Vehicle anti-collision method and device, vehicle, storage medium and program product

Abstract

The invention provides a vehicle anti-collision method and device, a vehicle, a storage medium and a computer program product, and the method comprises the steps: generating a first prediction track according to the operation state information, the current speed, the vehicle wheelbase and the steering angle of the vehicle; according to the current speed and the corresponding course angle variable quantity, a left side maximum correction angle and a right side maximum correction angle are obtained through the prediction model, and a first steering correction angle and a second steering correction angle are determined; and performing anti-collision processing according to the first prediction trajectory, the second prediction trajectory, the third prediction trajectory, the vehicle speed and the boundary dimension information of the vehicle and other vehicles. According to the invention, the precision and stability of trajectory prediction can be improved, and the limitation of single trajectory prediction is overcome; the generalization ability of the model can be enhanced, and the dependence on high-quality annotation data is reduced; high-risk collision scenes can be accurately recognized, effective avoidance measures are taken, and the false alarm rate caused by prediction errors is reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of vehicles, in particular to a vehicle anti-collision method and device, a vehicle, a storage medium and a computer program product. BACKGROUND

[0002] Vehicle trajectory prediction plays an important role in vehicle anti-collision technology. At present, vehicle trajectory prediction methods usually adopt trajectory prediction methods based on dynamic modeling, trajectory prediction methods based on neural networks, etc. The trajectory prediction method based on dynamic modeling mainly relies on the running state of the vehicle (such as speed, acceleration, heading angle, etc.) for prediction, but the behavior decision of the driver has high randomness, and the motion trajectory of the vehicle is not only subject to its own dynamics, but also subject to the dynamic influence of the surrounding environment and traffic participants, so the applicability of this method in highly dynamic traffic environment is limited; the trajectory prediction method based on neural network usually outputs a most possible future trajectory, which cannot fully consider the possible multiple driving intentions of the vehicle and lacks the ability to measure the uncertainty of the trajectory. The existing vehicle trajectory prediction methods have deficiencies in prediction accuracy, behavior uncertainty modeling, data dependency, etc., and using the predicted trajectory for anti-collision processing will lead to more misjudgments or omissions, thereby affecting the driving safety of the vehicle. SUMMARY

[0003] The present disclosure provides a vehicle anti-collision method, device, vehicle, storage medium and computer program product.

[0004] According to a first aspect of the present disclosure, a vehicle anti-collision method is provided, comprising: generating a first predicted trajectory according to the running state information, the current speed, the vehicle wheelbase and the steering angle of the vehicle; obtaining a heading angle change corresponding to the current speed according to the first predicted trajectory; obtaining a left maximum correction angle and a right maximum correction angle corresponding to the steering angle by using a trained prediction model according to the current speed and the corresponding heading angle change; determining a first steering correction angle and a second steering correction angle according to the steering angle and the left maximum correction angle and the right maximum correction angle; generating a second predicted trajectory according to the running state information, the current speed, the vehicle wheelbase and the first steering correction angle, and generating a third predicted trajectory according to the running state information, the current speed, the vehicle wheelbase and the second steering correction angle; and performing anti-collision processing according to the first predicted trajectory, the second predicted trajectory, the third predicted trajectory, the vehicle speed and the external size information of the vehicle and the first predicted trajectory, the second predicted trajectory, the third predicted trajectory, the vehicle speed and the external size information of the other vehicle.

[0005] Optionally, the prediction model comprises a semi-supervised learning model, and the method comprises: generating labeled trajectory sample data and unlabeled trajectory sample data by using an autonomous driving simulation tool; training the semi-supervised learning model by using the labeled trajectory sample data and the unlabeled trajectory sample data, to obtain the trained semi-supervised learning model.

[0006] Optionally, the labeled trajectory sample data comprises trajectory data and corresponding labels, and the unlabeled trajectory sample data comprises the trajectory data; the trajectory data comprises a simulated speed of a vehicle and a corresponding change in a heading angle; and the labels comprise a left maximum correction angle and a right maximum correction angle corresponding to a simulated steering angle of the vehicle.

[0007] Optionally, a first loss function corresponding to the labeled trajectory sample data is constructed; a second loss function corresponding to the unlabeled trajectory sample data after being disturbed by Gaussian noise is constructed; a composite loss function is generated according to the first loss function and the second loss function; and the semi-supervised learning model is adjusted by using the composite loss function, to obtain the trained semi-supervised learning model.

[0008] Optionally, generating the composite loss function according to the first loss function and the second loss function comprises: weighting the second loss function by using a weight parameter corresponding to the second loss function; and generating the composite loss function according to the first loss function and the weighted second loss function.

[0009] Optionally, adjusting the semi-supervised learning model by using the composite loss function, to obtain the trained semi-supervised learning model, comprises: determining a prediction error value by using the composite loss function; adjusting parameters of the semi-supervised learning model according to the prediction error value, until the prediction error value is lower than a preset error threshold, to obtain the trained semi-supervised learning model.

[0010] Optionally, the running state information comprises a vehicle position and a heading angle; and generating a first predicted trajectory according to the running state information, a current speed, a wheelbase of the vehicle, and a steering angle of the vehicle comprises: determining a first position and a first heading angle of the vehicle at a future time by using a vehicle trajectory algorithm, according to the running state information, the current speed, the wheelbase of the vehicle, and the steering angle of the vehicle; and generating the first predicted trajectory according to the running state information, the first position, and the first heading angle.

[0011] Optionally, the generating the second predicted trajectory according to the running state information, the current speed, the wheelbase of the vehicle and the first steering correction angle comprises: determining, by using the vehicle trajectory algorithm, a second position and a second heading angle of the vehicle at a future time according to the running state information, the current speed, the wheelbase of the vehicle and the first steering correction angle; and generating the second predicted trajectory according to the running state information, the second position and the second heading angle.

[0012] Optionally, the generating the third predicted trajectory according to the running state information, the current speed, the wheelbase of the vehicle and the second steering correction angle comprises: determining, by using the vehicle trajectory algorithm, a third position and a third heading angle of the vehicle at a future time according to the running state information, the current speed, the wheelbase of the vehicle and the second steering correction angle; and generating the third predicted trajectory according to the running state information, the third position and the third heading angle.

[0013] Optionally, the performing the anti-collision processing according to the first predicted trajectory, the second predicted trajectory, the third predicted trajectory, the speed and the size information of the vehicle and the first predicted trajectory, the second predicted trajectory, the third predicted trajectory, the speed and the size information of the other vehicle comprises: determining whether there is a collision point according to the first predicted trajectory, the second predicted trajectory, the third predicted trajectory and the size information of the vehicle and the first predicted trajectory, the second predicted trajectory, the third predicted trajectory and the size information of the other vehicle; when it is determined that there is a collision point, determining a time of collision according to the speed of the vehicle and the speed of the other vehicle; and triggering a warning and performing a deceleration processing when a time interval between the time of collision and a current time is less than a safety time threshold.

[0014] Optionally, the information transmitted by the other vehicle is received by using a vehicle-to-vehicle (V2V) communication device, and the information comprises the first predicted trajectory, the second predicted trajectory, the third predicted trajectory, the speed and the size information of the other vehicle.

[0015] According to a second aspect of the present disclosure, there is provided a vehicle anti-collision device, comprising: a memory; and a processor coupled to the memory, the processor being configured to execute a method as described above based on instructions stored in the memory.

[0016] According to a third aspect of the present disclosure, there is provided a vehicle comprising a vehicle anti-collision device as described above.

[0017] According to a fourth aspect of the present disclosure, there is provided a computer-readable storage medium storing computer instructions, the instructions being executed by a processor to implement steps of a method as described above.

[0018] According to a fifth aspect of this disclosure, a computer program product is provided, comprising a computer program that, when executed by a processor, implements the steps of the method described above.

[0019] The vehicle collision avoidance method, device, vehicle, storage medium, and computer program product disclosed herein can improve the accuracy and stability of trajectory prediction, more accurately determine changes in driver intent in complex traffic scenarios, and overcome the limitations of single trajectory prediction; it can enhance the generalization ability of the model and reduce the dependence on high-quality labeled data; it can effectively measure the uncertainty of future behavior, has faster inference speed and lower computational complexity, improve vehicle driving safety, accurately identify high-risk collision scenarios and take effective avoidance measures, reduce the false alarm rate caused by prediction errors, and improve the user experience. Attached Figure Description

[0020] The above and other objects, features, and advantages of this disclosure will become more apparent from the more detailed description of the embodiments thereof in conjunction with the accompanying drawings. The accompanying drawings are provided to further illustrate the embodiments of this disclosure and form part of the specification. They are used together with the embodiments of this disclosure to explain the disclosure and do not constitute a limitation thereof. The above and other objects and advantages of this disclosure will be further described below with reference to specific embodiments and the accompanying drawings. In the drawings, the same or corresponding technical features or components will be represented by the same or corresponding reference numerals.

[0021] Figure 1 This is a schematic flowchart of some embodiments of the vehicle collision avoidance method according to the present disclosure;

[0022] Figure 2 This is a schematic diagram of the process of training a semi-supervised learning model in some embodiments of the vehicle collision avoidance method according to the present disclosure;

[0023] Figure 3 An angle statistics diagram for some embodiments of the vehicle collision avoidance method according to this disclosure;

[0024] Figure 4 This is a flowchart illustrating the process of training a model using a loss function in some embodiments of the vehicle collision avoidance method according to this disclosure.

[0025] Figure 5 This is a schematic diagram illustrating the structure and training of a semi-supervised learning neural network model in some embodiments of the vehicle collision avoidance method according to this disclosure;

[0026] Figure 6 This is a schematic flowchart illustrating the generation of a first predicted trajectory in some embodiments of the vehicle collision avoidance method according to the present disclosure;

[0027] Figure 7 This is a schematic diagram of a trajectory iteration algorithm based on vehicle steering angle in some embodiments of the vehicle collision avoidance method according to the present disclosure;

[0028] Figure 8 This is a schematic diagram of a vehicle behavior uncertainty measurement in some embodiments of the vehicle collision avoidance method according to the present disclosure;

[0029] Figure 9 This is a schematic diagram of the process of performing collision avoidance processing in some embodiments of the vehicle collision avoidance method according to the present disclosure;

[0030] Figure 10 and Figure 11 This is a schematic diagram of an optimized TTC collision avoidance algorithm in some embodiments of the vehicle collision avoidance method according to the present disclosure;

[0031] Figure 12 This is a schematic diagram of collision avoidance detection in some embodiments of the vehicle collision avoidance method according to the present disclosure;

[0032] Figure 13 This is a schematic diagram of some embodiments of a vehicle collision avoidance device according to the present disclosure. Detailed Implementation

[0033] Exemplary embodiments of the present disclosure will be described below with reference to the accompanying drawings. For clarity and brevity, not all features of the embodiments are described in the specification. However, it should be understood that many implementation-specific settings must be made in carrying out the embodiments to achieve the developer's specific goals, such as complying with constraints related to the device and business, and these constraints may vary depending on the implementation. Furthermore, it should be understood that while development work can be very complex and time-consuming, such development work is merely a routine task for those skilled in the art who benefit from the present disclosure.

[0034] It should be noted that, unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps set forth in these embodiments do not limit the scope of this disclosure.

[0035] Those skilled in the art will understand that the terms "first," "second," etc., in the embodiments of this disclosure are only used to distinguish different steps, devices, or modules, and do not represent any specific technical meaning, nor do they indicate a necessary logical order between them.

[0036] It should also be understood that in the embodiments disclosed herein, "a plurality of" may refer to two or more, and "at least one" may refer to one, two or more.

[0037] It should also be understood that, for any component, data, or structure mentioned in the embodiments of the present disclosure, one or more can be generally understood without explicit limitation or in the context of the opposite implications given by the preceding or subsequent text.

[0038] In addition, the term "and / or" in the present disclosure is merely a description of the association relationship of the associated objects, which means that there can be three relationships, for example, A and / or B can represent the three cases of A existing alone, A and B existing together, and B existing alone. In addition, the character " / " in the present disclosure generally represents an "or" relationship between the front and rear associated objects.

[0039] It should also be understood that the description of the embodiments of the present disclosure emphasizes the differences between the embodiments, and the same or similar parts can be referred to each other, and for the sake of brevity, they will not be repeated one by one.

[0040] At the same time, it should be understood that, for the convenience of description, the size of each part shown in the drawings is not drawn in accordance with the actual proportional relationship.

[0041] The following description of at least one example embodiment is merely illustrative in nature and is in no way limiting on the disclosure and its applications or uses.

[0042] Techniques, methods, and devices known to those of ordinary skill in the relevant art can not be discussed in detail, but where appropriate, the described techniques, methods, and devices should be considered as part of the specification.

[0043] It should be noted that similar reference numbers and letters refer to similar items in the following drawings, and therefore, once an item is defined in one drawing, it need not be discussed further in subsequent drawings.

[0044] In addition, in order to avoid obscuring the present disclosure due to unnecessary details, only the processing steps and / or device structures closely related to the scheme according to the present disclosure are shown in the drawings, and other details not closely related to the present disclosure are omitted. It should be noted that similar reference numbers and letters in the drawings indicate similar items, and therefore, once an item is defined in one drawing, it need not be discussed further in subsequent drawings.

[0045] In the related art known to the inventors, the rapid development of automatic driving technology has promoted the evolution of intelligent transportation systems, and vehicle trajectory prediction has become an important technology to ensure driving safety and optimize traffic flow control. Accurate trajectory prediction not only improves the safety and stability of the automatic driving system, but also plays a key role in collision avoidance; existing collision avoidance methods first perform trajectory prediction, and then use the predicted trajectory for vehicle collision avoidance processing.

[0046] Currently, vehicle trajectory prediction methods usually adopt trajectory prediction methods based on dynamic modeling, trajectory prediction methods based on neural networks, etc. The trajectory prediction method based on dynamic modeling is mainly based on dynamic modeling, including kinematic model and Kalman filter method based on Bayesian filtering, which usually uses kinematic parameters such as speed, acceleration, and heading angle of the vehicle, and deduces the future trajectory based on physical laws, and predicts the motion trend. The trajectory prediction method based on dynamic modeling usually assumes that the vehicle motion conforms to a deterministic model. However, in a complex interactive scene (such as an intersection, a roundabout, a multi-lane lane change, etc.), the behavior decision of the driver has high randomness, and the motion trajectory of the vehicle is not only subject to its own dynamics, but also subject to the dynamic influence of the surrounding environment and traffic participants; therefore, this method is difficult to accurately depict the behavior change of the driver (such as left turn, right turn, lane change, avoidance, etc.), and its applicability in a highly dynamic traffic environment is limited, and it is difficult to model the uncertainty of driving behavior, which affects the reliability and generalization ability.

[0047] The trajectory prediction method based on neural networks can extract complex spatio-temporal features by learning large-scale historical trajectory data, thereby predicting the future trajectory; for example, using RNN (Recurrent Neural Network, recurrent neural network), LSTM (Long Short-Term Memory, long short-term memory network) and other networks can capture the time dependence of trajectory data, thereby predicting the vehicle trajectory. The trajectory prediction method based on neural networks usually focuses on single trajectory prediction, that is, outputting a most likely future trajectory, without fully considering the possible multiple driving intentions of the vehicle. In the actual driving process, the future trajectory of the vehicle often has multiple possible evolution paths (such as continuing straight, turning left, or changing lanes, etc.), and a single prediction result may not be able to fully depict these potential changes. In addition, the trajectory prediction method based on neural networks usually uses an end-to-end neural network learning mechanism, which lacks explicit uncertainty measurement capability, resulting in insufficient stability and interpretability of the prediction when dealing with complex scenarios.

[0048] The above trajectory prediction methods have deficiencies in prediction accuracy, behavior uncertainty modeling, computational efficiency, and data dependency. The existing anti-collision processing method relies on predicted trajectories for collision risk assessment. Due to the uncertainty of trajectory prediction, directly using the predicted trajectory for anti-collision processing may lead to misjudgment or omission, thereby affecting the driving safety and decision stability of the vehicle.

[0049] Figure 1 For the flowchart of some embodiments of the vehicle anti-collision method according to the present disclosure, as shown in FIG. 1, the vehicle anti-collision method includes steps S101-S106. Figure 1

[0050] ​In step S101, a first predicted trajectory is generated according to the running state information of the vehicle, the current speed, the wheelbase of the vehicle, and the steering angle.

[0051] The vehicle can be various vehicles, for example, a vehicle configured with an automatic driving system, etc. The running state information of the vehicle includes the vehicle position and the heading angle, etc. The vehicle position can be the x-coordinate and y-coordinate in various coordinate systems.

[0052] In step S102, a heading angle change corresponding to the current speed is obtained according to the first predicted trajectory.

[0053] In step S103, the left and right maximum correction angles corresponding to the steering angle are obtained by using the trained prediction model according to the current speed and the corresponding heading angle change.

[0054] The prediction model can be various neural network models. For example, the prediction model can be a semi-supervised learning model, etc. The semi-supervised learning is characterized by using a small amount of labeled data and a large amount of unlabeled data for model training, which is between supervised learning (only using labeled data) and unsupervised learning (only using unlabeled data). The semi-supervised learning model can be various neural network models trained by semi-supervised learning. The left maximum correction angle is the maximum correction angle of the steering angle in the left turning direction, and the right maximum correction angle is the maximum correction angle of the steering angle in the right turning direction.

[0055] In step S104, the first and second steering correction angles are determined according to the steering angle and the left and right maximum correction angles.

[0056] In step S105, a second predicted trajectory is generated according to the running state information, the current speed, the wheelbase of the vehicle, and the first steering correction angle, and a third predicted trajectory is generated according to the running state information, the current speed, the wheelbase of the vehicle, and the second steering correction angle.

[0057] In step S106, collision avoidance processing is performed according to the first, second, and third predicted trajectories of the vehicle, the vehicle speed, and the size information of the vehicle and the first, second, and third predicted trajectories of other vehicles, the vehicle speed, and the size information of the other vehicles. The size information can include the length, width, and height of the vehicle.

[0058] The vehicle collision avoidance method of the present disclosure can improve the accuracy and stability of trajectory prediction, more accurately determine driver intention changes in complex traffic scenarios, overcome the limitations of single trajectory prediction, enhance the generalization ability of the model, and reduce the dependence on high-quality labeled data, effectively measure the uncertainty of future behavior, have faster reasoning speed and lower computational complexity, improve the safety, real-time performance, and robustness of the autonomous driving system, enable the autonomous driving system to accurately identify high-risk collision scenarios and take effective avoidance measures, and reduce the false alarm rate caused by prediction errors.

[0059] Figure 2 For the flowchart of training the semi-supervised learning model in some embodiments of the vehicle collision avoidance method according to the present disclosure, as shown in Figure 2

[0060] Step S201, using an autonomous driving simulation tool to generate labeled trajectory sample data and unlabeled trajectory sample data.

[0061] The labeled trajectory sample data and the unlabeled trajectory sample data can be various data. For example, the labeled trajectory sample data includes trajectory data and corresponding labels, and the unlabeled trajectory sample data includes trajectory data; the trajectory data includes vehicle simulation speed and corresponding heading angle change, etc.; the labels include left and right maximum correction angles corresponding to the vehicle simulation steering angle.

[0062] Step S202, using the labeled trajectory sample data and the unlabeled trajectory sample data to train the semi-supervised learning model to obtain a trained semi-supervised learning model.

[0063] The autonomous driving simulation tool can be various tools, such as CARLA simulation tools, etc. CARLA simulation tools are designed for research and development of autonomous driving technology, providing a highly realistic virtual environment for testing and verifying autonomous driving algorithms. CARLA simulation tools support simulation of various sensors (such as cameras, lidar, etc.), and allow users to customize weather, traffic conditions, and scene layout, etc., so as to be able to simulate various complex driving scenarios.

[0064] In some embodiments, the CARLA simulation tool is used to construct a trajectory dataset, including labeled trajectory sample data and unlabeled trajectory sample data, to support training and optimization of the semi-supervised learning neural network model; the operation of constructing the trajectory dataset includes steering angle calculation, behavior uncertainty measurement statistics, and data labeling, etc.

[0065] ​The calculation of the vehicle steering angle is derived based on the curvature of the trajectory points, and the curvature is determined by the coordinates of the three consecutive vehicle trajectory points. According to the geometric principle of determining a circle by three points, the radius of the circumscribed circle is calculated, and its reciprocal is taken as the curvature value. The calculation formula of the curvature is as follows:

[0066]

[0067] where (x0, y0), (x1, y1), (x2, y2) are the coordinates of the three consecutive vehicle trajectory points, and κ is the curvature value.

[0068] The steering angle calculation formula derived in combination with the vehicle geometric parameters is as follows:

[0069] θ = arctan (κ·L) (1-2);

[0070] where L is the wheelbase of the vehicle, and the default value is 2.5 meters.

[0071] Behavioral uncertainty refers to the possibility of the vehicle turning left and right during normal driving. It can be represented by the maximum possible steering angles to the left and right, i.e., the left and right maximum correction angles corresponding to the vehicle simulation steering angle. The maximum left and right steering angles (the left and right maximum correction angles corresponding to the vehicle simulation steering angle) under specific conditions can be obtained by collecting data.

[0072] For example, during data collection, by analyzing a large number of vehicle turning trajectories, the maximum steering angle of the vehicle is calculated based on the above formulas (1-1) and (1-2) to be about 30 degrees. To further measure the uncertainty of the vehicle behavior, the maximum difference between the actual steering angle and the maximum steering angle is calculated, and the maximum correction angle obtained is about 11 degrees, i.e., the left and right maximum correction angles corresponding to the vehicle simulation steering angle are about 11 degrees.

[0073] For different vehicle motion states with different speed and heading angle change, a typical vehicle motion state can be selected, and the maximum correction angle of the corresponding behavioral uncertainty measure can be calculated in the CARLA simulation environment. For example, when driving on a straight road, if the vehicle speed is 38 km / h and the heading angle change is close to 0 degrees, the maximum correction angle is about 0.4 degrees. When turning, if the vehicle speed is 29 km / h and the heading angle change is about 3 degrees, the maximum correction angles on the left and right sides are 11 degrees and 1 degree, respectively.

[0074] Figure 3 The angle statistical chart for the behavioral uncertainty measure is shown in FIG. 1. Figure 3The statistical method of the maximum correction angle of the left and right sides in the case of a speed of 29 km / h and a change in the heading angle of about 3 degrees during the right turn of the vehicle is described. By screening the data that meets the driving characteristics, the original steering angle is calculated, and the actual steering angle is derived based on the curvature, and then the correction angle between the two is calculated. Statistical analysis of these correction angles is performed to determine the distribution of the maximum correction angle in this working condition. Figure 3 The positive value in the formula represents the right correction, and the negative value represents the left correction. After systematic analysis, 11 degrees and 1 degree are finally determined as the standard label values of the maximum correction angle of the left and right sides, i.e., the maximum correction angle of the left side corresponding to the simulated steering angle of the vehicle is 11 degrees, and the maximum correction angle of the right side is 1 degree, so as to facilitate subsequent model training and application.

[0075] For example, in the CARLA simulation environment, vehicle driving state data is collected, speed and heading angle change are extracted, and data labeling is performed, which can obtain three thousand data samples, of which one thousand are labeled trajectory sample data, and two thousand are unlabeled trajectory sample data. Together, they form a data set for training and optimizing the semi-supervised learning model.

[0076] Figure 4 For the process of training the model using the loss function in some embodiments of the vehicle anti-collision method according to the present disclosure, as shown in Figure 4 , the process includes the following steps:

[0077] Step S401, a first loss function corresponding to the labeled trajectory sample data is constructed.

[0078] Step S402, a second loss function corresponding to the unlabeled trajectory sample data after Gaussian noise disturbance is constructed.

[0079] Step S403, a composite loss function is generated according to the first loss function and the second loss function.

[0080] The composite loss function can be generated in various ways. For example, the second loss function is weighted using a weight parameter corresponding to the second loss function; and the composite loss function is generated according to the first loss function and the weighted second loss function.

[0081] Step S404, the semi-supervised learning model is adjusted using the composite loss function to obtain a trained semi-supervised learning model.

[0082] The semi-supervised learning model can be adjusted using the composite loss function in various ways. For example, the prediction error value is determined using the composite loss function, the parameters of the semi-supervised learning model are adjusted according to the prediction error value, and the trained semi-supervised learning model is obtained until the prediction error value is lower than the preset error threshold.

[0083] In some embodiments, a semi-supervised learning model based on a consistency regularization term can be constructed, and the semi-supervised learning model can be a fully connected neural network, such as a fully connected recurrent neural network model, a fully connected transform network model, and the like, for optimization requirements of inference speed and training efficiency. Since the input and output dimensions of the trajectory prediction task are low, the fully connected neural network can fully capture the mapping relationship between the input features and the output variables under the premise of ensuring computational efficiency, and achieve accurate trajectory prediction effect.

[0084] In order to improve the learning ability of the model, a total loss function including a supervised loss term and a consistency regularization loss term is constructed. The supervised loss term is used to minimize the prediction error of the labeled sample, to ensure the accuracy of the model on the labeled data; the consistency regularization loss term is used to apply Gaussian noise disturbance on the unlabeled sample, and to calculate the output error between the original sample and the disturbed sample, so that the prediction of the model on the unlabeled data remains stable, thereby improving the generalization ability; the two parts of the loss are optimized by using an adaptive weight coefficient, so that the model can balance the influence of supervised learning and semi-supervised learning in the training process, to ensure the effective use of labeled data and to enhance the learning ability of unlabeled data, thereby achieving better prediction performance while maintaining computational efficiency.

[0085] The mathematical expression of the total loss function (composite loss function) is as follows:

[0086] L total =L supervised +λ·L unsupervised (1-3);

[0087] Wherein, L supervised is a supervised loss term (i.e., a first loss function corresponding to the labeled trajectory sample data); L unsupervised is a consistency regularization loss term (i.e., a second loss function corresponding to the unlabeled trajectory sample data after Gaussian noise disturbance); λ is a hyperparameter for controlling the consistency regularization weight (i.e., a weight parameter corresponding to the second loss function), for balancing the contribution of the labeled trajectory sample data and the unlabeled trajectory sample data;

[0088] The supervised loss term is based on the error between the predicted value of the labeled sample (labeled trajectory sample data) and the true label, and is specifically as follows:

[0089]

[0090] Wherein, N l represents the total number of labeled samples (i.e., labeled trajectory sample data), f θ is a prediction function of the semi-supervised learning model, and xi and yi are the input feature and the corresponding true label of the ith labeled sample respectively.

[0091] To further improve the utilization efficiency of unlabeled data, a consistency regularization loss term is introduced. By constraining the difference between the original output and the output after adding noise of the unlabeled sample, the stability of the model is optimized, and its mathematical form is:

[0092]

[0093] where N u is the number of unlabeled samples, xi is the input feature of the ith unlabeled sample (i.e., the trajectory data in the unlabeled trajectory sample data), ∈ ~ N(0, σ 2 ) represents Gaussian noise with mean 0 and variance σ 2 .

[0094] Figure 5 The structure and training process of the semi-supervised learning neural network model are given. In the training process, the data set is divided into labeled data and unlabeled data, and different strategies are used for optimization. For the unlabeled data x ul , first, Gaussian noise disturbance is applied to generate enhanced samples to improve the robustness and generalization ability of the model. In the forward propagation stage of the model, the outputs z ul and of the original unlabeled sample and the disturbed sample after passing through the fully connected neural network MLP are calculated respectively, and the mean square error of the prediction results of the two is calculated to construct the consistency regularization loss, ensuring that the model has high stability under different input disturbances. For the labeled data x l , after passing through the fully connected neural network, the mean square error between the predicted output z l and the true label y l is calculated to measure the prediction accuracy of the model.

[0095] By coupling the supervised loss and the unsupervised loss with an adaptive weight coefficient, a composite loss function of the neural network is formed. This composite loss function not only improves the prediction accuracy of the model using labeled data, but also enhances the stability and uncertainty measurement ability of the model through self-supervised optimization of unlabeled data. Based on the composite loss function, gradient backpropagation is performed to optimize the neural network parameters, so that the model can efficiently learn the trajectory prediction rule under the condition of low labeled data demand and improve the uncertainty measurement ability of future trajectories.

[0096] The semi-supervised learning mechanism reduces the requirement for labeled data, so that a small amount of labeled data can be combined with a large amount of unlabeled data for training, effectively reducing the dependence on high-quality data and reducing the data labeling cost. The semi-supervised learning mechanism improves the generalization ability of the model, so that it can still maintain stable prediction performance in the case of insufficient data or large scene changes, and improves the applicability of the autonomous driving system.

[0097] The semi-supervised learning mechanism reduces the requirement for labeled data, so that a small amount of labeled data can be combined with a large amount of unlabeled data for training, effectively reducing the dependence on high-quality data and reducing the data labeling cost. The semi-supervised learning mechanism improves the generalization ability of the model, so that it can still maintain stable prediction performance in the case of insufficient data or large scene changes, and improves the applicability of the autonomous driving system.

[0098] Figure 6 For the flowchart of generating the first predicted trajectory in some embodiments of the vehicle anti-collision method according to the present disclosure, as shown in Figure 6

[0099] Step S601, using a vehicle trajectory algorithm, determining a first position and a first heading angle of the vehicle at a future time according to the running state information, the current speed, the vehicle wheelbase and the steering angle.

[0100] Step S602, generating a first predicted trajectory according to the running state information, the first position and the first heading angle.

[0101] The vehicle trajectory algorithm can be various algorithms, for example, the vehicle trajectory algorithm is a future trajectory iteration algorithm based on vehicle steering angle, which uses a step-by-step iteration calculation method, dynamically updates the position and heading angle based on the running state information, current speed, vehicle wheelbase and steering angle information of the vehicle, and generates a future trajectory.

[0102] Figure 7 For a schematic diagram of the trajectory iteration algorithm based on vehicle steering angle (vehicle trajectory algorithm), assuming that the initial running state of the vehicle is (x0, y0, θ0), where (x0, y0) represents the initial position of the vehicle and θ0 represents the initial heading angle. The state is updated at each time step dt. The position update formula is as follows:

[0103] x t+1 =x t +vcos(θ t )·dt (1-6);

[0104] y t+1 =y t +vsin(θ t )·dt (1-7);

[0105] ​Where v represents the vehicle's current speed, θ t The current heading angle. Formulas (1-6) and (1-7) decompose the velocity in a two-dimensional coordinate system based on the vehicle's motion characteristics, and calculate the vehicle's displacement in the x and y directions respectively, making the trajectory prediction more consistent with the vehicle's dynamic characteristics.

[0106] To ensure the physical validity of trajectory iteration, the heading angle update depends on the relationship between velocity and steering angle, and its calculation formula is as follows:

[0107]

[0108] Where v represents the initial velocity, L represents the vehicle wheelbase, and δ represents the steering angle. Formula (1-8) uses the vehicle kinematics model to calculate the change in heading angle through the relationship between velocity, wheelbase, and steering angle, ensuring that the trajectory prediction conforms to the vehicle's driving pattern. By iteratively calculating the vehicle's position information and heading angle changes, this method can efficiently generate future trajectories, providing accurate data support for path planning and trajectory prediction in autonomous driving systems.

[0109] For example, vehicle trajectory algorithms include formulas (1-6), (1-7), and (1-8), etc. These algorithms utilize the vehicle trajectory information (x) to determine the trajectory based on the vehicle's operational status. t y t θ t The vehicle's first position (x) at a future time is determined by using the current speed (v), wheelbase (L), and steering angle (δ). t+1 y t+1 ) and first heading angle (θ) t+1 The first predicted trajectory can be generated by iterating based on the operating status information, the first position, and the first heading angle.

[0110] In some embodiments, based on the first predicted trajectory, the change in heading angle corresponding to the current speed is obtained; based on the current speed and the corresponding change in heading angle, the maximum left correction angle and the maximum right correction angle corresponding to the steering angle are obtained using a trained prediction model; based on the steering angle and the maximum left correction angle and the maximum right correction angle, the first steering correction angle and the second steering correction angle are determined.

[0111] Figure 8 This is a flowchart illustrating the process of determining vehicle behavior uncertainty metrics using a semi-supervised learning neural network based on consistency regularization terms. The process begins by obtaining the heading angle change Δθ corresponding to the vehicle's current speed v based on the first predicted trajectory. The current speed v and the heading angle change Δθ are then input as features into the semi-supervised learning neural network model. By learning the vehicle's motion patterns and behavioral uncertainties, the semi-supervised learning neural network model ultimately outputs the maximum correction angle Δδ for the left and right side behavior uncertainty metrics.l and Δδ r , Δδ l and Δδ r are the left and right maximum correction angles corresponding to the steering angle, respectively, for predicting the maximum deflection angle range that the vehicle can possibly occur. Δδ l and Δδ r are added or subtracted from the current vehicle steering angle δ, respectively, to obtain the maximum steering angle of the vehicle behavior uncertainty measure, i.e., to obtain the first and second steering correction angles, thereby further modeling the future motion trajectory of the vehicle.

[0112] Based on the same method as generating the first predicted trajectory, the second position and the second heading angle of the vehicle at the future time are determined according to the running state information, the current speed, the vehicle wheelbase and the first steering correction angle, using the vehicle trajectory algorithm; and the second predicted trajectory is generated according to the running state information, the second position and the second heading angle. Based on the same method as generating the first predicted trajectory, the third position and the third heading angle of the vehicle at the future time are determined according to the running state information, the current speed, the vehicle wheelbase and the second steering correction angle, using the vehicle trajectory algorithm; and the third predicted trajectory is generated according to the running state information, the third position and the third heading angle.

[0113] In the trajectory prediction process, future trajectory iterative calculation can be used to deduce the future trajectory of the vehicle according to different steering angles, and finally output three possible driving trajectories: the standard future trajectory (first trajectory): the future trajectory calculated based on the current steering angle, reflecting the predicted driving path of the vehicle under the condition of no additional interference; the left uncertainty trajectory (second trajectory): by introducing the maximum left correction angle, simulating the future trajectory of the vehicle under the possible left deflection condition, representing the maximum left deflection behavior that the vehicle can possibly occur; the right uncertainty trajectory (third trajectory): by introducing the maximum right correction angle, simulating the future trajectory of the vehicle under the possible right deflection condition, representing the maximum right deflection behavior that the vehicle can possibly occur.

[0114] By deducing the future trajectory of the vehicle according to different steering angles, not only more accurate future trajectory estimation can be provided, but also the steering uncertainty of the vehicle can be effectively characterized, so that the automatic driving system has stronger safety and adaptability in the process of collision detection, and provides reliable support for driving decision-making in complex traffic environment.

[0115] Figure 9 The flowchart for performing the collision avoidance process in some embodiments of the vehicle collision avoidance method according to the present disclosure is shown as follows: Figure 9

[0116] ​Step S901, according to the first predicted trajectory, the second predicted trajectory, the third predicted trajectory of the vehicle and the first predicted trajectory, the second predicted trajectory, the third predicted trajectory and the size information of the other vehicle, whether there is a collision point is determined.

[0117] Step S902, when it is determined that there is a collision point, the time of collision is determined according to the speed of the vehicle and the speed of the other vehicle.

[0118] Step S903, in the case where the time interval between the time of collision and the current time is less than the safety time threshold, the pre-warning is triggered and the deceleration processing is carried out.

[0119] For example, whether there is a collision point can be determined according to the first predicted trajectory, the second predicted trajectory, the third predicted trajectory and the size information of the vehicle A and the first predicted trajectory, the second predicted trajectory, the third predicted trajectory and the size information of the other vehicle B. If it is determined that there is a collision point, the time t1 of collision is determined according to the speed of the vehicle A and the speed of the other vehicle B. In the case where the time interval between the time t1 of collision and the current time is less than the safety time threshold 5 seconds, the pre-warning is triggered and the deceleration processing is carried out on the vehicle A, and the pre-warning can be that the pre-warning information is popped up on the control screen of the vehicle A, etc.

[0120] In some embodiments, TTC (Time To Collision) refers to the time required for two vehicles (or other moving bodies) to collide at the current speed and trajectory, which is an important indicator for evaluating potential collision risk in the field of vehicle safety (such as autonomous driving, V2V communication and anti-collision system). A TTC anti-collision algorithm based on uncertainty measurement trajectory can be constructed, and the behavior uncertainty measurement trajectory is incorporated into the anti-collision decision framework. The core idea of the TTC anti-collision algorithm is to analyze whether there is an intersection point between the future trajectories of the two vehicles. If the intersection point exists, it indicates a potential collision risk. When the time interval between the two vehicles reaching the collision point is lower than the safety threshold, the system will trigger a pre-warning and take deceleration measures.

[0121] The optimized TTC anti-collision algorithm provided by the present disclosure is shown in Figure 10 and Figure 11 The influence of vehicle geometry is introduced in the TTC framework to avoid misjudgment caused by ignoring the length and width of the vehicle. By using the vehicle size information and correcting the position of the collision point in the TTC calculation process, the calculation result is more consistent with the actual scene. Under the existing TTC algorithm, the intersection point of the trajectories of the two vehicles is considered as the potential collision point, and the optimized TTC anti-collision algorithm of the present disclosure considers the vehicle size and corrects two collision points that are more consistent with the physical reality, as shown in Figure 10The optimized TTC anti-collision algorithm can calculate appropriate deceleration according to the current vehicle speed and TTC, so that the braking process is more stable and intelligent, and the single intersection judgment is extended to regional judgment, which improves the robustness of collision detection and reduces false positives caused by small deviations in the trajectory. Figure 11 As shown in FIG. 6, if the vehicle trajectories do not directly intersect, but due to the existence of vehicle size, collision may still occur, therefore, the optimized TTC anti-collision algorithm calculates the closest distance between the two vehicle trajectories, and if it is less than the safety distance, a collision point based on the closest point is constructed for TTC calculation.

[0122] The TTC anti-collision method based on uncertainty measurement trajectory provided by the present disclosure is as shown in FIG. 7, which introduces three types of trajectories (standard future trajectory, left uncertainty trajectory, right uncertainty trajectory) into the anti-collision detection framework, and uses the maximum deflection trajectory derived by the uncertainty measurement for collision detection. Figure 12 For example, according to the first prediction trajectory (standard future trajectory), the second prediction trajectory (left uncertainty trajectory), the third prediction trajectory (left uncertainty trajectory) of vehicle A and the first prediction trajectory, the second prediction trajectory, the third prediction trajectory and the size information of other vehicle B, it is determined whether there is a collision point between any prediction trajectory of vehicle A and other vehicle B, so as to improve the safety, real-time performance and robustness of the automatic driving system, reduce the false alarm rate caused by trajectory prediction error, and provide more stable and reliable decision support for the automatic driving system.

[0123] The present disclosure improves the reliability of anti-collision detection by combining uncertainty measurement; the use of the maximum deflection trajectory derived by the uncertainty measurement for anti-collision detection improves the safety, real-time performance and robustness of the automatic driving system, significantly reduces the false alarm rate caused by prediction error, and provides more stable decision support for the automatic driving system; the introduction of the geometric size of the vehicle avoids the misjudgment caused by the neglect of the length and width of the vehicle in the TTC method; the use of the current speed, acceleration and TTC of the vehicle to select the deceleration strategy optimizes the traditional single-point collision judgment method to regional collision analysis, and improves the robustness and applicability of the detection.

[0124] In some embodiments, V2V (Vehicle to Vehicle) is a wireless communication technology for direct exchange of information between vehicles, such as speed, position, direction, etc., which can improve road safety and optimize traffic efficiency, and plays a key role in applications such as collision prevention, cooperative driving and autonomous driving.

[0125] The V2V device receives information transmitted by other vehicles, including first, second, and third predicted trajectories, vehicle speed, and size information. The V2V device can be various devices that rely on V2V communication technology to enable vehicles to broadcast their predicted trajectories, current position information, heading angle, speed, and other key dynamic parameters in real time, ensuring that surrounding vehicles can accurately perceive the motion state of adjacent traffic subjects and implement more efficient risk assessment and collision avoidance strategies.

[0126] In traditional collision avoidance processing methods, vehicles usually rely on sensors (such as laser radar, millimeter wave radar, and cameras) mounted on the vehicle for environment perception and calculate future trajectories based on autonomous prediction models. However, in complex dynamic traffic environments, factors such as line-of-sight obstruction, sensor failure, and uncertainty in target trajectory prediction can cause collision warning delays or false positives. The introduction of V2V communication breaks through the limitations of single-vehicle perception, allowing vehicles to exchange position information, speed, heading angle, and predicted trajectories in real time, thereby building a global traffic situation awareness model and improving the accuracy of collision risk assessment. Based on this, not only does the vehicle rely on autonomous prediction, but it can also optimize decision-making strategies based on the dynamic characteristics of adjacent traffic flow, thereby providing more forward-looking warning and avoidance measures in emergency situations. The introduction of V2V communication enables the automatic driving system to adopt a distributed computing architecture, with each vehicle acting as both a data source and a computing node, collaboratively performing collision risk assessment and avoidance decision-making. Through dynamic weight adjustment strategies, the system can select the most influential adjacent vehicles for collision detection within the communication range to optimize data fusion efficiency and reduce communication load. While ensuring system safety, this method improves the real-time performance and robustness of collision avoidance detection.

[0127] The present disclosure optimizes collision avoidance detection based on vehicle information exchange through V2V communication, enabling efficient data sharing between vehicles and improving the accuracy of trajectory prediction, thereby optimizing the collision avoidance detection system. Through V2V communication, vehicles can obtain real-time trajectory information and other information from adjacent vehicles, enabling the collision avoidance detection system to rely not only on single-vehicle prediction but also on the global situation of adjacent traffic flow, thereby providing more accurate warning and avoidance strategies in emergency situations.

[0128] The vehicle collision avoidance method of the present disclosure, by using the consistent regularization term constraint neural network model and the trajectory iteration algorithm based on the vehicle steering angle, combines the accuracy of the traditional trajectory prediction method and the generalization ability of the multi-modal trajectory prediction model, which can not only accurately predict the possible future driving trajectory of the vehicle, but also effectively quantify the uncertainty of different trajectories and improve the reliability of the prediction results. Using the future trajectory iteration algorithm based on the vehicle steering angle can more accurately depict the vehicle motion trend, so that the automatic driving system can better respond to the dynamic changes in the complex traffic environment and provide more robust support for collision avoidance detection and path planning.

[0129] As shown in Figure 13 The vehicle collision avoidance device can include a memory 1301, a processor 1302, a communication interface 1303, and a bus 1304. The memory 1301 is configured to store instructions, and the processor 1302 is coupled to the memory 1301 and configured to execute the vehicle collision avoidance method as described above based on the instructions stored in the memory 1301.

[0130] The memory 1301 can be a high-speed RAM memory, a non-volatile memory, etc., and the memory 81 can also be a memory array. The memory 1301 can also be divided into blocks, and the blocks can be combined into a virtual volume according to certain rules. The processor 1302 can be a central processing unit CPU, or an application specific integrated circuit ASIC, or one or more integrated circuits configured to implement the vehicle collision avoidance method of the present disclosure. The vehicle collision avoidance device can be a device in an automatic driving system.

[0131] In some embodiments, the present disclosure provides a computer readable storage medium storing computer instructions, which when executed by a processor, implement the vehicle collision avoidance method as described in any of the above embodiments.

[0132] In some embodiments, the present disclosure provides a vehicle including the vehicle collision avoidance device as described in any of the above embodiments. The vehicle can be various vehicles, such as a vehicle configured with an automatic driving system, etc.

[0133] The computer readable storage medium can be a combination of one or more computer readable media. The computer readable media can be a computer readable signal medium or a computer readable storage medium. The computer readable storage medium can include, for example, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of the computer readable storage medium include an electrical connection having one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0134] Embodiments of the present disclosure can also be a computer program product that includes a computer program instruction, which, when executed by a processor, causes the processor to perform the steps of the methods according to various embodiments of the present disclosure described in the above "Exemplary Methods" section of the specification.

[0135] The above describes the basic principles of the present disclosure in conjunction with specific embodiments, but it should be noted that the advantages, benefits, effects and the like mentioned in the present disclosure are only examples and are not limiting, and these advantages, benefits, effects and the like cannot be considered as necessary for each embodiment of the present disclosure. In addition, the above specific details are only for the purpose of example and for the purpose of understanding, and the above details do not limit the present disclosure to be necessarily implemented with the above specific details.

[0136] Each embodiment in the specification is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between each embodiment can be understood by mutual reference. For system embodiments, since they basically correspond to method embodiments, the description is relatively simple, and the relevant parts can be understood by referring to the part of the method embodiment.

[0137] The block diagrams of the devices, apparatuses, equipment, systems involved in the present disclosure are only exemplary examples and are not intended to require or imply the connection, arrangement, configuration as shown in the block diagram. As those skilled in the art will recognize, these devices, apparatuses, equipment, systems can be connected, arranged, configured in any manner. Words such as "include", "contain", "have" and the like are open-ended words, which mean "including but not limited to", and can be used interchangeably. The words "or" and "and" used herein mean the word "and / or", and can be used interchangeably unless the context clearly indicates otherwise. The word "such as" used herein means the phrase "such as but not limited to", and can be used interchangeably.

[0138] It is also important to note that the construction and arrangement of the devices, equipment, and methods as shown in the various examples is illustrative only. Although only a few embodiments have been described, many modifications, changes, and substitutions can be made by one having ordinary skill in the art without departing from the application. Therefore, the above description is not intended to limit the scope of the application and the spirit, which is defined by the appended claims.

[0139] The above description of disclosed aspects is intended to be illustrative, and not restrictive. Many embodiments and applications other than the examples provided would be apparent upon reading the above description. The scope of the application should be determined, not with reference to the above description, but should instead be determined with reference to the appended claims, along with the full scope of equivalents to which such claims are entitled. It is anticipated and intended that future developments will occur in the arts discussed herein, and that the disclosed aspects will be incorporated into such future embodiments. In summary, it is intended that the application be construed as including all such future developments as fall within the scope of the appended claims.

[0140] The above description has been presented for the purpose of illustration and description. It is not intended to be exhaustive or to limit the embodiments of the disclosure to the precise forms disclosed. Although several example aspects and embodiments have been discussed, it should be understood that modifications, combinations, sub-combinations, and alternatives can become apparent to those skilled in the art having the benefit of this disclosure. It is intended that the scope of the disclosure should be interpreted by reference to the claims and should not be limited by reference to the description or the examples.

Claims

1. A vehicle anti-collision method, comprising: generating a first predicted trajectory according to running state information, a current speed, a wheelbase and a steering angle of a vehicle; obtaining a heading angle change corresponding to the current speed according to the first predicted trajectory; obtaining a left maximum correction angle and a right maximum correction angle corresponding to the steering angle according to the current speed and the corresponding heading angle change by using a trained prediction model; determining a first steering correction angle and a second steering correction angle according to the steering angle and the left maximum correction angle and the right maximum correction angle; generating a second predicted trajectory according to the running state information, the current speed, the wheelbase and the first steering correction angle, and generating a third predicted trajectory according to the running state information, the current speed, the wheelbase and the second steering correction angle; performing anti-collision processing according to the first predicted trajectory, the second predicted trajectory, the third predicted trajectory, the speed and the size information of the vehicle and the first predicted trajectory, the second predicted trajectory, the third predicted trajectory, the speed and the size information of another vehicle.

2. The method of claim 1, wherein, The prediction model comprises a semi-supervised learning model, and the method comprises: generating labeled trajectory sample data and unlabeled trajectory sample data by using an autonomous driving simulation tool; training the semi-supervised learning model using the labeled trajectory sample data and the unlabeled trajectory sample data to obtain the trained semi-supervised learning model. 3.The method of claim 2, wherein: the labeled trajectory sample data comprises trajectory data and corresponding labels, and the unlabeled trajectory sample data comprises the trajectory data; the trajectory data comprises a vehicle simulation speed and a corresponding heading angle change; the labels comprise a left maximum correction angle and a right maximum correction angle corresponding to a vehicle simulation steering angle. 4.The method of claim 2, comprising: constructing a first loss function corresponding to the labeled trajectory sample data; constructing a second loss function corresponding to the unlabeled trajectory sample data after Gaussian noise disturbance; generating a composite loss function according to the first loss function and the second loss function; adjusting the semi-supervised learning model by using the composite loss function to obtain the trained semi-supervised learning model.

5. The method of claim 4, wherein, The generating of the composite loss function according to the first loss function and the second loss function comprises: weighting the second loss function using a weight parameter corresponding to the second loss function; generating the composite loss function according to the first loss function and the weighted second loss function.

6. The method of claim 4, wherein, The adjusting of the semi-supervised learning model by using the composite loss function to obtain the trained semi-supervised learning model comprises: determining a prediction error value by using the composite loss function; adjusting parameters of the semi-supervised learning model according to the prediction error value until the prediction error value is lower than a preset error threshold to obtain the trained semi-supervised learning model.

7. The method of claim 1, wherein, The operating state information includes a vehicle position and a heading angle; the generating a first predicted trajectory according to the operating state information, the current speed, the vehicle wheelbase and the steering angle includes: determining, by a vehicle trajectory algorithm, a first position and a first heading angle of the vehicle at a future time according to the operating state information, the current speed, the vehicle wheelbase and the steering angle; generating the first predicted trajectory according to the operating state information, the first position and the first heading angle.

8. The method of claim 7, wherein, The generating a second predicted trajectory according to the operating state information, the current speed, the vehicle wheelbase and the first steering correction angle includes: determining, by the vehicle trajectory algorithm, a second position and a second heading angle of the vehicle at the future time according to the operating state information, the current speed, the vehicle wheelbase and the first steering correction angle; generating the second predicted trajectory according to the operating state information, the second position and the second heading angle.

9. The method of claim 7, wherein, The generating a third predicted trajectory according to the operating state information, the current speed, the vehicle wheelbase and the second steering correction angle includes: determining, by the vehicle trajectory algorithm, a third position and a third heading angle of the vehicle at the future time according to the operating state information, the current speed, the vehicle wheelbase and the second steering correction angle; generating the third predicted trajectory according to the operating state information, the third position and the third heading angle.

10. The method of claim 1, wherein, The performing a collision avoidance process according to the first predicted trajectory, the second predicted trajectory, the third predicted trajectory, the vehicle speed and the outer dimension information of the vehicle and the first predicted trajectory, the second predicted trajectory, the third predicted trajectory, the vehicle speed and the outer dimension information of the other vehicle includes: determining whether there is a collision point according to the first predicted trajectory, the second predicted trajectory, the third predicted trajectory and the outer dimension information of the vehicle and the first predicted trajectory, the second predicted trajectory, the third predicted trajectory and the outer dimension information of the other vehicle; when it is determined that there is a collision point, determining a time of collision according to the vehicle speed and the vehicle speed of the other vehicle; when a time interval between the time of collision and a current time is less than a safety time threshold, triggering a warning and performing a deceleration process.

11. The method of any one of claims 1 to 10, comprising: receiving, by a vehicle-to-vehicle communication (V2V) device, information transmitted by the other vehicle, including the first predicted trajectory, the second predicted trajectory, the third predicted trajectory, the vehicle speed and the outer dimension information of the other vehicle.

12. A vehicle collision avoidance apparatus, comprising: a memory; and a processor coupled to the memory, the processor configured to perform the method of any one of claims 1 to 11 based on instructions stored in the memory.

13. A vehicle, comprising: the vehicle collision avoidance apparatus of claim 12.

14. A computer-readable storage medium storing computer instructions, the instructions, when executed by a processor, implementing the steps of the method of any one of claims 1 to 11.

15. A computer program product comprising a computer program, characterized in that, The computer program, which when executed by the processor, implements the steps of the method of any one of claims 1 to 11.

Images