outside the sensor field of view boundary.

[0142] According to another aspect of the embodiments of this application, a vehicle is also provided, including: a memory storing an executable program; and a processor for running the executable program, wherein the executable program executes the state prediction method described in any of the above embodiments when running on the processor.

[0143] Optionally, in this embodiment, the processor in the vehicle can be configured to run a computer program to perform the following steps:

[0144] Step S11: Obtain the disappearance time of the edge target, wherein the edge target is located at the boundary of the sensor's field of view, and the disappearance time is used to indicate the time when the edge target leaves the sensor's detection range;

[0145] Step S12: Determine the context parameters related to the edge target based on the disappearance time. The context parameters are used to quantify the motion state, environmental prior constraints, and geometric positional relationships of the edge target at the disappearance time.

[0146] Step S13: Construct a noise covariance matrix based on context parameters, wherein the noise covariance matrix is ​​used to differentially adjust the cumulative rate of state prediction error corresponding to edge targets.

[0147] Step S14: Predict the state prediction vector and state covariance matrix of the edge target based on the noise covariance matrix;

[0148] Step S15: Determine the motion state of the edge target based on the state prediction vector and the state covariance matrix.

[0149] According to another aspect of the embodiments of this application, a computer-readable storage medium is also provided, wherein a computer program is stored in the computer-readable storage medium, wherein the computer program is configured to execute the state prediction method in any of the above embodiments when running on a computer or processor.

[0150] Optionally, in this embodiment, the computer-readable storage medium may be configured to store a computer program for performing the following steps:

[0151] Step S11: Obtain the disappearance time of the edge target, wherein the edge target is located at the boundary of the sensor's field of view, and the disappearance time is used to indicate the time when the edge target leaves the sensor's detection range;

[0152] Step S12: Determine the context parameters related to the edge target based on the disappearance time. The context parameters are used to quantify the motion state, environmental prior constraints, and geometric positional relationships of the edge target at the disappearance time.

[0153] Step S13: Construct a noise covariance matrix based on context parameters, wherein the noise covariance matrix is ​​used to differentially adjust the cumulative rate of state prediction error corresponding to edge targets.

[0154] Step S14: Predict the state prediction vector and state covariance matrix of the edge target based on the noise covariance matrix;

[0155] Step S15: Determine the motion state of the edge target based on the state prediction vector and the state covariance matrix.

[0156] 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 state prediction methods of various embodiments of this application.

[0157] Optionally, in this embodiment, the computer program in the above-described computer program product can be configured to perform the following steps when executed by a processor:

[0158] Step S11: Obtain the disappearance time of the edge target, wherein the edge target is located at the boundary of the sensor's field of view, and the disappearance time is used to indicate the time when the edge target leaves the sensor's detection range;

[0159] Step S12: Determine the context parameters related to the edge target based on the disappearance time. The context parameters are used to quantify the motion state, environmental prior constraints, and geometric positional relationships of the edge target at the disappearance time.

[0160] Step S13: Construct a noise covariance matrix based on context parameters, wherein the noise covariance matrix is ​​used to differentially adjust the cumulative rate of state prediction error corresponding to edge targets.

[0161] Step S14: Predict the state prediction vector and state covariance matrix of the edge target based on the noise covariance matrix;

[0162] Step S15: Determine the motion state of the edge target based on the state prediction vector and the state covariance matrix.

[0163] 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, which, when executed by a processor, implements the state prediction method in various embodiments of this application.

[0164] Optionally, in this embodiment, the computer program in the above-described computer program product can be configured to perform the following steps when executed by a processor:

[0165] Step S11: Obtain the disappearance time of the edge target, wherein the edge target is located at the boundary of the sensor's field of view, and the disappearance time is used to indicate the time when the edge target leaves the sensor's detection range;

[0166] Step S12: Determine the context parameters related to the edge target based on the disappearance time. The context parameters are used to quantify the motion state, environmental prior constraints, and geometric positional relationships of the edge target at the disappearance time.

[0167] Step S13: Construct a noise covariance matrix based on context parameters, wherein the noise covariance matrix is ​​used to differentially adjust the cumulative rate of state prediction error corresponding to edge targets.

[0168] Step S14: Predict the state prediction vector and state covariance matrix of the edge target based on the noise covariance matrix;

[0169] Step S15: Determine the motion state of the edge target based on the state prediction vector and the state covariance matrix.

[0170] According to another aspect of the embodiments of this application, a computer program is also provided, which, when executed by a processor, implements the state prediction methods of various embodiments of this application.

[0171] Optionally, in this embodiment, the computer program described above can be configured to perform the following steps when executed by the processor:

[0172] Step S11: Obtain the disappearance time of the edge target, wherein the edge target is located at the boundary of the sensor's field of view, and the disappearance time is used to indicate the time when the edge target leaves the sensor's detection range;

[0173] Step S12: Determine the context parameters related to the edge target based on the disappearance time. The context parameters are used to quantify the motion state, environmental prior constraints, and geometric positional relationships of the edge target at the disappearance time.

[0174] Step S13: Construct a noise covariance matrix based on context parameters, wherein the noise covariance matrix is ​​used to differentially adjust the cumulative rate of state prediction error corresponding to edge targets.

[0175] Step S14: Predict the state prediction vector and state covariance matrix of the edge target based on the noise covariance matrix;

[0176] Step S15: Determine the motion state of the edge target based on the state prediction vector and the state covariance matrix.

[0177] 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.

[0178] 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.

[0179] 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.

[0180] 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.

[0181] 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 USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0182] 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 state prediction method, characterized in that, include: Obtain the disappearance time of an edge target, wherein the edge target is located at the boundary of the sensor's field of view, and the disappearance time is used to represent the time when the edge target leaves the sensor's detection range; The context parameters related to the edge target are determined based on the disappearance time, wherein the context parameters are used to quantify the motion state, environmental prior constraints and geometric positional relationships of the edge target at the disappearance time; A noise covariance matrix is ​​constructed based on the context parameters, wherein the noise covariance matrix is ​​used to differentially adjust the state prediction error accumulation rate corresponding to the edge target; Predict the state prediction vector and state covariance matrix of the edge target based on the noise covariance matrix; The motion state of the edge target is determined based on the state prediction vector and the state covariance matrix.

2. The method according to claim 1, characterized in that, The context parameters include: motion urgency factor, environmental constraint factor, and geometric factor. The motion urgency factor represents the maneuverability of the edge target at the vanishing moment. The environmental constraint factor represents the road structure information corresponding to the edge target at the vanishing moment. The geometric factor represents the vanishing angle of the edge target relative to the sensor's field of view boundary at the vanishing moment.

3. The method according to claim 1, characterized in that, The method further includes: At the disappearance time, a preset feature vector of the edge target is obtained, wherein the preset feature vector is bound and stored with the track identifier corresponding to the edge target; The disappearance duration of the edge target is determined based on the disappearance time and the current time; The attenuation coefficient is determined based on the preset feature vector; The confidence score of the preset feature vector is determined based on the disappearance duration and the decay coefficient, wherein the confidence score is used to re-associate the edge target.

4. The method according to claim 3, characterized in that, The method further includes: The edge target is re-associated based on the state prediction vector, the preset feature vector, and the confidence score; Upon successful association, the track corresponding to the edge target is restored; In response to association failure, update the noise covariance matrix.

5. The method according to claim 3, characterized in that, The method further includes: The survival confidence of the edge target's track at the current time is determined based on the survival confidence of the track at the disappearance time, the state covariance matrix, and the disappearance duration, wherein the survival confidence is used to determine whether to delete the track corresponding to the edge target; In response to the survival confidence score falling below a preset confidence threshold, the track corresponding to the edge target is deleted.

6. The method according to claim 5, characterized in that, The method further includes: In response to the edge target's position exceeding the upper limit of the cooperative detection range of all sensors, the track corresponding to the edge target is deleted.

7. The method according to claim 5, characterized in that, The method further includes: In response to the disappearance duration exceeding a preset duration threshold, the track corresponding to the edge target is deleted.

8. The method according to claim 1, characterized in that, The method further includes: Acquire the motion state information of the target in the sensor coordinate system; The velocity vector of the target and the relative distance between the target and the sensor's field of view boundary are determined based on the motion state information. In response to the relative distance being less than a preset distance threshold and the velocity vector pointing outside the sensor's field of view boundary, the target is determined to be the edge target.

9. A vehicle, characterized in that, include: Memory, which stores executable programs; A processor for running the executable program, wherein the executable program, when running on the processor, performs the state prediction method as described in any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, wherein the computer program is configured to execute the state prediction method as described in any one of claims 1 to 8 when run on a computer or processor.

Images