Vehicle early warning method and device, electronic equipment and vehicle

Abstract

The application relates to a vehicle early warning method and device, electronic equipment and a vehicle. The method comprises the following steps: acquiring multi-dimensional sensing data of a target object and weights of the sensing data in real time, determining a target covariance matrix according to the sensing data and the corresponding weights, wherein the target covariance matrix is used for indicating a distribution range of a target object position estimation; performing smoothing filtering processing on the target covariance matrix to obtain a smoothed target covariance matrix; performing eigenvalue decomposition on the smoothed target covariance matrix to obtain an eigenvalue ratio, and updating the weights of the sensing data according to the eigenvalue ratio to iteratively update the target covariance matrix; and if it is determined that the target object has an intention to cut into a lane where a target vehicle currently locates according to the real-time updated and smoothed target covariance matrix, performing a pre-warning operation. The application improves the accuracy of pre-warning execution.

Description

Technical Field

[0001] This application relates to the field of vehicle control technology, and in particular to a vehicle early warning method, device, electronic equipment, and vehicle. Background Technology

[0002] During vehicle operation, the movement and driving trends of surrounding objects such as motor vehicles, pedestrians, and non-motorized vehicles directly affect the vehicle's driving safety. When driving in complex road conditions, it is necessary to identify and predict intrusive behaviors such as lane cutting and close-range crossings by surrounding objects in real time, and issue timely safety warnings to avoid potential collision risks.

[0003] Existing vehicle-mounted early warning solutions are prone to frequent jumps and violent oscillations in sensor data weights when calculating the position and motion status of surrounding targets. This can lead to unstable intrusion identification criteria, making it easy to make false or false judgments. Ultimately, this results in inaccurate vehicle safety warning outputs, making it difficult to ensure vehicle safety during driving. Summary of the Invention

[0004] This application provides a vehicle warning method, device, electronic equipment, and vehicle to solve the problem of inaccurate vehicle safety warnings.

[0005] In a first aspect, this application provides a vehicle warning method, the method comprising: The system acquires multidimensional sensor data and weights of each sensor data point in real time for the target object, and determines the target covariance matrix based on each sensor data point and its corresponding weight. The target covariance matrix is ​​used to indicate the distribution range of the target object's estimated location. The target covariance matrix is ​​subjected to a smoothing filter to obtain a smoothed target covariance matrix; The target covariance matrix after smoothing and filtering is subjected to eigenvalue decomposition to obtain the eigenvalue ratio, and the weights of each sensor data are updated according to the eigenvalue ratio to iteratively update the target covariance matrix. If, based on the real-time updated and smoothed target covariance matrix, it is determined that the target object intends to cut into the lane currently occupied by the target vehicle, a warning operation is executed.

[0006] Optionally, updating the weights of each sensor data based on the eigenvalue ratio includes: If the eigenvalue ratio is less than the first threshold, it is determined that the target object is in a shaking motion state, and the camera weight is increased while the radar weight is decreased. If the eigenvalue ratio is greater than or equal to the second threshold, it is determined that the target object is in a linear motion state, the radar weight is increased and the camera weight is decreased. If the feature ratio is greater than or equal to the first threshold and less than the second threshold, then the camera weight and the radar weight remain unchanged.

[0007] Optionally, After performing eigenvalue decomposition on the smoothed target covariance matrix to obtain the eigenvalue ratio, the method further includes: performing eigenvalue decomposition on the smoothed target covariance matrix to obtain the error distribution amplitude and error extension direction, wherein the error distribution amplitude is used to indicate the overall dispersion of the target object position estimation, and the error extension direction is used to indicate the potential change trend of the target object's motion; Determining whether the target object intends to enter the lane currently occupied by the target vehicle based on the real-time updated and smoothed target covariance matrix includes: if the error distribution amplitude is greater than a preset amplitude threshold, and the deviation between the error extension direction and the actual movement direction of the target object is greater than a preset deviation threshold, then it is determined that the target object intends to enter the lane currently occupied by the target vehicle.

[0008] Optionally, performing the alert operation includes: Determine the relative distance, relative speed, and spatiotemporal collision probability between the target vehicle and the target object; The warning probability threshold is dynamically adjusted based on the error distribution amplitude, wherein the error distribution amplitude is negatively correlated with the warning probability threshold; If the relative distance is less than a preset distance threshold, the relative motion speed is greater than a preset speed threshold, and the spatiotemporal collision probability is greater than the warning probability threshold, then a warning operation is performed according to the determined entry intention.

[0009] Optionally, determining the spatiotemporal collision probability between the target vehicle and the target object includes: Determine the position difference based on the current position of the target object and the current position of the target vehicle; The target covariance matrix of the target object and the covariance matrix of the target vehicle are superimposed to determine the joint covariance matrix; The Mahalanobis distance is determined based on the position difference and the joint covariance matrix. The spatiotemporal collision probability between the target vehicle and the target object is determined based on the Mahalanobis distance, wherein the Mahalanobis distance is negatively correlated with the spatiotemporal collision probability.

[0010] Optionally, after performing eigenvalue decomposition on the smoothed target covariance matrix to obtain the error distribution magnitude and error propagation direction, the method further includes: If a state frame that satisfies the condition that the feature value ratio is less than the target threshold and the error distribution amplitude is greater than the set amplitude threshold is detected to occur multiple times, then it is determined that the position fluctuation of the target object is abnormal, wherein the target threshold is less than the first threshold and the set amplitude threshold is greater than the preset amplitude threshold. The current state of the target object is corrected, and the calculation is re-iterated based on the corrected current state to obtain a new target covariance matrix.

[0011] Optionally, the multidimensional sensing data includes camera data and radar data, and the target covariance matrix is ​​determined based on each sensing data and its corresponding weight, including: Determine the camera weights corresponding to the camera data and the radar weights corresponding to the radar data; According to the camera weight and the radar weight, the camera data and the radar data are weighted and summed to obtain the perception fusion data; Based on the perception fusion data, the optimal state estimation is performed to update the current state of the target object, wherein the current state includes the current position and the current velocity; The current state of the target object is processed to obtain the target covariance matrix.

[0012] Secondly, this application provides a vehicle warning device, the device comprising: The determination module is used to acquire multi-dimensional sensing data of the target object and the weight of each sensing data in real time, and determine the target covariance matrix based on each sensing data and the corresponding weight, wherein the target covariance matrix is ​​used to indicate the distribution range of the target object's position estimate; The smoothing module is used to perform smoothing filtering on the target covariance matrix to obtain a smoothed target covariance matrix. The update module is used to perform eigenvalue decomposition on the smoothed target covariance matrix to obtain the eigenvalue ratio, and update the weight of each sensor data according to the eigenvalue ratio to iteratively update the target covariance matrix. The judgment and warning module is used to perform a warning operation if it is determined, based on the real-time updated and smoothed target covariance matrix, that the target object intends to cut into the lane currently occupied by the target vehicle.

[0013] Thirdly, this application provides an electronic device, comprising: at least one communication interface; at least one bus connected to the at least one communication interface; at least one processor connected to the at least one bus; and at least one memory connected to the at least one bus.

[0014] Fourthly, this application also provides a vehicle, the vehicle including a memory and a processor, wherein the memory stores executable program code, and the processor is used to call and execute the executable program code to implement the method described in any of the above.

[0015] The technical solutions provided in this application have the following advantages compared with the prior art: This application first constructs a target covariance matrix using multidimensional sensor data and their corresponding weights. This target covariance matrix represents the distribution fluctuation range of the target object's position estimate. Then, it performs a smoothing filter on the target covariance matrix to ensure that the eigenvalue ratios obtained from the decomposition of the smoothed target covariance matrix remain stable. Based on the stable eigenvalue ratios, the weights of each sensor data are dynamically updated, allowing for smooth weight adjustment without significant oscillations. Subsequently, combining the multidimensional sensor data and the updated stable weights, the target covariance matrix is ​​iteratively updated, ensuring that the target covariance matrix always remains stable and accurately reflects the actual movement state of the target object. Based on the stable and reliable target covariance matrix, it determines whether the target object intends to cut into the lane currently occupied by the target vehicle, improving the accuracy of intrusion intent determination and thus accurately executing warning actions. This application accurately restores the actual movement state of the target object through the stability of the weights corresponding to each sensor data, and improves the reliability of target intrusion intent determination by relying on a stable target covariance matrix, thereby enhancing the accuracy of warning execution. Attached Figure Description

[0016] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] One or more embodiments are illustrated by way of example with reference numerals in the accompanying drawings. These illustrations do not constitute a limitation on the embodiments. Elements with the same reference numerals in the drawings are denoted as similar elements. Unless otherwise stated, the figures in the drawings are not to be limited by scale.

[0019] Figure 1 This is a schematic diagram of the hardware environment of the vehicle warning system provided in the embodiments of this application; Figure 2 A flowchart of a vehicle early warning method provided in this application embodiment; Figure 3A flowchart of a vehicle early warning method provided in this application embodiment; Figure 4 This is a schematic diagram of the structure of a vehicle warning device provided in an embodiment of this application; Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0021] The following disclosure provides numerous different embodiments or examples for implementing various structures of this application. To simplify the disclosure, specific examples of components and arrangements are described below. These are merely examples and are not intended to limit the scope of this application. Furthermore, reference numerals and / or letters may be repeated in different examples. Such repetition is for simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or arrangements discussed.

[0022] The vehicle warning method provided in this application can be applied to processing scenarios where the vehicle processor operates independently, as well as to hardware environments where the vehicle and server process collaboratively. In collaborative processing scenarios, such as Figure 1 As shown, server 103 is connected to vehicle 101 via a network. Vehicle 101 sends multi-dimensional sensor data to server 103. Server 103 generates a warning command based on the multi-dimensional sensor data and feeds it back to vehicle 101. Vehicle 101 executes the warning command. Database 105 can be set up on the server or independently of the server to provide data storage services for server 103. The aforementioned network includes, but is not limited to, wide area network, metropolitan area network or local area network.

[0023] Furthermore, embodiments of this application can also be implemented as functional plug-ins, integrated into the vehicle processor, directly generating and executing warning commands based on multi-dimensional sensing data of the target object.

[0024] The following will describe in detail a vehicle warning method provided in this application embodiment, taking a processor applied to a vehicle as an example. Figure 2 As shown, the specific steps are as follows: Step 201: Real-time acquisition of multi-dimensional sensor data for the target object and the weight of each sensor data, and determination of the target covariance matrix based on each sensor data and the corresponding weight, wherein the target covariance matrix is ​​used to indicate the distribution range of the target object's position estimate; Step 202: Perform smoothing filtering on the target covariance matrix to obtain the smoothed target covariance matrix; Step 203: Perform eigenvalue decomposition on the smoothed target covariance matrix to obtain the eigenvalue ratio, and update the weights of each sensor data according to the eigenvalue ratio to iteratively update the target covariance matrix; Step 204: If, based on the real-time updated and smoothed target covariance matrix, it is determined that the target object intends to cut into the lane currently occupied by the target vehicle, then a warning operation is executed.

[0025] In step 201, the processor collects multi-dimensional sensing data of the target object in real time through on-board sensors. The multi-dimensional sensing data mainly includes camera data and radar data. The camera data is used to provide attribute information such as the target object's category, bounding box position, and semantic confidence. The target object's category refers to common traffic participants such as vehicles, pedestrians, and non-motorized vehicles. The radar data is used to provide the target object's three-dimensional coordinates, radial velocity, and RCS (Radar Cross Section) data. The radar data is a high-precision source of information on the relative distance and relative velocity between the target vehicle and the target object. The camera data and radar data complement each other to ensure comprehensive perception of the target object.

[0026] The weights corresponding to each sensor data point refer to the initial weights or updated weights during the iteration process. If it is an initial weight, then the weight is preset based on the environmental data, which includes data such as rainfall, light intensity, and fog.

[0027] The processor performs weighted fusion processing on the multi-dimensional sensor data according to the weights corresponding to each sensor data point, and then iteratively solves the problem using a state estimation algorithm to construct the target covariance matrix. The target covariance matrix is ​​essentially a second-order statistical matrix characterizing the position estimation error of the target object. It is used to quantify the distribution fluctuation range of the target object's position estimation. The target covariance matrix can be represented by an error ellipse model, corresponding to an error ellipse shape. This error ellipse shape can dynamically evolve into a near-circular or elongated straight line shape according to the target's motion state.

[0028] The magnitude of the elements in the target covariance matrix corresponds to the degree of dispersion of the position estimation error. The larger the value, the more scattered the position distribution of the target object, corresponding to a near-circular scattered trajectory; the smaller the value, the more concentrated the position distribution of the target object, corresponding to a narrow, straight trajectory.

[0029] The formula for calculating the target covariance matrix is: Σpos=Q*Λ*Q^T, where Λ=diag(λ1,λ2) and λ1≥λ2.

[0030] Where Σpos is the target covariance matrix, Q is the orthogonal eigenvector matrix, Λ is the eigenvalue diagonal matrix, Q^T is the transpose of matrix Q, λ1 is the largest eigenvalue of the target covariance matrix, and λ2 is the smallest eigenvalue of the target covariance matrix.

[0031] In step 202, the processor determines the corresponding time memory factor β based on the eigenvalue ratio k. This time memory factor β is used to adjust the weighted ratio of historical state prediction information and current perception fusion data. The eigenvalue ratio k and the time memory factor β adopt a piecewise mapping correspondence: if κ≥3.0, then β=0.8; if 1.5≤κ<3.0, then β=0.6; if κ<1.5, then β=0.4.

[0032] The processor performs weighted iterative correction on the target covariance matrix across multiple consecutive frames based on the time memory factor β. This processing method can preserve the true motion trend of the target object, while suppressing matrix value fluctuations caused by sudden changes in single-frame data, filtering out random noise in the sensor data, and eliminating high-frequency interference signals.

[0033] The formula for smoothing filtering is: Σpos_smooth=β×Σpos(t-1)+(1 β)×Σpos Where Σpos_smooth: the target covariance matrix after smoothing at time t, β: time memory factor, pos(t-1): the target covariance matrix after smoothing at time t-1, and Σpos: the target covariance matrix before filtering at time t.

[0034] After smoothing filtering, the shape of the target covariance matrix is ​​more stable, avoiding numerical jumps caused by noise interference. The smoothed covariance matrix can more realistically reflect the motion state of the target object.

[0035] In step 203, the processor performs eigenvalue decomposition on the smoothed target covariance matrix to obtain the eigenvalue ratio k. The formula for calculating the eigenvalue ratio k is k=λ1 / λ2. The eigenvalue ratio k is used to quantify the distribution pattern of the target object position estimation error, directly reflecting the motion state of the target object. The value of k determines the ellipse shape corresponding to the target object position error, and thus corresponds to different motion states.

[0036] When k≈1, it means that the largest eigenvalue λ1 and the smallest eigenvalue λ2 of the target covariance matrix are close in value. At this time, the error ellipse corresponding to the target covariance matrix is ​​close to a circle. This situation corresponds to the target object being in an unstable motion state, such as the irregular swaying of pedestrians or the random shaking of non-motorized vehicles.

[0037] when When the maximum eigenvalue λ1 of the covariance matrix is ​​much larger than the minimum eigenvalue λ2, the error ellipse corresponding to the target covariance matrix is ​​elongated. This situation corresponds to the target object being in a stable linear motion state, such as a vehicle traveling in a normal straight line.

[0038] The eigenvalue k has a mapping relationship with the camera weights and radar data weights. The processor can adjust the camera weights and radar data weights based on the eigenvalue k. The perception advantages of the camera and radar differ under different motion states. For example, when the target object is swaying or its motion is unstable (k≈1), the camera's visual perception is more stable and reliable, and it can more clearly capture the outline and trajectory of the target object; when the target object is moving smoothly in a straight line (k≈1), the camera's visual perception is more stable and reliable, and it can more clearly capture the outline and trajectory of the target object. In this case, radar has higher accuracy in measuring velocity and position, and can more accurately acquire the target's motion parameters. By adjusting the weights of the two types of sensors according to the k-value, the sensor advantages of each can be fully utilized, avoiding the limitations of a single sensor, and allowing the weighted fused sensor data to more accurately reflect the actual position and motion state of the target object.

[0039] Because the target covariance matrix undergoes smoothing filtering, the eigenvalues ​​k obtained from the matrix decomposition after smoothing are more stable. Consequently, the weights of the corresponding sensor data will not oscillate and will remain in a stable adjustment or unchanged state. Thus, when updating the target covariance matrix based on multidimensional sensor data and corresponding weights, the stability of the weights avoids matrix value jumps and shape distortions caused by abrupt weight changes. Therefore, when the processor iteratively updates the target covariance matrix based on the adjusted stable weights and multidimensional sensor data, the updated target covariance matrix can also remain stable and accurately reflect the actual motion state of the target object.

[0040] The processor continuously optimizes the target covariance matrix through iterative updates that ensure smooth filtering, stable k-values, stable weights, and stable covariance matrix, thereby ensuring that the target covariance matrix always accurately and stably reflects the distribution range of the target object's location estimate.

[0041] In step 204, the processor relies on the real-time iteratively updated and smooth and stable target covariance matrix to analyze the distribution characteristics of the target object's position estimate and determine whether the target object intends to cut into the lane currently occupied by the target vehicle.

[0042] Because the target covariance matrix remains stable, the corresponding error ellipse shape can stably reflect the motion state and positional distribution range of the target object, preventing misjudgments of the target's movement trend due to matrix fluctuations. Matrix stability directly determines that the judgment result of the target object's intrusion behavior will also remain stable, preventing misjudgments and false triggers caused by matrix fluctuations, effectively avoiding alarm disorder caused by weight oscillations and matrix instability.

[0043] When the processor determines that the target object intends to enter the vehicle's lane, it immediately activates the vehicle's warning system to issue a warning. The system uses HUD (Head-Up Display) visual prompts, cabin sound alarms, steering wheel vibrations, and other methods to convey intrusion risk information to the driver.

[0044] This application first constructs a target covariance matrix using multidimensional sensor data and their corresponding weights. This target covariance matrix represents the distribution fluctuation range of the target object's position estimate. Then, it performs a smoothing filter on the target covariance matrix to ensure that the eigenvalue ratios obtained from the decomposition of the smoothed target covariance matrix remain stable. Based on the stable eigenvalue ratios, the weights of each sensor data are dynamically updated, allowing for smooth weight adjustment without significant oscillations. Subsequently, combining the multidimensional sensor data and the updated stable weights, the target covariance matrix is ​​iteratively updated, ensuring that the target covariance matrix always remains stable and accurately reflects the actual movement state of the target object. Based on the stable and reliable target covariance matrix, it determines whether the target object intends to cut into the lane currently occupied by the target vehicle, improving the accuracy of intrusion intent determination and thus accurately executing warning actions. This application accurately restores the actual movement state of the target object through the stability of the weights corresponding to each sensor data, and improves the reliability of target intrusion intent determination by relying on a stable target covariance matrix, thereby enhancing the accuracy of warning execution.

[0045] As an optional implementation, in step 201, the multidimensional sensing data includes camera data and radar data. Determining the target covariance matrix based on each sensing data and its corresponding weight includes: Step S11: Determine the camera weights corresponding to the camera data and the radar weights corresponding to the radar data; Step S12: According to the camera weight and radar weight, perform weighted summation on the camera data and radar data to obtain the perception fusion data; Step S13: Perform state optimization estimation based on the perception fusion data and update the current state of the target object, where the current state includes the current position and the current velocity; Step S14: Process the current state of the target object to obtain the target covariance matrix.

[0046] In step S11, the processor determines the camera weight corresponding to the camera data and the radar weight corresponding to the radar data. The camera weight and radar weight are used to represent the trust ratio of the two types of sensor data in the data fusion process. The magnitude of the weight directly determines the contribution of the camera data and radar data to the subsequent fusion result.

[0047] In step S12, the processor performs weighted summation on the camera data and radar data according to the determined camera weights and radar weights, respectively, and linearly fuses the two types of heterogeneous sensor data according to their corresponding weights to obtain perceptual fusion data. This method can take into account the perceptual advantages of both cameras and radar, and suppress the error interference of single sensor data through weight allocation, forming more robust perceptual fusion data.

[0048] In step S13, the processor performs state-optimal estimation based on the perceptual fusion data and uses the standard Kalman filter formula to use the information generated after fusion to iteratively update the current state of the target object. The current state specifically includes the current position and current velocity of the target object.

[0049] The state update formula is: x_hat(t|t)=x_hat(t|t-1)+K(t)*y_fused(t) Where, x_hat(t|t): the target object state estimated at time t using the current perception fusion data; x_hat(t|t-1): the target object state predicted at time t based on the perception fusion data at time t-1; K(t): Kalman gain, used to balance the correction ratio between the state prediction value and the perception fusion data; y_fused(t): the perception fusion data at the current time.

[0050] In step S14, after updating the current state of the target object, the processor continues to use the Kalman filter covariance update formula to perform covariance recursion processing on the current state of the target object. The formula is as follows: Sigma_pos(t)=(IK(t)*H)*Sigma_pos(t|t-1) Where Sigma_pos(t): the target covariance matrix at time t after the update; I: identity matrix; K(t): Kalman gain; H: observation matrix, used to map the system state dimension to the observation dimension; Sigma_pos(t|t-1): the target covariance matrix obtained recursively at time t-1.

[0051] This application combines camera and radar weights, achieving effective fusion of the two types of sensor data through weighted summation to obtain perceptual fusion data. Then, a standard Kalman filter framework is introduced, utilizing perceptual fusion information combined with Kalman gain to achieve optimal state estimation of the target object's position and velocity. Simultaneously, the target covariance matrix is ​​updated in real-time using a covariance recursive formula. The entire process relies on fixed weight ratios and Kalman filter closed-loop iteration, achieving both the complementary advantages of multi-source sensor information and quantifying the error distribution range of position estimation, ensuring that the output target covariance matrix closely matches the actual motion characteristics of the target object.

[0052] As an optional implementation, step 203, updating the weights of each sensor data according to the eigenvalue ratio, includes: If the eigenvalue ratio is less than the first threshold, it is determined that the target object is in a shaking motion state, and the camera weight is increased while the radar weight is decreased. If the eigenvalue ratio is greater than or equal to the second threshold, it is determined that the target object is in a linear motion state, the radar weight is increased and the camera weight is decreased. If the eigenvalue ratio is greater than or equal to the first threshold and less than the second threshold, then the camera weight and radar weight remain unchanged.

[0053] If the processor determines that the eigenvalue ratio is less than the first threshold, for example, eigenvalue ratio k < 1.5, then it determines that the target object is in a swaying motion state. At this time, the error ellipse corresponding to the target covariance matrix is ​​close to a circle, the target position estimation has strong dispersion and large fluctuation amplitude, the radar observation data is affected by the target swaying, and the noise interference is relatively large. However, the visual observation information of the camera can more clearly capture the target outline and instantaneous motion trajectory, and has better stability and reliability. At this time, the weight of the camera is increased and the weight of the radar is decreased, so that the weight of the camera is not less than 0.7, and the data fusion is completed with visual perception as the main force.

[0054] If the processor determines that the eigenvalue ratio is greater than or equal to the second threshold, for example, if the eigenvalue ratio k ≥ 3.0, then it is determined that the target object is in a linear motion state. At this time, the error ellipse corresponding to the target covariance matrix is ​​narrow and elongated, the discreteness of the target position estimation is weak, and the motion state is stable. The radar's measurement accuracy of the target position and velocity is much better than that of the camera. At this time, the radar weight is increased and the camera weight is decreased, so that the radar weight ratio is not less than 0.7, and the data fusion is completed with radar perception as the main force.

[0055] If the processor determines that the eigenvalue ratio is greater than or equal to the first threshold and less than the second threshold, for example, if the eigenvalue ratio is 1.5≤k<3.0, then the target motion state is determined to be in a transitional state. At this time, the error ellipse shape corresponding to the target covariance matrix is ​​between a circle and a narrow shape, the target motion state is relatively stable, and the radar and camera observation information can form an effective complementarity. No additional parameter adjustment is required, and the camera weight and radar weight remain constant with the initial environmental weights.

[0056] This application constructs a three-stage fusion mode switching mechanism based on the eigenvalue ratio k. It automatically matches the corresponding weight adjustment strategy according to the numerical range of the eigenvalue ratio, and solves the optimal motion state of the target object based on the sensor data and the matched weights, achieving optimal allocation of radar and camera hardware sensing resources. This application can adaptively switch the fusion dominant mode according to three different working conditions: actual swaying motion, transitional motion, and linear stable motion of the target object. It dynamically adapts to the sensing advantages of the two types of sensors under different motion states, fundamentally solving the problem of decreased fusion accuracy caused by the mismatch between fixed weights and actual working conditions. This makes the fusion results of multi-source sensor data more closely match the real motion characteristics of the target, thereby ensuring the accuracy of the target covariance matrix.

[0057] As an optional implementation, after performing eigenvalue decomposition on the smoothed target covariance matrix to obtain the eigenvalue ratio, the method further includes: performing eigenvalue decomposition on the smoothed target covariance matrix to obtain the error distribution amplitude and error extension direction, wherein the error distribution amplitude is used to indicate the overall dispersion of the target object position estimation, and the error extension direction is used to indicate the potential change trend of the target object's motion. Based on the real-time updated and smoothed target covariance matrix, determining whether a target object intends to cut into the lane currently occupied by a target vehicle includes: if the error distribution amplitude is greater than a preset amplitude threshold, and the deviation between the error extension direction and the actual movement direction of the target object is greater than a preset deviation threshold, then it is determined that the target object intends to cut into the lane currently occupied by the target vehicle.

[0058] In this embodiment, the processor performs eigenvalue decomposition on the smoothed target covariance matrix to obtain the eigenvalue ratio k, and then further performs eigenvalue decomposition on the smoothed target covariance matrix to obtain the error distribution amplitude Acov and the error extension direction φ.

[0059] Among them, the error distribution amplitude Acov corresponds to the area of ​​the ellipse of the error ellipse model, and its calculation formula is: Acov=π×λ1×λ2. Acov can intuitively reflect the overall dispersion of the target object's position estimation. The larger the error distribution amplitude, the larger the ellipse area, indicating that the target object's position estimation is more discrete and the motion state is more unstable, corresponding to a near-circular scattered trajectory. The error extension direction φ corresponds to the major axis direction of the error ellipse model. The change in the error extension direction can intuitively reflect the potential change trend of the target object's motion.

[0060] The processor compares the error distribution amplitude Acov with the preset amplitude threshold Ath. If the error distribution amplitude is less than or equal to the preset amplitude threshold, it indicates that the target object's motion state is stable, and the corresponding error ellipse presents a narrow, straight trajectory with no obvious tendency to deviate from the surrounding lanes. The target object does not intend to cut into the lane currently occupied by the target vehicle. If the error distribution amplitude is greater than the preset amplitude threshold, it indicates that the target object's motion state is unstable, presenting a nearly circular, scattered trajectory with a tendency to deviate from the surrounding lanes.

[0061] Simultaneously, the processor calculates the deviation θdev between the error extension direction φ and the actual movement direction θ of the target object. This deviation represents the degree of deviation between the major axis of the error ellipse and the actual driving trajectory of the target object, reflecting the magnitude of the shift in the target's movement trend. If the deviation is less than or equal to a preset deviation threshold, for example, θdev < 45°, it indicates that the target object's movement trend is stable, the major axis of the error ellipse is basically consistent with the actual movement direction, and there is no obvious change in direction or tendency to cut in. At this time, it is determined that the target object has no intention to cut into the lane currently occupied by the target vehicle. If the deviation is greater than the preset deviation threshold, it indicates that the major axis of the error ellipse deviates from the actual movement trajectory of the target object, indicating that the target object's movement trend has changed and there is a tendency to move closer to the lane currently occupied by the target vehicle.

[0062] If the processor determines that the following conditions are met simultaneously: the error distribution amplitude is greater than the preset amplitude threshold, and the deviation between the error extension direction and the actual movement direction of the target object is greater than the preset deviation threshold, then it is determined that the target object has the intention to cut into the lane currently occupied by the target vehicle. By using dual-condition linkage for determination, misjudgment caused by single parameter judgment is avoided, ensuring the accuracy and reliability of intent determination.

[0063] This application, based on the eigenvalue ratio decomposition, further utilizes the target covariance matrix to construct a covariance ellipse. The error distribution amplitude is quantified by calculating the area of ​​the covariance ellipse, and the error extension direction is determined by solving for the major axis direction angle of the covariance ellipse. The overall dispersion of the target object's position estimation is constrained by comparing the error distribution amplitude with a preset amplitude threshold. The deviation of the angle between the error extension direction and the actual movement direction of the target object is used to measure the degree of deviation of the target's potential movement trend. By relying on the dual thresholds of error distribution amplitude and direction deviation for joint judgment, the intention of the target object to enter the lane is identified, avoiding the one-sidedness of single-parameter judgment and improving the sensitivity and accuracy of lane entry intention determination.

[0064] As an optional implementation, in step 204, performing the warning operation includes: Step S21: Determine the relative distance, relative speed, and spatiotemporal collision probability between the target vehicle and the target object; Step S22: Dynamically adjust the early warning probability threshold based on the error distribution amplitude, wherein the error distribution amplitude and the early warning probability threshold are negatively correlated; Step S23: If the relative distance is less than the preset distance threshold, the relative motion speed is greater than the preset speed threshold, and the spatiotemporal collision probability is greater than the warning probability threshold, then execute the warning operation according to the determined cutting intention.

[0065] In step S21, the processor calculates and determines the relative distance, relative speed, and spatiotemporal collision probability between the target vehicle and the target object in real time. The spatiotemporal collision probability is used to quantify the likelihood of a collision between the target vehicle and the target object within a preset time period in the future.

[0066] In step S22, the processor dynamically adjusts the warning probability threshold Pth based on the error distribution amplitude Acov obtained by splitting the target covariance matrix after smoothing and filtering. The formula for calculating the warning probability threshold is as follows: Pth=max(0.65,0.85 α×(Acov / Aref)) Where Pth is the warning probability threshold, Acov is the error distribution amplitude, and Aref is the reference area of ​​the variance ellipse.

[0067] The magnitude of the error distribution is negatively correlated with the warning probability threshold. Setting this negative correlation adapts to the uncertainty of target object position estimation, enabling adaptive adjustment of the warning threshold. When the error distribution magnitude is large, it indicates a high degree of overall dispersion in the target object position estimation and strong position uncertainty. In this case, the processor actively lowers the warning probability threshold, reducing the difficulty of triggering the warning and ensuring that potential collision risks can be captured in a timely manner even if the target position fluctuates significantly, avoiding missed warnings. Conversely, when the error distribution magnitude is small, it indicates a more concentrated target object position estimation, a more stable motion state, and low position uncertainty. In this case, the processor correspondingly raises the warning probability threshold, increasing the warning triggering threshold and avoiding false warnings triggered when the target state is stable and the risk is low, reducing unnecessary interference to the driver. This dynamic adjustment method breaks the limitations of the traditional one-size-fits-all fixed warning threshold, allowing the warning threshold to accurately match the actual risk level of the target.

[0068] In step S23, the processor sets dual warning constraints. The physical constraints involve a relative distance less than a preset distance threshold and a relative speed greater than a preset speed threshold, for example, a relative distance d < 5m and a relative speed vrel > 15km / h. This physical constraint is used to filter out high-risk basic scenarios involving close-range, high-speed approach. The probabilistic constraint is that the spatiotemporal collision probability Pcollision is greater than the warning probability threshold Pth, used to further accurately determine the collision risk level. When both the physical and probabilistic constraints are met, the processor, combined with the previously determined intention of the target object to enter the lane currently occupied by the target vehicle, formally executes the warning operation, linking the vehicle's warning system to deliver risk warning information to the driver.

[0069] This application constructs a dual judgment logic of basic physical constraints and dynamic probabilistic constraints when performing early warning operations. Firstly, it filters out high-risk basic scenarios by setting preset distance and speed thresholds, avoiding invalid warnings in low-risk scenarios. Secondly, it dynamically adapts the warning threshold in real time according to the negative correlation between the error distribution amplitude and the warning probability threshold. This ensures that the greater the fluctuation in the target object's position and the more difficult the risk is to predict, the lower the warning threshold, ensuring no risk is overlooked; conversely, the more stable the target object's position and the more controllable the risk, the higher the warning threshold, avoiding false warnings. Compared to traditional fixed warning threshold solutions, this application achieves adaptive matching of the warning threshold, improving the timeliness of warnings in high-risk scenarios and reducing the false warning rate under stable operating conditions. This makes the warning triggering logic more closely resemble real driving conditions, improving the accuracy and reliability of the warning operation and further ensuring driving safety.

[0070] As an optional implementation, after performing eigenvalue decomposition on the smoothed target covariance matrix to obtain the error distribution amplitude, the method further includes: if a state frame satisfying the eigenvalue ratio being less than a target threshold and the error distribution amplitude being greater than a set amplitude threshold is detected to occur multiple times continuously, then it is determined that the position fluctuation of the target object is abnormal, wherein the target threshold is less than a first threshold and the set amplitude threshold is greater than a preset amplitude threshold; the current state of the target object is corrected, and a new target covariance matrix is ​​updated by re-iteratio calculation based on the corrected current state.

[0071] In addition to basic physical constraints and dynamic probabilistic constraints, this application also sets up feedback recalibration. The feedback recalibration process is as follows: the processor performs eigenvalue decomposition on the smoothed and filtered target covariance matrix to obtain the error distribution amplitude and error extension direction. If the eigenvalue ratio k is detected to be less than the target threshold and the error distribution amplitude Acov is detected to be greater than the set amplitude threshold, and the state frames corresponding to the above two abnormal conditions appear continuously multiple times, such as appearing continuously for 3 frames, then the processor determines that the position fluctuation of the target object is abnormal. At this time, the target object is in a continuous and unstable shaking state, and the dispersion of the position estimation is far beyond the normal range, which can easily lead to the distortion of subsequent intrusion intent judgment and false triggering of warnings.

[0072] When the target object's position fluctuation is determined to be abnormal, the processor performs a backsliding and smoothing correction on the target object's current state, appropriately weakens the confidence weight of the current abnormal observation information, constrains the correction magnitude of the Kalman gain, and suppresses the interference of single-frame fluctuation noise on position and velocity state estimation. This corrects and calibrates the target object's current state, eliminates the state estimation bias caused by abnormal position fluctuation, and then, based on the corrected target object's current state, iteratively solves the new target covariance matrix using the Kalman filter covariance update formula, thus completing the update and calibration of the target covariance matrix.

[0073] In addition, during the abnormal operating condition identification process, the processor simultaneously implements false trigger suppression. When the target object is detected as a vehicle traveling in the same direction and at the same speed as the target vehicle, or a parked vehicle on the roadside, a fixed obstacle, or when a sensor malfunction causes abnormal data, the processor will forcibly suppress the warning trigger and completely eliminate the false alarm phenomenon.

[0074] As an optional implementation method, determining the spatiotemporal collision probability between the target vehicle and the target object includes: Step S31: Determine the position difference based on the current position of the target object and the current position of the target vehicle; Step S32: Perform a superposition operation on the target covariance matrix of the target object and the covariance matrix of the target vehicle to determine the joint covariance matrix; Step S33: Determine the Mahalanobis distance based on the position difference and the joint covariance matrix; Step S34: Determine the spatiotemporal collision probability between the target vehicle and the target object based on the Mahalanobis distance, where the Mahalanobis distance is negatively correlated with the spatiotemporal collision probability.

[0075] In step S31, the processor obtains the current position of the target object and the current position of the target vehicle, performs a difference operation on the two sets of position information, and solves the position difference between the two to characterize the degree of spatial offset between the target vehicle and the target object.

[0076] In step S32, the processor performs a superposition operation on the target covariance matrix corresponding to the target object and the covariance matrix corresponding to the target vehicle to synthesize and obtain the joint covariance matrix. The joint covariance matrix can simultaneously incorporate the position estimation uncertainty and motion uncertainty of both the target vehicle and the target object, realizing the fusion representation of the two-end error information.

[0077] In step S33, the processor uses the position difference as the spatial offset input, combines it with the constructed joint covariance matrix, and calculates the Mahalanobis distance according to the Mahalanobis distance calculation logic.

[0078] The formula for calculating Mahalanobis distance is as follows: dM=sqrt(Delta_X^T*Sigma_combined^(-1)*Delta_X) Where dM: Mahalanobis distance, Delta_X: position difference vector, Delta_X^T: transpose of position difference vector, and Sigma_combined^(-1): inverse of joint covariance matrix.

[0079] In step S34, the processor calculates the spatiotemporal collision probability based on the Mahalanobis distance dM between the vehicle and the target trajectory using the chi-square cumulative function. The formula for calculating the spatiotemporal collision probability is as follows: Pcollision=1 Fχ2(dM²) Where Pcollision: spatiotemporal collision probability, dM: Mahalanobis distance; Fχ2( ): Chi-square cumulative function with 2 degrees of freedom.

[0080] As can be seen from the above formula, Mahalanobis distance is negatively correlated with the probability of spacetime collision. The smaller the Mahalanobis distance, the stronger the spatial correlation and the higher the probability of spacetime collision; the larger the Mahalanobis distance, the more obvious the spatial deviation and the lower the probability of spacetime collision.

[0081] This application first obtains the position difference, then superimposes the covariance matrices of both parties to construct a joint covariance matrix, fully considering the positional and motion uncertainties of the target vehicle and the target object. Based on the position difference and the joint covariance matrix, the Mahalanobis distance is calculated, and then the spatiotemporal collision probability is solved using a chi-square cumulative function mapping. The negative correlation between Mahalanobis distance and collision probability is used to quantify the risk level. Compared to traditional methods that rely solely on fixed distances and speed thresholds, this application incorporates dual-end state estimation error information, characterizing vehicle collision risk from a probabilistic statistical perspective. This avoids the shortcomings of single physical threshold judgments being one-sided and lacking precision, thus improving the accuracy of spatiotemporal collision probability.

[0082] This application also provides a schematic diagram of the overall process of a vehicle warning method, such as... Figure 3 As shown, the steps include the following.

[0083] Step 301: Determine the target covariance matrix based on the multidimensional sensor data and corresponding weights.

[0084] By relying on multi-dimensional sensor data from radar and cameras, and combining them with corresponding weights, a target covariance matrix is ​​constructed.

[0085] Step 302: Perform matrix smoothing on the target covariance matrix.

[0086] By performing time-series smoothing filtering on the target covariance matrix through the outer loop, the matrix value jump caused by single-frame sensing noise is suppressed.

[0087] Step 303: Perform feature decomposition on the smoothed matrix to obtain the eigenvalue ratio k, the error distribution magnitude Acov, and the error extension direction φ.

[0088] Eigenvalue decomposition and geometric analysis are performed on the smoothed target covariance matrix to calculate three types of characteristic parameters: eigenvalue ratio k, error distribution magnitude Acov, and error extension direction φ.

[0089] Step 304: Update the weights based on the eigenvalue ratio k.

[0090] The corresponding weights are updated based on the interval distribution of the eigenvalue ratio k.

[0091] Step 305: Dynamically adjust the early warning probability threshold Pth based on the error distribution amplitude Acov, and update the target covariance matrix based on the updated weights.

[0092] By using the inner loop, the negative correlation between the error distribution amplitude Acov and the early warning probability threshold is utilized to adaptively adjust the early warning probability threshold. At the same time, the target covariance matrix is ​​iteratively updated using the updated weights.

[0093] Step 306: Use the target object's current position data and covariance matrix to solve for the Mahalanobis distance dM, and then determine the spatiotemporal collision probability Pcollision.

[0094] The position difference between the target object and the target vehicle is determined based on their current positions. The covariance matrices of both parties are superimposed to obtain the joint covariance matrix. The Mahalanobis distance dM is then calculated based on this matrix, and the spatiotemporal collision probability Pcollision is solved using the chi-square distribution cumulative function.

[0095] Step 307: Determine the lane cutting intention based on the error distribution amplitude Acov and the error extension direction φ.

[0096] If the error distribution amplitude Acov is greater than the preset amplitude threshold, and the angle deviation between the error extension direction φ and the actual movement direction of the target object is greater than the preset deviation threshold, then it is determined that the target object has the intention to cut into the lane where the target vehicle is currently located.

[0097] Step 308: Set dual constraints as the basis for triggering the early warning.

[0098] Set physical constraints and probabilistic constraints; the physical constraints are that the relative distance and relative motion speed meet the preset thresholds, and the probabilistic constraints are that the spatiotemporal collision probability Pcollision is greater than the dynamically adjusted warning probability threshold Pth.

[0099] Step 309: Combine multiple conditions to make a joint judgment and execute the early warning operation.

[0100] When both physical and probabilistic constraints are met, and the target is determined to have the intention to cut into the lane, the system executes the corresponding vehicle warning operation and performs false trigger suppression for stationary targets, co-moving targets, and sensor anomaly scenarios.

[0101] In this application, eigenvalue decomposition is used to accurately generate three types of motion features from the smoothed target covariance matrix: eigenvalue ratio k, error distribution amplitude Acov, and error extension direction φ. These features comprehensively characterize the motion morphology, positional dispersion, and motion trend of the target object, effectively avoiding fusion mismatch issues. A three-stage switching mechanism is used to dynamically adjust the weights of the camera and radar in segments based on the eigenvalue ratio k, achieving smooth weight transitions without abrupt changes and completely eliminating weight oscillation defects. A dual-loop feedback architecture is constructed: the outer loop adaptively adjusts the weights based on the eigenvalue ratio k, achieving dynamic adaptation of global parameters; the inner loop breaks through the limitations of traditional rigid thresholds. Probability calculation is introduced, fusing the positional difference between target objects and the joint covariance matrix. The spatiotemporal collision probability Pcollision is accurately solved using Mahalanobis distance and the chi-square distribution cumulative function, quantifying the collision risk caused by positional and motion uncertainties. Combined with a dynamically adjusted preset collision threshold, the lane cutting intention is jointly determined, solving the intention lag problem and improving the reliability of lane cutting warnings in complex driving scenarios. This application fundamentally solves the four major pain points of traditional solutions—fusion mismatch, weight oscillation, threshold rigidity, and intent lag—through the coordinated linkage of four core algorithmic links: eigenvalue decomposition, three-stage switching, dual closed-loop feedback, and probability calculation.

[0102] Based on the same technical concept, this application also provides a vehicle warning device, such as... Figure 4 As shown, the device includes: The determination module 401 is used to acquire multi-dimensional sensing data of the target object and the weight of each sensing data in real time, and determine the target covariance matrix based on each sensing data and the corresponding weight. The target covariance matrix is ​​used to indicate the distribution range of the target object's position estimate. The smoothing module 402 is used to perform smoothing filtering on the target covariance matrix to obtain the smoothed target covariance matrix. The update module 403 is used to perform eigenvalue decomposition on the smoothed target covariance matrix to obtain the eigenvalue ratio, and update the weights of each sensor data according to the eigenvalue ratio to iteratively update the target covariance matrix. The judgment and warning module 404 is used to perform a warning operation if it is determined, based on the real-time updated and smoothed target covariance matrix, that the target object intends to cut into the lane currently occupied by the target vehicle.

[0103] Optionally, update module 403 is used for: If the eigenvalue ratio is less than the first threshold, it is determined that the target object is in a shaking motion state, and the camera weight is increased while the radar weight is decreased. If the eigenvalue ratio is greater than or equal to the second threshold, it is determined that the target object is in a linear motion state, the radar weight is increased and the camera weight is decreased. If the eigenvalue ratio is greater than or equal to the first threshold and less than the second threshold, then the camera weight and radar weight remain unchanged.

[0104] Optionally, the device is also used to: perform eigenvalue decomposition on the smoothed target covariance matrix to obtain the error distribution magnitude and error extension direction, wherein the error distribution magnitude is used to indicate the overall dispersion of the target object position estimation, and the error extension direction is used to indicate the potential change trend of the target object's motion. The judgment and warning module 404 is used to: if the error distribution amplitude is greater than a preset amplitude threshold, and the deviation between the error extension direction and the actual movement direction of the target object is greater than a preset deviation threshold, then it is determined that the target object has the intention to cut into the lane where the target vehicle is currently located.

[0105] Optionally, the determination and warning module 404 is used for: Determine the relative distance, relative speed, and spatiotemporal collision probability between the target vehicle and the target object; The warning probability threshold is dynamically adjusted based on the error distribution amplitude, where the error distribution amplitude and the warning probability threshold are negatively correlated. If the relative distance is less than a preset distance threshold, the relative motion speed is greater than a preset speed threshold, and the probability of spatiotemporal collision is greater than a warning probability threshold, then a warning operation will be performed according to the determined entry intention.

[0106] Optionally, the determination and early warning module 404 is specifically used for: Determine the position difference based on the current position of the target object and the current position of the target vehicle; The target covariance matrix of the target object and the covariance matrix of the target vehicle are superimposed to determine the joint covariance matrix; The Mahalanobis distance is determined based on the positional difference and the joint covariance matrix. The spatiotemporal collision probability between the target vehicle and the target object is determined based on Mahalanobis distance, where Mahalanobis distance is negatively correlated with the spatiotemporal collision probability.

[0107] Optionally, the device is also used for: If a state frame that satisfies the condition that the feature ratio is less than the target threshold and the error distribution amplitude is greater than the set amplitude threshold is detected to appear multiple times, then it is determined that the position fluctuation of the target object is abnormal. Here, the target threshold is less than the first threshold and the set amplitude threshold is greater than the preset amplitude threshold. The current state of the target object is corrected, and the calculation is re-iterated based on the corrected current state to obtain a new target covariance matrix.

[0108] Optionally, the multidimensional sensing data includes camera data and radar data, and the determination module 401 is used for: Determine the camera weights corresponding to camera data and the radar weights corresponding to radar data; According to the camera weight and radar weight, the camera data and radar data are weighted and summed to obtain the perception fusion data. The current state of the target object is updated by performing state-optimal estimation based on the perception fusion data, where the current state includes the current position and the current velocity. The current state of the target object is processed to obtain the target covariance matrix.

[0109] like Figure 5 As shown, this application provides an electronic device including a processor 501, a communication interface 502, a memory 503, and a communication bus 504, wherein the processor 501, the communication interface 502, and the memory 503 communicate with each other through the communication bus 504.

[0110] Memory 503 is used to store computer programs.

[0111] In one embodiment of this application, the processor 501, when executing the program stored in the memory 503, implements the vehicle warning method provided in any of the foregoing method embodiments.

[0112] This application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the vehicle warning method provided in any of the foregoing method embodiments.

[0113] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and 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 network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0114] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented using software plus a general-purpose hardware platform, or of course, using hardware. Based on this understanding, the above technical solutions, in essence or the parts that contribute to the related technology, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0115] It should be understood that the terminology used herein is for the purpose of describing particular exemplary embodiments only and is not intended to be limiting. Unless the context clearly indicates otherwise, the singular forms “a,” “an,” and “described” as used herein may also include the plural forms. The terms “comprising,” “including,” “containing,” and “having” are inclusive and therefore indicate the presence of the stated features, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, elements, components, and / or combinations thereof. The method steps, processes, and operations described herein are not construed as requiring them to be performed in a particular order described or illustrated unless the order of performance is explicitly indicated. It should also be understood that additional or alternative steps may be used.

[0116] The above description is merely a specific embodiment of this application, enabling those skilled in the art to understand or implement this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features claimed herein.

Claims

1. A vehicle early warning method, characterized in that, The method includes: The system acquires multidimensional sensor data and weights of each sensor data point in real time for the target object, and determines the target covariance matrix based on each sensor data point and its corresponding weight. The target covariance matrix is ​​used to indicate the distribution range of the target object's estimated location. The target covariance matrix is ​​subjected to a smoothing filter to obtain a smoothed target covariance matrix; The target covariance matrix after smoothing and filtering is subjected to eigenvalue decomposition to obtain the eigenvalue ratio, and the weights of each sensor data are updated according to the eigenvalue ratio to iteratively update the target covariance matrix. If, based on the real-time updated and smoothed target covariance matrix, it is determined that the target object intends to cut into the lane currently occupied by the target vehicle, a warning operation is executed.

2. The method according to claim 1, characterized in that, Updating the weights of each sensor data based on the eigenvalue ratio includes: If the eigenvalue ratio is less than the first threshold, it is determined that the target object is in a shaking motion state, and the camera weight is increased while the radar weight is decreased. If the eigenvalue ratio is greater than or equal to the second threshold, it is determined that the target object is in a linear motion state, the radar weight is increased and the camera weight is decreased. If the feature ratio is greater than or equal to the first threshold and less than the second threshold, then the camera weight and the radar weight remain unchanged.

3. The method according to claim 1, characterized in that, After performing eigenvalue decomposition on the smoothed target covariance matrix to obtain the eigenvalue ratio, the method further includes: The target covariance matrix after smoothing and filtering is subjected to eigenvalue decomposition to obtain the error distribution magnitude and error extension direction. The error distribution magnitude is used to indicate the overall dispersion of the target object position estimation, and the error extension direction is used to indicate the potential change trend of the target object's motion. Determining whether the target object intends to enter the lane currently occupied by the target vehicle based on the real-time updated and smoothed target covariance matrix includes: if the error distribution amplitude is greater than a preset amplitude threshold, and the deviation between the error extension direction and the actual movement direction of the target object is greater than a preset deviation threshold, then it is determined that the target object intends to enter the lane currently occupied by the target vehicle.

4. The method according to claim 3, characterized in that, Performing early warning operations includes: Determine the relative distance, relative speed, and spatiotemporal collision probability between the target vehicle and the target object; The warning probability threshold is dynamically adjusted based on the error distribution amplitude, wherein the error distribution amplitude is negatively correlated with the warning probability threshold. If the relative distance is less than a preset distance threshold, the relative motion speed is greater than a preset speed threshold, and the spatiotemporal collision probability is greater than the warning probability threshold, then a warning operation is performed according to the determined entry intention.

5. The method according to claim 4, characterized in that, Determining the spatiotemporal collision probability between the target vehicle and the target object includes: Determine the position difference based on the current position of the target object and the current position of the target vehicle; The target covariance matrix of the target object and the covariance matrix of the target vehicle are superimposed to determine the joint covariance matrix; The Mahalanobis distance is determined based on the position difference and the joint covariance matrix. The spatiotemporal collision probability between the target vehicle and the target object is determined based on the Mahalanobis distance, wherein the Mahalanobis distance is negatively correlated with the spatiotemporal collision probability.

6. The method according to claim 3, characterized in that, After performing eigenvalue decomposition on the smoothed target covariance matrix to obtain the error distribution magnitude and error propagation direction, the method further includes: If a state frame that satisfies the condition that the feature value ratio is less than the target threshold and the error distribution amplitude is greater than the set amplitude threshold is detected to occur multiple times, then it is determined that the position fluctuation of the target object is abnormal, wherein the target threshold is less than the first threshold and the set amplitude threshold is greater than the preset amplitude threshold. The current state of the target object is corrected, and the calculation is re-iterated based on the corrected current state to obtain a new target covariance matrix.

7. The method according to claim 1, characterized in that, The multidimensional sensor data includes camera data and radar data. The target covariance matrix is ​​determined based on each sensor data and its corresponding weight, including: Determine the camera weights corresponding to the camera data and the radar weights corresponding to the radar data; According to the camera weight and the radar weight, the camera data and the radar data are weighted and summed to obtain perception fusion data; Based on the perception fusion data, the optimal state estimation is performed to update the current state of the target object, wherein the current state includes the current position and the current velocity; The current state of the target object is processed to obtain the target covariance matrix.

8. A vehicle warning device, characterized in that, The device includes: The determination module is used to acquire multi-dimensional sensing data of the target object and the weight of each sensing data in real time, and determine the target covariance matrix based on each sensing data and the corresponding weight, wherein the target covariance matrix is ​​used to indicate the distribution range of the target object's position estimate; The smoothing module is used to perform smoothing filtering on the target covariance matrix to obtain the smoothed target covariance matrix. The update module is used to perform eigenvalue decomposition on the smoothed target covariance matrix to obtain the eigenvalue ratio, and update the weight of each sensor data according to the eigenvalue ratio to iteratively update the target covariance matrix. The judgment and warning module is used to perform a warning operation if it is determined, based on the real-time updated and smoothed target covariance matrix, that the target object intends to cut into the lane currently occupied by the target vehicle.

9. An electronic device, characterized in that, It includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; Memory, used to store computer programs; A processor, when executing a program stored in memory, implements the method described in any one of claims 1-7.

10. A vehicle, characterized in that, The vehicle includes a memory and a processor, wherein the memory stores executable program code, and the processor is used to call and execute the executable program code to implement the method as described in any one of claims 1 to 7.

Images