Method for dynamic assessment of vehicle collision risk based on multi-modal perception

Abstract

The present application relates to the technical field of intelligent driving, and especially relates to a vehicle collision risk dynamic evaluation method based on multi-modal perception. In the process of vehicle driving, the driving information and the environmental characteristic information of the vehicle are collected through a sensor array, the vehicle incremental movement is predicted according to the environmental characteristic information, the second driving speed of the vehicle is obtained, the vehicle incremental movement is identified, and the identified vehicle incremental movement distance is obtained. According to the first driving speed and the second driving speed, the fusion driving speed is calculated and obtained, the predicted vehicle incremental movement distance and the identified vehicle incremental movement distance are weighted and calculated, and the vehicle incremental movement distance is obtained. According to the first monitoring distance and the vehicle incremental movement distance, the second monitoring distance is calculated and obtained, and the fusion driving speed is used for correction to obtain the third monitoring distance, the collision risk is evaluated, and the evaluation result is obtained. The technical effect of accurately and quickly evaluating the collision risk is achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent driving, and in particular to a vehicle collision risk dynamic assessment method based on multi-modal perception. BACKGROUND

[0002] The existing vehicle collision risk assessment technology mainly relies on a single distance threshold to determine the vehicle collision risk, and only measures the vehicle distance through sensors without considering the influence of environmental factors (such as wet road surface and weather conditions) and vehicle dynamic changes (such as displacement caused by skidding) on the collision risk, resulting in high misjudgment and omission rates in complex road conditions and failing to meet the demand of intelligent driving for accurate collision risk assessment. In the process of vehicle driving, multi-modal data cannot be effectively fused, and collision risk cannot be dynamically and accurately assessed, and the problems of large assessment error, insufficient real-time performance and reliability caused by environmental and vehicle dynamic changes cannot be solved. In summary, the existing technology has the technical problem of being unable to accurately and quickly assess the vehicle collision risk. SUMMARY

[0003] The present application provides a vehicle collision risk dynamic assessment method based on multi-modal perception to solve the technical problem of being unable to accurately and quickly assess the vehicle collision risk in the prior art.

[0004] The technical solution of the present application to solve the above technical problem is as follows:

[0005] In the first aspect, the present application provides a vehicle collision risk dynamic assessment method based on multi-modal perception, which comprises: a vehicle collision risk dynamic assessment method based on multi-modal perception, the method comprising: in the process of vehicle driving, collecting the first driving speed, the first monitoring distance and the environmental feature information of the vehicle through a sensor array, predicting the incremental movement of the vehicle according to the environmental feature information to obtain the predicted incremental movement distance of the vehicle; obtaining the second driving speed of the vehicle, combining the first driving speed to identify the incremental movement of the vehicle to obtain the identified incremental movement distance of the vehicle; calculating the fusion driving speed according to the first driving speed and the second driving speed, configuring the prediction weight and the identification weight, and performing weighted calculation on the predicted incremental movement distance of the vehicle and the identified incremental movement distance of the vehicle to obtain the incremental movement distance of the vehicle; calculating the second monitoring distance according to the first monitoring distance and the incremental movement distance of the vehicle, correcting the third monitoring distance by using the fusion driving speed, and performing collision risk assessment to obtain the assessment result.

[0006] Optionally, during the driving of the vehicle, the first driving speed, the first monitoring distance and the environmental feature information of the vehicle are collected by the sensor array, the vehicle incremental movement is predicted according to the environmental feature information, and the predicted vehicle incremental movement distance is obtained, including: during the driving of the vehicle, the first driving speed, the first monitoring distance and the environmental feature information of the vehicle are collected by the sensor array, wherein the sensor array includes a ranging sensor, a speed sensor and a visual sensor, and the environmental feature information includes an environmental visual image; a trained environmental incremental movement predictor is called; the environmental feature information is input into the environmental incremental movement predictor, and a predicted vehicle incremental movement distance is output.

[0007] The training step of the environmental incremental movement predictor includes: collecting a sample environmental feature information set according to the test data of the vehicle incremental movement, and collecting the vehicle incremental movement distance under different sample environmental feature information to obtain a sample predicted vehicle incremental movement distance set; based on machine learning, an environmental incremental movement predictor is constructed; the sample environmental feature information set and the sample predicted vehicle incremental movement distance set are used to supervise the training of the environmental incremental movement predictor, and the training is completed after convergence.

[0008] Optionally, the second driving speed of the vehicle is obtained, and the vehicle incremental movement is identified in combination with the first driving speed to obtain an identified vehicle incremental movement distance, including: the second driving speed is obtained through the control system of the vehicle; the first driving speed and the second driving speed are input into a speed incremental movement identifier to identify and output an identified vehicle incremental movement distance, wherein the speed incremental movement identifier is trained using a sample first driving speed set, a sample second driving speed set and a sample identified vehicle incremental movement distance set.

[0009] Optionally, the predicted vehicle incremental movement distance and the identified vehicle incremental movement distance are weighted to obtain the vehicle incremental movement distance, including: the predicted vehicle incremental movement distance and the identified vehicle incremental movement distance are weighted using the prediction weight and the identification weight to obtain the vehicle incremental movement distance.

[0010] Optionally, the second monitoring distance is calculated according to the first monitoring distance and the vehicle incremental movement distance, and the third monitoring distance is obtained by correcting the fusion driving speed, the collision risk is evaluated, and the evaluation result is obtained, including: the second monitoring distance is obtained by subtracting the vehicle incremental movement distance from the first monitoring distance; the second monitoring distance is corrected and calculated according to the ratio of the preset average driving speed to the fusion driving speed to obtain the third monitoring distance; the collision risk coefficient is obtained by calculating the ratio of the safe collision distance threshold to the third monitoring distance, and is taken as the evaluation result.

[0011] By implementing the present application, the first driving speed, the first monitoring distance and the environmental feature information of the vehicle can be collected by the sensor array during the driving of the vehicle, the vehicle incremental movement prediction is carried out according to the environmental feature information, the predicted vehicle incremental movement distance is obtained, the potential skidding risk is predicted in advance by using the environmental visual image and other information, and the forward-looking perception of the driving risk of the vehicle is realized.

[0012] By implementing the present application, the second driving speed of the vehicle can be obtained, the vehicle incremental movement identification is carried out in combination with the first driving speed, the identified vehicle incremental movement distance is obtained, and the vehicle skidding state can be quickly detected to provide instant data support for the collision risk assessment.

[0013] By implementing the present application, the fusion driving speed can be calculated according to the first driving speed and the second driving speed, the prediction weight and the identification weight are configured, the predicted vehicle incremental movement distance and the identified vehicle incremental movement distance are weighted and calculated, the vehicle incremental movement distance is obtained, the advantages of different information sources are effectively integrated, and the accuracy and robustness of the vehicle incremental movement distance calculation are improved.

[0014] By implementing the present application, the second monitoring distance can be calculated according to the first monitoring distance and the vehicle incremental movement distance, the fusion driving speed is used for correction to obtain the third monitoring distance, the collision risk assessment is carried out, the assessment result is obtained, a reliable decision basis is provided for the vehicle active safety system, and the reliability and practicality of the risk assessment are enhanced.

[0015] In summary, by implementing the present application, the technical effect of accurately and quickly assessing the collision risk can be achieved. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 A flowchart of a vehicle collision risk dynamic assessment method based on multi-modal perception provided by the present application is shown. DETAILED DESCRIPTION

[0017] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0018] In the description of the present application, the terms "first", "second" are only for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the technical features indicated. Therefore, the features defined with "first", "second" can explicitly or implicitly include one or more features. In the description of the present application, the meaning of "a plurality of" is two or more, unless otherwise explicitly specified.

[0019] In the description of the present application, the term "for example" is used to indicate "as an example, illustration or description". Any embodiment described as "for example" in the present application is not necessarily interpreted as more preferred or more advantageous than other embodiments. The following description is given in order to enable any person skilled in the art to implement and use the present application. In the following description, details are listed for the purpose of explanation. It should be understood that those skilled in the art can recognize that the present application can be implemented without using these specific details. In other examples, well-known structures and processes will not be described in detail to avoid unnecessary details making the description of the present application obscure. Therefore, the present application is not intended to be limited to the embodiments shown, but is consistent with the broadest scope in accordance with the principles and characteristics disclosed.

[0020] Embodiment one, as Figure 1 shown, the embodiment of the present application provides a vehicle collision risk dynamic assessment method based on multi-modal perception, comprising:

[0021] S100: During vehicle driving, a sensor array is used to collect a first driving speed, a first monitoring distance and environmental feature information of the vehicle, to perform vehicle incremental movement prediction according to the environmental feature information, and to obtain a predicted vehicle incremental movement distance;

[0022] S200: Obtain a second driving speed of the vehicle, combine the first driving speed, and perform vehicle incremental movement identification to obtain an identified vehicle incremental movement distance;

[0023] S300: According to the first driving speed and the second driving speed, a fusion driving speed is calculated and obtained, a prediction weight and an identification weight are configured, and the predicted vehicle incremental movement distance and the identified vehicle incremental movement distance are weighted and calculated to obtain a vehicle incremental movement distance;

[0024] S400: According to the first monitoring distance and the vehicle incremental movement distance, a second monitoring distance is calculated and obtained, and the fusion driving speed is used for correction to obtain a third monitoring distance, and collision risk assessment is performed to obtain an assessment result.

[0025] In step S100 of the embodiment, during vehicle driving, a sensor array is used to collect a first driving speed, a first monitoring distance, and environmental feature information of the vehicle, and a vehicle incremental movement prediction is performed according to the environmental feature information to obtain a predicted vehicle incremental movement distance, including:

[0026] During vehicle driving, a sensor array is used to collect a first driving speed, a first monitoring distance, and environmental feature information of the vehicle, wherein the sensor array includes a ranging sensor, a speed sensor, and a vision sensor, and the environmental feature information includes an environmental vision image.

[0027] The trained environmental incremental movement predictor is called.

[0028] The environmental feature information is input into the environmental incremental movement predictor to predict and output a predicted vehicle incremental movement distance.

[0029] In the embodiment, the first driving speed is the actual speed of the vehicle driving, i.e., the movement speed relative to the ground; the first monitoring distance is the distance monitored from the vehicle to a neighboring vehicle or an obstacle during vehicle driving; and the environmental feature information is an environmental vision image, including environmental feature information such as road surface features (e.g., wet, icy, dry), weather conditions (rainy, foggy), and the like.

[0030] The first driving speed can be measured by a speed sensor (e.g., a GPS speed measurement module); the first monitoring distance can be measured by a ranging sensor (e.g., a laser radar, a millimeter wave radar); and the environmental feature information can be collected by a vision sensor (e.g., a camera, an infrared imaging system).

[0031] In the embodiment, the environmental incremental movement predictor is used to predict an incremental movement distance according to environmental feature information. When the road is wet, it can cause the vehicle to slip during braking, resulting in vehicle incremental movement and further causing a collision. Therefore, the vehicle incremental movement distance is the distance that the vehicle can slip due to the road features.

[0032] The training steps of the environmental incremental movement predictor include:

[0033] According to the test data of the vehicle incremental movement, a sample environmental feature information set is collected, and the vehicle incremental movement distance under different sample environmental feature information is collected to obtain a sample predicted vehicle incremental movement distance set.

[0034] Based on machine learning, an environmental incremental movement predictor is constructed.

[0035] The sample environmental feature information set and the sample predicted vehicle incremental movement distance set are used to supervise the training of the environmental incremental movement predictor, and the training is completed after convergence.

[0036] In the embodiments of the present application, for the training samples of the environment incremental movement predictor (i.e. the environment visual images labeled with the actual incremental movement distance of the vehicle), different environments (such as wet road surface, icy road surface) can be simulated in a controllable test site, or natural scene collection can be performed on actual roads (such as rainy day, night), but the driving safety needs to be ensured. Then the environment label (such as “wet road surface”, “dry road surface”, “icy road surface”, etc.) of each frame of image is manually labeled.

[0037] For each collected environment visual image, the actual incremental movement distance of the vehicle in the corresponding environment (through the IMU or high-precision positioning system) is calculated as the label of supervised learning (such as labeling the actual incremental movement distance as 10 m). The incremental movement distance of the vehicle under different sample environment characteristic information is collected to obtain a set of sample predicted incremental movement distance of the vehicle; finally, no less than 10,000 groups of sample data (including different environment scenes) are collected, and are divided into training set: 70%, verification set: 15%, and test set: 15% in proportion, which are used to train the environment incremental movement predictor.

[0038] Optionally, the environment incremental movement predictor can be built using a convolutional neural network. The main structure of the environment incremental movement predictor is as follows from the input layer to the output layer: the input layer receives the environment visual image, the input color image, and the resolution can be 224x224; the convolutional layer 1 has 32 convolutional kernels, the kernel size is 3x3, the step is 1, the same padding is used, and the ReLU activation function is used; the pooling layer 1 uses maximum pooling with a pooling kernel size of 2x2 and a step of 2; the convolutional layer 2 has 64 convolutional kernels, the kernel size is 3x3, the step is 1, the same padding is used, and the ReLU activation function is used; the pooling layer 2 uses maximum pooling with a pooling kernel size of 2x2 and a step of 2; the flattening layer is used to expand the multi-dimensional feature map into a one-dimensional vector; the fully connected layer 1 has 256 neurons, the activation function is ReLU, and the Dropout rate is 0.2; the output layer has 1 neuron, and the linear activation function is used.

[0039] During the training of the environment incremental movement predictor, the optimizer uses Adam; the learning rate is set to 0.001; L2 regularization is used with a coefficient of 0.001; the mean square error (MSE) is used as the loss function; the early stopping strategy is added, i.e. the validation set loss is monitored, and if it does not decrease for 10 consecutive rounds, the training is stopped; the maximum number of training rounds is preset to 500 rounds. Combined with the early stopping strategy, the training usually converges within 100-200 rounds; the convergence standard of the environment incremental movement predictor can be set to the MSE of the verification set being less than 0.01.

[0040] The environment incremental movement predictor is trained by the method until a convergence criterion is reached, and the environment incremental movement predictor is obtained. The environment feature information (environmental visual image) is input into the environment incremental movement predictor, and the predicted vehicle incremental movement distance (such as 10 m) is output.

[0041] In step S200 of the embodiment of the present application, the second driving speed of the vehicle is obtained, and the vehicle incremental movement is identified in combination with the first driving speed to obtain the identified vehicle incremental movement distance, including:

[0042] The second driving speed is obtained by the control system of the vehicle.

[0043] The first driving speed and the second driving speed are input into the speed incremental movement identifier, and the identified vehicle incremental movement distance is obtained by identification output, wherein the speed incremental movement identifier is trained by using a sample first driving speed set, a sample second driving speed set, and a sample identified vehicle incremental movement distance set.

[0044] In the embodiment of the present application, the second driving speed is the expected speed when it is assumed that the tire does not slip with the ground, that is, the theoretical driving speed, that is, the rotational linear speed of the outer edge of the tire, which can be calculated by obtaining the vehicle information through the vehicle control system (such as engine ECU, gearbox gear ratio). The theoretical driving speed can be calculated by reading the engine speed, gearbox gear ratio, tire radius, etc. through the CAN bus or OBD-II interface. For example, the second driving speed can be: second driving speed = (engine speed * tire circumference) / (gearbox gear ratio * main reducer speed ratio).

[0045] As for the speed incremental movement identifier, it needs to be trained by using a sample first driving speed set, a sample second driving speed set, and a sample identified vehicle incremental movement distance set. The sample first driving speed and the sample second driving speed can be directly obtained by the foregoing method, which will not be described here. The sample identified vehicle incremental movement distance is the actual measured vehicle incremental movement distance under a specific sample first driving speed and a sample second driving speed. For example, the actual incremental movement distance can be obtained by using a high-precision positioning system (such as RTK-GPS) or IMU measurement in real time through acceleration, deceleration, emergency braking, or sharp turning tests under different road conditions (dry, wet, icy) in a closed field or a simulation environment, and a plurality of sample identified vehicle incremental movement distances are collected to obtain the sample identified vehicle incremental movement distance set.

[0046] The sample first driving speed set, the sample second driving speed set, and the sample identified vehicle incremental movement distance set are divided into a training set, a validation set, and a test set according to the proportions of 70%, 20%, and 10% respectively, which are used to train the speed incremental movement identifier.

[0047] In the embodiment of the present application, the speed increment movement identifier identifies the abnormal movement distance of the vehicle caused by factors such as slipping by analyzing the difference between the two speeds (the first driving speed and the second driving speed), and therefore can be built using a gated recurrent unit (GRU) network.

[0048] Optionally, the speed increment movement identifier comprises the following parts: an input layer for receiving the speed pair (the first driving speed and the second driving speed) in a sliding time window (such as 3 seconds, a sampling frequency of 30 Hz, a total of 90 time steps), with a feature dimension of 90x2; a GRU layer with 64 neurons for learning the time sequence features of the speed difference; an attention mechanism layer for weighting the time sequence features output by the GRU to highlight key time points (such as the start and end time of slipping); a fully connected layer for reducing the features to 1 dimension to output the identified increment movement distance.

[0049] In the training of the speed increment movement identifier, the Adam optimizer can be used with an initial learning rate of 0.001; the Huber loss function is used; the Dropout (0.2) is used for regularization to prevent overfitting; the batch size is 64, the number of training rounds is 100, the early stopping strategy is introduced, and the validation set loss does not decrease for 5 consecutive rounds; the hierarchical training strategy is used, the GRU layer weight is frozen in the first stage, and the GRU layer is unfrozen for overall fine-tuning in the second stage. In the continuous 100 rounds of training, if the MSE of the validation set does not decrease significantly (such as a decrease of ≤0.05%), it is considered to be converged, and the speed increment movement identifier is obtained. The first driving speed and the second driving speed are input into the speed increment movement identifier, and the identified increment movement distance of the vehicle is obtained.

[0050] In step S300 of the embodiment of the present application, the fusion driving speed is calculated according to the first driving speed and the second driving speed, and the prediction weight and the identification weight are configured, including:

[0051] The mean value is calculated according to the first driving speed and the second driving speed to obtain the fusion driving speed;

[0052] The ratio of the fusion driving speed to the preset average driving speed is calculated to correct and calculate the preset identification weight, and the identification weight is obtained;

[0053] The prediction weight is calculated based on the identification weight.

[0054] In the embodiment of the present application, the first driving speed is the actual driving speed (such as 50 km / h) directly measured by the vehicle-mounted sensor. The second driving speed is the theoretical driving speed (such as 60 km / h) calculated based on the engine output and the transmission system parameters. The fusion driving speed is the mean value of the aforementioned two speeds (such as 55 km / h).

[0055] The preset average driving speed is a typical driving speed obtained according to a large amount of historical data (for example, 30 km / h for urban roads and 80 km / h for expressways), and is used as a reference speed for evaluating the particularity of the current driving state.

[0056] In the embodiment of the present application, the preset recognition weight can be set to 0.5, indicating that the importance of environment prediction and speed recognition is the same at the initial time, and the correction formula can be recognition weight = preset recognition weight * (fused driving speed / preset average driving speed) a, wherein a is an adjustment coefficient (which can be 0.8) for controlling the correction strength. When the fused driving speed is greater than the preset average driving speed (for example, high-speed driving), the risk of vehicle skidding increases, the importance of speed recognition increases, and the recognition weight increases; when the fused driving speed is less than the preset average driving speed (for example, congested road conditions), the influence of environmental factors (such as wet road surface) is more significant, and the recognition weight decreases. After obtaining the recognition weight, the prediction weight can be further calculated, and the calculation method can be prediction weight = 1 - recognition weight. Through this dynamic weight configuration method, the system can adapt to different driving states, optimize the fusion effect of environment prediction and speed recognition, and improve the accuracy of collision risk assessment.

[0057] In step S300 of the embodiment of the present application, the predicted vehicle incremental movement distance and the recognized vehicle incremental movement distance are weighted and calculated to obtain the vehicle incremental movement distance, including:

[0058] The predicted vehicle incremental movement distance and the recognized vehicle incremental movement distance are weighted and calculated using the prediction weight and the recognition weight to obtain the vehicle incremental movement distance.

[0059] In the vehicle collision risk assessment in the embodiment of the present application, the output (predicted vehicle incremental movement distance) of the environment incremental movement predictor is weighted and fused with the output (recognized vehicle incremental movement distance) of the speed incremental movement recognizer, which is a key step to improve the prediction accuracy.

[0060] Optionally, the weighting method can be: vehicle incremental movement distance = prediction weight * vehicle incremental movement distance + recognition weight * recognized vehicle incremental movement distance. For example, when the prediction weight = 0.4, the recognition weight = 0.6, the predicted vehicle incremental movement distance = 1.2 m, and the recognized vehicle incremental movement distance = 0.8 m, the vehicle incremental movement distance = 0.4 * 1.2 m + 0.6 * 0.8 = 0.96 m. The fused vehicle incremental movement distance will be used to correct the actual distance between the vehicle and the obstacle

[0061] In step S400 of the embodiment of the present application, the second monitoring distance is obtained according to the first monitoring distance and the incremental moving distance of the vehicle, and the third monitoring distance is obtained by correcting the fusion driving speed, the collision risk is evaluated, and the evaluation result is obtained, including:

[0062] The second monitoring distance is obtained by subtracting the incremental moving distance of the vehicle from the first monitoring distance;

[0063] The third monitoring distance is obtained by correcting the second monitoring distance according to the ratio of the preset average driving speed to the fusion driving speed;

[0064] The collision risk coefficient is obtained by calculating the ratio of the safety collision distance threshold to the third monitoring distance, and is taken as the evaluation result.

[0065] In the embodiment of the present application, the first monitoring distance is the real-time distance between the vehicle and the adjacent vehicle / obstacle directly measured by the sensor (such as laser radar, millimeter wave radar). The incremental moving distance of the vehicle is obtained by the method in S300. The second monitoring distance is the remaining safety distance after the correction of the slip influence, and the calculation method can be: second monitoring distance = first monitoring distance - incremental moving distance of the vehicle.

[0066] In the embodiment of the present application, the higher the driving speed, the longer the braking distance of the vehicle, and the larger the safety distance to be reserved. Therefore, the second monitoring distance needs to be dynamically adjusted according to the current speed. The specific adjustment method is that the preset average driving speed is taken as the reference speed (such as 30 km / h on urban roads and 80 km / h on expressways), and the fusion driving speed (the average of the first and second driving speeds, reflecting the current actual speed) is combined for calibration.

[0067] The specific correction method can be: third monitoring distance = second monitoring distance * (preset average driving speed / fusion driving speed).

[0068] Further, a ratio of a safe collision distance threshold value and the third monitoring distance is calculated to obtain a collision risk coefficient as the evaluation result. The safe collision distance threshold value is a minimum safe distance calculated according to a vehicle dynamics model (such as a braking distance formula), and can be 50 m under certain environmental conditions and vehicle speed. The safe collision distance threshold value is obtained in the prior art, and will not be described herein. The collision risk coefficient can be calculated as collision risk coefficient = safe collision distance threshold value / third monitoring distance. Further, it can be defined that when the collision risk coefficient ≤ 1, it indicates that it is safe and risk-free, and when the collision risk coefficient > 1, it indicates that there is a collision risk, and the driver needs to be reminded to avoid in time. It should be noted that in the above embodiments, the description of each embodiment is focused on, and the parts not described in detail in a certain embodiment can be referred to the related description of other embodiments.

[0069] Those skilled in the art will appreciate that embodiments of the application can be supplied as a method, a system, or a computer program product. Therefore, the application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage media, etc.) having computer usable program code embodied therein.

[0070] The application is described with reference to flowcharts and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded computer, or other programmable data processing apparatus to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing apparatus produce an apparatus that implements the functions specified in the flowcharts and / or block diagrams. Figure 1 one or more functions specified in one or more of the flowcharts and / or blocks. Figure 1 an apparatus that implements one or more functions specified in one or more of the flowcharts and / or blocks.

[0071] These computer program instructions can also be stored in a computer-readable memory that can direct the computer or other programmable data processing apparatus to work in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including instruction apparatus that implements the functions specified in the flowcharts and / or block diagrams. Figure 1 one or more functions specified in one or more of the flowcharts and / or blocks. Figure 1 an apparatus that implements one or more functions specified in one or more of the flowcharts and / or blocks.

[0072] These computer program instructions can also be loaded into a computer or other programmable data processing devices, so that a series of operational steps are performed on the computer or other programmable devices to generate computer-implemented processes, thus the instructions executed on the computer or other programmable devices provide the function for implementing the processes specified in the flowchart Figure 1 one or more flows and / or blocks Figure 1 one or more blocks or steps of the functions specified in the flowchart

[0073] Although the preferred embodiments of the application have been described, those skilled in the art will be able to make additional modifications and variations to these embodiments without departing from the spirit and scope of the application.

[0074] Obviously, various modifications and changes can be made to the present application by those skilled in the art without departing from the spirit and scope of the present application. Accordingly, it is intended that the present application embrace all such modifications and changes as fall within the scope of the present application and its equivalents.

Claims

1. A method for dynamic assessment of vehicle collision risk based on multimodal perception, characterized in that, The method includes: During vehicle operation, the sensor array collects the vehicle's first driving speed, first monitoring distance, and environmental feature information. Based on the environmental feature information, the incremental movement of the vehicle is predicted to obtain the predicted incremental movement distance of the vehicle. The vehicle's second driving speed is obtained, and combined with the first driving speed, incremental vehicle movement is identified to obtain the identified incremental vehicle movement distance. The second driving speed is the ideal driving speed under conditions where the wheels do not slip. Based on the first driving speed and the second driving speed, a fused driving speed is calculated, prediction weights and recognition weights are configured, and the predicted vehicle incremental movement distance and the recognized vehicle incremental movement distance are weighted to obtain the vehicle incremental movement distance. Based on the first monitoring distance and the incremental vehicle movement distance, the second monitoring distance is calculated and corrected using the fused driving speed to obtain the third monitoring distance. A collision risk assessment is then performed to obtain the assessment result. During vehicle operation, a sensor array collects the vehicle's first speed, first monitoring distance, and environmental characteristic information. Based on the environmental characteristic information, incremental vehicle movement is predicted to obtain the predicted incremental vehicle movement distance, including: During vehicle operation, the sensor array collects the vehicle's first driving speed, first monitoring distance, and environmental feature information. The sensor array includes a ranging sensor, a speed sensor, and a vision sensor, and the environmental feature information includes environmental visual images. Invoke the trained environment incremental movement predictor; The environmental feature information is input into the environmental incremental movement predictor, and the predicted vehicle incremental movement distance is output. Specifically, the process involves obtaining the vehicle's second speed, combining it with the first speed, performing incremental vehicle movement recognition, and obtaining the identified incremental vehicle movement distance, including: The second driving speed is obtained through the vehicle's control system; The first driving speed and the second driving speed are input into the speed increment movement recognizer, and the recognition output is the recognized vehicle increment movement distance. The speed increment movement recognizer is obtained by training with a sample first driving speed set, a sample second driving speed set, and a sample recognized vehicle increment movement distance set. Specifically, based on the first and second driving speeds, a fused driving speed is calculated, and prediction weights and recognition weights are configured, including: The average of the first and second driving speeds is used to obtain the merged driving speed. Calculate the ratio of the fused driving speed to the preset average driving speed, and perform correction calculations on the preset recognition weights to obtain the recognition weights; Based on the identification weights, the prediction weights are calculated. The process of weighting the predicted vehicle incremental movement distance and the identified vehicle incremental movement distance to obtain the vehicle incremental movement distance includes: Using the prediction weight and the identification weight, the predicted vehicle incremental movement distance and the identified vehicle incremental movement distance are weighted and calculated to obtain the vehicle incremental movement distance. Specifically, based on the first monitoring distance and the incremental vehicle movement distance, a second monitoring distance is calculated and corrected using the fused driving speed to obtain a third monitoring distance. A collision risk assessment is then performed to obtain the assessment results, including: The second monitoring distance is obtained by subtracting the incremental movement distance of the vehicle from the first monitoring distance; The second monitoring distance is corrected and calculated based on the ratio of the preset average driving speed to the fused driving speed to obtain the third monitoring distance; The ratio of the safe collision distance threshold to the third monitoring distance is calculated to obtain the collision risk coefficient, which is used as the evaluation result.

2. The method for dynamic assessment of vehicle collision risk based on multimodal perception according to claim 1, characterized in that, The training steps for the environmental incremental movement predictor include: Based on the test data of incremental vehicle movement, a set of sample environmental feature information is collected, and the incremental vehicle movement distance under different sample environmental feature information is collected to obtain a set of sample predicted incremental vehicle movement distances. Based on machine learning, we construct an incremental environmental movement predictor. The environmental incremental movement predictor is trained under supervision using the sample environmental feature information set and the sample predicted vehicle incremental movement distance set, and the training is completed after convergence.

Images