Vehicle trajectory prediction method, electronic equipment and storage medium

Abstract

The invention provides a vehicle track prediction method, electronic equipment and a storage medium, and relates to the field of automatic driving track prediction.The method comprises the steps that state information of a target vehicle is acquired, and a kinematics track of the target vehicle is determined; road information is obtained, and the intention of a target vehicle is predicted based on the state information of the target vehicle and the road information; determining a planned trajectory of the target vehicle based on the intention of the target vehicle; and carrying out weighted fusion on the kinematic trajectory and the planned trajectory to obtain a predicted trajectory. According to the method, smooth transition from short-term accurate extrapolation to long-term reasonable planning of the predicted trajectory is realized through a staged and multi-model collaborative prediction framework, the overall precision and continuity of trajectory prediction and the adaptability to complex driving scenes are remarkably improved, and a more reliable basis is provided for automatic driving decision making.

Description

Technical Field

[0001] This invention relates to the field of autonomous driving trajectory prediction, and more particularly to a vehicle trajectory prediction method, electronic device, and storage medium. Background Technology

[0002] Accurately predicting the future trajectory of a target vehicle, especially its key driving intentions such as lane changes and steering, is crucial for the decision-making module of an autonomous driving system. Currently, most mainstream solutions adopt a two-stage framework of "intent reasoning + trajectory generation," that is, first identifying the driving intention of the target vehicle, and then generating its future driving trajectory based on that intention.

[0003] Regarding intent reasoning, existing technologies mainly employ three implementation paths. The first is rule-based methods, which determine intent by setting specific rules such as speed thresholds, acceleration changes, or turn signal status. This method is logically simple and requires no training data, but its generalization ability is limited. The second is methods based on traditional machine learning, such as Support Vector Machines (SVM), which classify vehicle state features, achieving a balance between model complexity and interpretability. The third is deep learning-based methods, which train neural networks using large-scale data. While achieving high inference accuracy, these methods are complex and computationally expensive. In trajectory generation, common methods can be summarized into four categories: First, methods based on physical models (such as uniform speed and uniform acceleration models), which are computationally simple, have high real-time performance, and are suitable for short-term prediction; second, methods based on deep learning models (such as LSTM), which take the state and intent as input and output a trajectory, achieving high accuracy; third, methods based on retrieval, which match the most similar trajectory from a historical trajectory database; and fourth, methods based on planning, which plan the trajectory in real time based on the current state and intent.

[0004] However, all of the aforementioned existing technologies have significant drawbacks in practical applications. In the intent reasoning stage, rule-based methods struggle to handle complex and ever-changing real-world traffic scenarios, resulting in a high false positive rate; while the computational overhead of deep learning methods makes them unsuitable for environments with limited onboard computing power. In the trajectory generation stage, physics-based methods, although fast, have short prediction time domains, typically not exceeding one second, failing to meet medium- to long-term prediction needs; deep learning methods are also limited by computing power; and retrieval methods rely on large and comprehensive trajectory sample libraries, leading to high construction and maintenance costs. Summary of the Invention

[0005] This invention aims to at least solve the aforementioned technical problems existing in the prior art. To this end, a first aspect of this invention proposes a vehicle trajectory prediction method, the method comprising:

[0006] Obtain the state information of the target vehicle and determine its kinematic trajectory;

[0007] Obtain road information and predict the target vehicle's intention based on the target vehicle's status information and road information;

[0008] Based on the target vehicle's intent, determine the target vehicle's planned trajectory;

[0009] The predicted trajectory is obtained by weighted fusion of the kinematic trajectory and the planned trajectory.

[0010] Optionally, predicting the target vehicle's intent includes:

[0011] Based on the target vehicle's status information and road information, the lateral distance is obtained; the target vehicle's status information includes lateral velocity; the lateral distance is the distance along the lane's lateral dimension between the target vehicle's center point and the lane centerline of the target vehicle's current lane; the lateral velocity is the velocity component of the target vehicle along the lane's lateral dimension along the lane centerline of the target vehicle's current lane, positive to the left and negative to the right;

[0012] Intent prediction is made based on whether the directions of lateral distance and lateral velocity are consistent.

[0013] Optionally, the step of predicting intent based on whether the directions of lateral distance and lateral velocity are consistent includes:

[0014] Based on the vehicle's direction of travel, if the center point of the target vehicle is located to the left of the lane center line of the lane in which the target vehicle is currently located, the lateral distance is positive; if the center point is located to the right of the lane center line, the lateral distance is negative.

[0015] The vehicle's intention is predicted based on the lateral distance and lateral velocity of the target vehicle.

[0016] Optionally, predicting the vehicle's intention based on the sign of its lateral distance and lateral velocity includes:

[0017] If both the lateral distance and lateral velocity are 0, the prediction result is no lane change;

[0018] If both the lateral distance and the lateral velocity are non-negative and at most one of them is 0, input them into the first support vector machine for prediction;

[0019] If both the lateral distance and the lateral velocity are non-positive and at most one of them is 0, input them into the second support vector machine for prediction;

[0020] Otherwise, the prediction result is no lane change.

[0021] Optionally, the positive and negative prediction of vehicle intent based on the lateral distance and lateral velocity of the target vehicle includes:

[0022] If the product of lateral distance and lateral velocity is negative, the prediction result is no lane change;

[0023] Otherwise, if the lateral distance or lateral velocity is positive, it is input into the first support vector machine for prediction; if the lateral distance or lateral velocity is negative, it is input into the second support vector machine for prediction.

[0024] Optionally, the prediction result of the first support vector machine is either a left lane change or no lane change; the prediction result of the second support vector machine is either a right lane change or no lane change.

[0025] Optionally, the weighted fusion of the kinematic trajectory and the planned trajectory includes:

[0026] The weight of the kinematic trajectory decreases over time until it reaches 0; the weight of the planned trajectory increases over time until it reaches 1; at any given time, the sum of the weights of the kinematic trajectory and the planned trajectory is 1.

[0027] A second aspect of the present invention provides an electronic device comprising a processor and a memory, wherein the memory stores at least one instruction or at least one program, the at least one instruction or at least one program being loaded and executed by the processor to implement the vehicle trajectory prediction method as proposed in the first aspect.

[0028] A third aspect of the present invention provides a computer-readable storage medium storing at least one instruction or at least one program, wherein the at least one instruction or at least one program is loaded and executed by a processor to implement the vehicle trajectory prediction method as proposed in the first aspect.

[0029] The beneficial effects of a vehicle trajectory prediction method are:

[0030] This invention first acquires the target vehicle's state information to determine its kinematic trajectory, ensuring that short-term predictions closely match the vehicle's real-time motion state and guaranteeing the accuracy of early predictions. Next, it acquires road information and combines it with the target vehicle's state information to predict the vehicle's intent, making subsequent trajectory planning more consistent with actual road scenarios and vehicle behavior logic. Then, it determines the planned trajectory based on the vehicle's intent, effectively avoiding trajectory deviation problems caused by relying solely on kinematic models in long-term predictions, thus ensuring long-term prediction stability. Finally, by weighted fusion of the kinematic trajectory and the planned trajectory, it successfully balances the accuracy of trajectory prediction in the early stages with the reliability in the later stages, overcoming the shortcomings of a single trajectory prediction model. This significantly improves the overall accuracy and adaptability of vehicle trajectory prediction in autonomous driving scenarios, providing a more reliable trajectory reference for the decision-making and control modules of autonomous driving systems.

[0031] Furthermore, by defining calculation rules for lateral distance and lateral velocity, and based on the design intent filtering rules of directional consistency or the positive or negative product of the two, targets with a negative product of lateral distance and lateral velocity are directly classified as not changing lanes. This can quickly eliminate targets that cannot change lanes in the short term, especially those that have just entered the current lane through a lane change, avoiding invalid lane change intent judgments. At the same time, the traditional three-class support vector machine model is improved into two two-class support vector machine models. Unlike the three-class model, there is no need to make meaningless distinctions between left and right lane changes. Instead, the first support vector machine is dedicated to predicting left lane changes or not changing lanes, and the second support vector machine is dedicated to predicting right lane changes or not changing lanes. The two models can also be trained separately, which greatly reduces the difficulty of model training and improves the efficiency and accuracy of vehicle intent prediction, laying a reliable foundation for the accurate determination of the planned trajectory. Attached Figure Description

[0032] Figure 1 A flowchart of a vehicle trajectory prediction method provided in an embodiment of the present invention;

[0033] Figure 2 A schematic diagram of the lateral distance provided for an embodiment of the present invention;

[0034] Figure 3 This is a flowchart of a vehicle intent prediction method provided in an embodiment of the present invention. Detailed Implementation

[0035] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0036] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of embodiments of this disclosure, unless otherwise stated, "a plurality of" means two or more. Furthermore, the use of "based on" or "according to" implies openness and inclusiveness, because processes, steps, calculations, or other actions "based on" or "according to" one or more of the stated conditions or values ​​may in practice be based on additional conditions or beyond the stated values.

[0037] This invention provides a vehicle trajectory prediction method, such as... Figure 1 As shown, the method may include the following steps:

[0038] Step 101: Obtain the status information of the target vehicle and determine the kinematic trajectory of the target vehicle.

[0039] Specifically, the state information of the target vehicle is obtained, including the center point coordinates (x, y), heading angle θ, speed v, and angular velocity w of the target vehicle.

[0040] The kinematic trajectory of the target vehicle is calculated using a model of uniform circular motion, and the calculation formula is as follows:

[0041]

[0042] θ(t)=θ0+wt

[0043] Where x0 and y0 are the initial positions of the vehicle center, and θ0 is the initial heading angle.

[0044] Specifically, when w is too small (|w|≤0.05rad / s), the model of uniform linear motion is adopted, and the formula is:

[0045] x(t) = x0 + vtcos(θ0)

[0046] y(t) = y0 + vtsin(θ0)

[0047] This yields the kinematic trajectory at time t.

[0048] Step 102: Obtain road information. Based on the target vehicle's status information and road information, predict the target vehicle's intention.

[0049] Specifically, road information includes lane line information and lane width.

[0050] In one possible implementation, predicting the target vehicle's intent includes:

[0051] Based on the target vehicle's status information and road information, the lateral distance is obtained; the target vehicle's status information includes lateral velocity; the lateral distance is the distance along the lane's lateral dimension between the target vehicle's center point and the lane centerline of the target vehicle's current lane; the lateral velocity is the velocity component of the target vehicle along the lane's lateral dimension along the lane centerline of the target vehicle's current lane, positive to the left and negative to the right;

[0052] Intent prediction is made based on whether the directions of lateral distance and lateral velocity are consistent.

[0053] In one possible implementation, intention prediction is performed based on whether the directions of the lateral distance and lateral velocity are consistent, including:

[0054] like Figure 2As shown, based on the vehicle's direction of travel, if the center point of the target vehicle is located to the left of the lane center line of the lane where the target vehicle is currently located, the lateral distance is positive; if the center point is located to the right of the lane center line, the lateral distance is negative.

[0055] The vehicle's intention is predicted based on the lateral distance and lateral velocity of the target vehicle.

[0056] In one possible implementation, predicting the vehicle's intention based on the lateral distance and lateral velocity of the target vehicle includes:

[0057] If both the lateral distance and lateral velocity are 0, the prediction result is no lane change;

[0058] If both the lateral distance and the lateral velocity are non-negative and at most one of them is 0, input them into the first support vector machine for prediction;

[0059] If both the lateral distance and the lateral velocity are non-positive and at most one of them is 0, input them into the second support vector machine for prediction;

[0060] Otherwise, the prediction result is no lane change, meaning the target vehicle will continue to travel in its current lane and will not change lanes to other lanes.

[0061] In one possible implementation, such as Figure 3 As shown, based on the lateral distance and lateral velocity of the target vehicle, the predicted vehicle intention includes:

[0062] If the product of lateral distance and lateral velocity is negative, the prediction result is no lane change;

[0063] Otherwise, if the lateral distance or lateral velocity is positive, it is input into the first support vector machine for prediction; if the lateral distance or lateral velocity is negative, it is input into the second support vector machine for prediction.

[0064] In one possible implementation, the prediction result of the first support vector machine is either a left lane change or no lane change; the prediction result of the second support vector machine is either a right lane change or no lane change.

[0065] Specifically, the support vector machine in this embodiment of the invention is a binary classification support vector machine. Both the first support vector machine and the second support vector machine are binary classification models. The training data includes the lateral distance, lateral speed and real lane change labels of historical vehicles. The kernel function is a linear kernel, and the training evaluation index is the accuracy of lane change intention prediction.

[0066] Step 103: Determine the planned trajectory of the target vehicle based on its intent.

[0067] Specifically, the planned trajectory is implemented in the Frenet coordinate system, which converts the obtained Cartesian coordinates (x,y) and velocity into Frenet coordinates (s,d) based on the centerline of the target lane. If the target vehicle intends not to change lanes, the centerline of the target lane is the centerline of the current lane; otherwise, the centerline of the target lane is the centerline of the lane to which the vehicle is changing lanes.

[0068] The starting point of the planned trajectory is the center point (s0, d0) of the target vehicle, and the ending point is set at a point on the center line of the target lane, that is, the ending point (s1, d1) of the planned trajectory has d1 as 0. Assuming that the speed along the positive d direction is also 0, the target vehicle representing the ending point will travel along the center line of the lane and will no longer have left or right speed.

[0069] The coordinate s1 of the endpoint is determined by the velocity v of the target vehicle in the positive s direction from the starting point. s Confirm, set as

[0070] s1 = s0 + v s t

[0071] The starting speed is v, and the ending speed along the positive s direction is set to v, meaning the vehicle completely reverses its speed direction to the positive s direction through an angle change.

[0072] The planned trajectory of the target vehicle is calculated. The s-coordinate or d-coordinate of the trajectory at time t is determined by the formula k(t), where k(t) is a cubic equation with t as the independent variable:

[0073] k(t) = a0 + a1t + a2t 2 +a3t 3

[0074] In the formula, α0, α1, α2, and α3 are formula parameters, a0 = k0, and a1 = v. k0 α2 and α3 are obtained by solving the following equation:

[0075]

[0076] In the formula, k0 is the s-coordinate or d-coordinate of the starting point, and k1 is the s-coordinate or d-coordinate of the ending point; v k0 The velocity component in the s-direction or d-direction from the starting point, v k1 The component rate in the s-direction or d-direction with the endpoint.

[0077] Step 104: Weighted fusion of kinematic trajectory and planned trajectory to obtain predicted trajectory.

[0078] In one possible implementation, the weighted fusion of the kinematic trajectory and the planned trajectory includes:

[0079] The weight of the kinematic trajectory decreases over time until it reaches 0; the weight of the planned trajectory increases over time until it reaches 1; at any given time, the sum of the weights of the kinematic trajectory and the planned trajectory is always 1.

[0080] Specifically, taking the predicted trajectory of T=3s as an example, a specific method of weighted fusion is as follows:

[0081] The predicted trajectory is divided into three trajectories according to time sequence: the first trajectory, the second trajectory, and the third trajectory. The first trajectory is the kinematic trajectory at second 1, the third trajectory is the planned trajectory at second 3, and the second trajectory is determined by weighting the kinematic trajectory at second 2 with the planned trajectory. The specific weighting method is as follows:

[0082] Obtain the kinematic trajectory and planned trajectory at the 2nd second, define the range of values ​​for t, and discretize it at certain intervals to obtain the discrete time t. i Determine t i kinematic trajectory coordinates at time (x) 2i ,y 2i ,θ 2i ) and planned trajectory coordinates (s 2i ,d 2i ), the coordinates (s) of the planned trajectory 2i ,d 2i Convert to Cartesian coordinates (x') 2i ,y' 2i ,θ' 2i The coordinates (x, y) of the predicted trajectory at time ti are calculated using a weighted average. 2i ,y" 2i ,θ" 2i The calculation formula is as follows:

[0083] m"=γm+(1-γ)m'

[0084] γ=1-(t i -t 20 ) / (t 21 -t 20 )

[0085] In the formula, m represents the coordinates of the kinematic trajectory, m' represents the Cartesian coordinates of the planned trajectory, m" represents the coordinates of the predicted trajectory, γ represents the weighting coefficient of the kinematic trajectory, and its value ranges from 0 ≤ γ ≤ 1. 20 t is the start time of the second trajectory. 21 This is the end time of the second trajectory.

[0086] Specifically, another weighted fusion method is as follows:

[0087] Define the total prediction time of the predicted trajectory as T, and discretize T at certain intervals to obtain discrete time t. i Determine t ikinematic trajectory coordinates at time (x) 2i ,y 2i ,θ 2i ) and planned trajectory coordinates (s 2i ,d 2i ), the coordinates (s) of the planned trajectory 2i ,d 2i Convert to Cartesian coordinates (x') 2i ,y' 2i ,θ' 2i The coordinates (x, y) of the predicted trajectory at time ti are calculated using a weighted average. 2i ,y" 2i ,θ" 2i The calculation formula is as follows:

[0088] m"=γm+(1-γ)m'

[0089]

[0090] In the formula, m is the coordinate of the kinematic trajectory, m' is the Cartesian coordinate of the planned trajectory, m" is the coordinate of the predicted trajectory, and γ is the weight coefficient of the kinematic trajectory.

[0091] In summary, in this embodiment of the invention, the vehicle trajectory prediction method significantly improves the accuracy and practicality of trajectory prediction through a phased, multi-model collaborative prediction framework. The method first acquires the target vehicle's state information and determines its short-term kinematic trajectory based on a kinematic model, ensuring a high degree of consistency between the initial prediction and the vehicle's real-time motion state, thus guaranteeing the initial accuracy of the trajectory prediction. Subsequently, by fusing road structure information and vehicle state, the driving intention of the target vehicle is reliably inferred, enabling subsequent trajectory generation to strictly adhere to actual road constraints and behavioral logic. Then, based on the inferred vehicle intention, a path planning method is used to generate a long-term planned trajectory that conforms to the scene's semantics, effectively avoiding the problems of trajectory deviating from the road network or exhibiting unreasonable behavior caused by simply relying on kinematic extrapolation. Finally, by fusing the kinematic trajectory and the planned trajectory with time-varying weights, a complete and smooth future trajectory is generated. This fusion strategy assigns a higher weight to the kinematic trajectory in the early stages of prediction to maintain responsiveness to the vehicle's real-time dynamics; in the later stages of prediction, it gradually transitions to prioritizing the planned trajectory, thereby ensuring the long-term rationality and stability of the trajectory. This method overcomes the inherent defects of single prediction models (such as kinematic models which are prone to distortion in long-term predictions, and planning models which are prone to lag in short-term predictions), and achieves a balance between short-term accuracy and long-term rationality, providing a more reliable environmental perception input for the decision-making and planning module of autonomous driving systems.

[0092] Furthermore, this invention specifically optimizes the vehicle intent prediction process. It clearly defines key decision features—lateral distance and lateral velocity—and designs a rapid filtering rule based on the consistency of their directions. Specifically, when the product of lateral distance and lateral velocity is negative, the target vehicle can be directly determined to have a "no lane change" intent. This rule efficiently filters out targets that do not have the conditions to change lanes in the short term (e.g., vehicles that have just completed a lane change and are converging within the lane), reducing unnecessary complex calculations and improving system real-time performance. Simultaneously, the traditional three-class (left lane change, no lane change, right lane change) support vector machine model is decomposed into two independent binary classification support vector machine models: the first model specifically distinguishes between "left lane change" and "non-left lane change," and the second model specifically distinguishes between "right lane change" and "non-right lane change." This structure avoids the direct competition and confusion between "left lane change" and "right lane change" in the feature space in the three-class classification model, and allows for independent and more targeted training of the two binary classifiers, which greatly reduces the complexity and difficulty of model training, thereby significantly improving the accuracy and reliability of intent classification and laying a solid foundation for generating reasonable planned trajectories in the future.

[0093] Furthermore, this invention features a refined trajectory fusion strategy. The predicted trajectory is obtained by weighted fusion of the kinematic trajectory and the planned trajectory, initially favoring the kinematic trajectory and later favoring the planned trajectory. Within the predicted trajectory, the weight of the kinematic trajectory decreases linearly or non-linearly as the prediction time progresses. This time-varying weighting fusion mechanism ensures a smooth transition between the predicted trajectory and the vehicle's current state at the starting point. Simultaneously, by gradually enhancing the guiding role of the planned trajectory, the final trajectory smoothly and progressively converges to a reasonable path that conforms to the road structure and driving intentions. This design effectively prevents abrupt changes or discontinuities at connection points, guaranteeing the continuity of the output trajectory and driving comfort, further enhancing the overall reliability and usability of the trajectory prediction results.

[0094] In another embodiment of the present invention, an electronic device is also provided, the electronic device including a processor and a memory, the memory storing at least one instruction or at least one program, the at least one instruction or at least one program being loaded and executed by the processor to implement the vehicle trajectory prediction method proposed in the embodiment of the present invention.

[0095] In another embodiment of the present invention, a computer-readable storage medium is also provided, wherein at least one instruction or at least one program is stored in the storage medium, and the at least one instruction or at least one program is loaded and executed by a processor to implement the vehicle trajectory prediction method proposed in the embodiment of the present invention.

[0096] The foregoing primarily describes the solutions provided by the embodiments of the present invention from the perspective of the device. It is understood that, in order to achieve the above functions, the device includes corresponding hardware structures and / or software modules for executing each function. Those skilled in the art should readily recognize that, in conjunction with the algorithmic steps of the various examples described in the embodiments disclosed herein, the present invention can be implemented in hardware or a combination of hardware and computer software. Whether a function is executed in hardware or by computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present invention.

[0097] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A vehicle trajectory prediction method, characterized by, The method comprises: obtaining state information of a target vehicle, and determining a kinematic trajectory of the target vehicle; obtaining road information, and predicting an intention of the target vehicle based on the state information of the target vehicle and the road information; determining a planned trajectory of the target vehicle based on the intention of the target vehicle; weighting and fusing the kinematic trajectory and the planned trajectory to obtain a predicted trajectory.

2. The vehicle trajectory prediction method of claim 1, wherein, The prediction of the intention of the target vehicle comprises: obtaining a lateral distance according to the state information of the target vehicle and the road information; the state information of the target vehicle comprises a lateral velocity; the lateral distance is a distance of a center point of the target vehicle from a lane center line of a lane in which the target vehicle is currently located in a lane lateral dimension; the lateral velocity is a velocity component of a velocity of the target vehicle on the lane center line of the lane in which the target vehicle is currently located in the lane lateral dimension, and is positive to the left and negative to the right; performing intention prediction according to whether the direction of the lateral distance and the direction of the lateral velocity are consistent.

3. The vehicle trajectory prediction method of claim 2, wherein, The intention prediction according to whether the direction of the lateral distance and the direction of the lateral velocity are consistent comprises: based on a vehicle driving direction, if the center point of the target vehicle is located on the left side of the lane center line of the lane in which the target vehicle is currently located, the lateral distance is positive; if the center point is located on the right side of the lane center line, the lateral distance is negative; predicting a vehicle intention based on the positive and negative of the lateral distance and the lateral velocity of the target vehicle.

4. The vehicle trajectory prediction method of claim 3, wherein, The prediction of the vehicle intention based on the positive and negative of the lateral distance and the lateral velocity of the target vehicle comprises: if the lateral distance and the lateral velocity are both 0, the prediction result is no lane change; if the lateral distance and the lateral velocity are both non-negative and at most one of them is 0, input a first support vector machine for prediction; if the lateral distance and the lateral velocity are both non-positive and at most one of them is 0, input a second support vector machine for prediction; otherwise, the prediction result is no lane change.

5. The vehicle trajectory prediction method of claim 3, wherein, The prediction of the vehicle intention based on the positive and negative of the lateral distance and the lateral velocity of the target vehicle comprises: if the product of the lateral distance and the lateral velocity is negative, the prediction result is no lane change; otherwise, if the lateral distance or the lateral velocity is positive, input the first support vector machine for prediction, and if the lateral distance or the lateral velocity is negative, input the second support vector machine for prediction.

6. The vehicle trajectory prediction method according to claim 4 or 5, characterized in that, The prediction result of the first support vector machine is left lane change or no lane change; and the prediction result of the second support vector machine is right lane change or no lane change.

7. The vehicle trajectory prediction method of claim 1, wherein, The weighting and fusing of the kinematic trajectory and the planned trajectory comprises: the weight of the kinematic trajectory decreases over time until it decreases to 0; the weight of the planned trajectory increases over time until it increases to 1; and the sum of the weight of the kinematic trajectory and the weight of the planned trajectory at any time is 1.

8. An electronic device, comprising: The electronic device comprises a processor and a memory, and the memory stores at least one instruction or at least one program, which is loaded and executed by the processor to implement the vehicle trajectory prediction method according to any one of claims 1-7.

9. A computer-readable storage medium, characterized in that, The storage medium stores at least one instruction or at least one program, which is loaded and executed by the processor to implement the vehicle trajectory prediction method according to any one of claims 1-7.

Images