Vehicle collision avoidance warning methods, computer program products and vehicles

Abstract

This application relates to a collision avoidance warning method, computer program product, and vehicle. The method includes: acquiring a set of parameter values ​​for each of multiple operating parameters of the vehicle during a preset time period; the set of parameter values ​​includes the parameter values ​​of the corresponding operating parameter at different time points; preprocessing the multiple set of parameter values ​​to obtain a time parameter matrix; inputting the time parameter matrix into a pre-trained collision avoidance warning model to obtain multiple warning strategies output by the collision avoidance warning model, and a probability value corresponding to each warning strategy; determining the current target warning strategy for the vehicle from the multiple warning strategies based on the probability value, and issuing a warning prompt to the vehicle according to the target warning strategy. This enables more accurate and rapid prediction of collision events during vehicle operation, thereby providing early warnings to users, preventing collisions between vehicles, improving driving safety, and enhancing user experience.

Description

Technical Field

[0001] This application relates to the field of vehicle safety technology, and in particular to a collision avoidance warning method, computer program product, and vehicle. Background Technology

[0002] With the rapid development of the automotive industry and people's increasing attention to traffic safety, automotive collision avoidance warning technology has become a research hotspot.

[0003] In related technologies, collision avoidance warnings for automobiles typically rely on a single sensor and a simple algorithm to determine whether a collision is imminent. For example, a radar sensor detects the distance between the current vehicle and the vehicle in front, and if this distance exceeds a preset distance threshold, it determines that a collision is likely and issues an alert.

[0004] However, because sensors are susceptible to environmental factors (severe weather, electromagnetic interference, etc.), the above-mentioned method of detecting whether a vehicle has collided with other vehicles using a single sensor and a simple algorithm is prone to inaccurate detection, resulting in missed or false detections. This leads to an inability to provide accurate warnings to users and affects the user experience. Summary of the Invention

[0005] This application provides a collision avoidance warning method, computer program product, and vehicle to solve the technical problem that in the prior art, when a vehicle is detected to be in collision with other vehicles by a single sensor and a simple algorithm, it is easy to cause inaccurate detection, resulting in missed detections or false detections, which leads to the inability to accurately provide warnings to users and affects the user experience.

[0006] In a first aspect, this application provides a collision avoidance warning method for vehicles, the method comprising:

[0007] The system obtains a set of parameter values ​​for each of the multiple operating parameters of a vehicle during a preset time period; the set of parameter values ​​includes the parameter values ​​of the corresponding operating parameter at different time points.

[0008] Preprocessing the multiple sets of parameter values ​​yields a time parameter matrix;

[0009] The time parameter matrix is ​​input into a pre-trained collision avoidance warning model to obtain multiple warning strategies output by the collision avoidance warning model, and the probability value corresponding to each warning strategy.

[0010] Based on the probability value, a target warning strategy for the vehicle is determined from multiple warning strategies, and a warning prompt is issued to the vehicle according to the target warning strategy.

[0011] As an optional implementation, the operating parameters include the operating parameters of the vehicle, the operating parameters of the target vehicle, and the operating parameters between the vehicle and the target vehicle, wherein the target vehicle is the vehicle closest to the vehicle.

[0012] The set of parameter values ​​corresponding to each of the multiple operating parameters obtained from the vehicle's operation within a preset time period includes:

[0013] The speed set, acceleration set, and steering angle set of the vehicle are measured respectively during the preset time period;

[0014] Determine the set of relative distances and relative velocities between the vehicle and the target vehicle, as well as the set of anti-collision accelerations of the target vehicle;

[0015] The set of velocities, the set of accelerations, the set of steering angles, the set of relative distances, the set of relative velocities, and the set of anti-collision accelerations are all determined as the set of parameter values.

[0016] As an optional implementation, the collision avoidance warning model includes an input layer, a convolutional layer, a pooling layer, a fully connected layer, and an output layer; the collision avoidance warning model outputs multiple warning strategies and the probability value corresponding to each warning strategy in the following manner:

[0017] The time parameter matrix is ​​obtained through the input layer, and multiple convolutional features are obtained by performing convolution calculations on the time parameter matrix through the convolutional layer.

[0018] The pooling layer is used to extract features from the multiple convolutional features to obtain a pooling feature matrix;

[0019] The pooling feature matrix is ​​processed by the fully connected layer to obtain vector features;

[0020] The output layer performs probability calculations on the vector features to obtain the probability value corresponding to each of the pre-set multiple early warning strategies, and outputs multiple early warning strategies and the probability value corresponding to each of the early warning strategies.

[0021] As an optional implementation, the convolutional layer includes N convolutional kernels, where N is a positive integer. The convolutional layer performs convolution calculations on the time parameter matrix to obtain multiple convolutional features, including:

[0022] For each convolution kernel, the convolution kernel is slid in the time parameter matrix according to a preset stride.

[0023] For each sliding submatrix of the convolution kernel, a convolution calculation is performed on the submatrix to obtain the sub-features corresponding to the submatrix;

[0024] A matrix composed of multiple sub-features is determined as the convolutional feature after the convolution kernel convolves the time parameter matrix.

[0025] As an optional implementation, the step of performing convolution calculations on the sub-matrix to obtain the sub-features corresponding to the sub-matrix includes:

[0026] Obtain the weight matrix and the first bias term corresponding to the convolution kernel;

[0027] The elements in the submatrix are weighted and summed according to the weight matrix to obtain the initial convolution value;

[0028] The initial convolution value is added to the first bias term to obtain the sub-feature corresponding to the sub-matrix.

[0029] As an optional implementation, the step of extracting features from the multiple convolutional features through the pooling layer to obtain a pooled feature matrix includes:

[0030] Determine the pooling window of the pooling layer;

[0031] For each of the convolutional features, the convolutional features are divided into multiple sub-window matrices according to the pooling window;

[0032] Obtain the maximum value in each of the sub-window matrices, and use the matrix composed of the maximum values ​​as the sub-pooling feature matrix corresponding to the convolutional feature;

[0033] The multiple sub-pooling feature matrices are combined to obtain a pooling feature matrix with the number of channels equal to the number of convolutional features.

[0034] As an optional implementation, the fully connected layer includes M neurons, where M is a positive integer. The process of performing fully connected computation on the pooling feature matrix through the fully connected layer to obtain vector features includes:

[0035] Convert the pooling feature matrix into a one-dimensional vector;

[0036] For each neuron, obtain the first weight vector and the second bias term corresponding to the neuron;

[0037] Based on the first weight vector, the one-dimensional vector is weighted and summed to obtain the initial sub-vector features;

[0038] The initial sub-vector features are added to the second bias term to obtain the sub-vector features;

[0039] The sub-vector features corresponding to each of the M neurons are combined to obtain the vector features.

[0040] As an optional implementation, the output layer includes T neurons, each neuron corresponding to a warning strategy, where T is a positive integer. The output layer performs probability calculations on the vector features to obtain the probability value corresponding to each of a pre-defined plurality of warning strategies, including:

[0041] For each neuron, determine the second weight vector and the third bias term corresponding to that neuron;

[0042] Based on the second weight vector, the vector features are weighted and summed to obtain the first value;

[0043] Add the first value to the third bias term to obtain the second value;

[0044] The probability value corresponding to each of the T second values ​​is obtained by using a preset activation function.

[0045] As an optional implementation, the warning strategy includes normal driving, deceleration, and emergency braking;

[0046] The step of issuing a warning to the vehicle according to the target warning strategy includes:

[0047] When the target warning strategy is slow driving or sudden braking, the warning prompt sound corresponding to the target warning strategy is played through the preset sound playback module.

[0048] And / or,

[0049] When the target warning strategy is slow driving or sudden braking, the warning symbol corresponding to the target warning strategy is displayed through a preset display interface.

[0050] Secondly, this application provides a computer program product that, when run on a computer, causes the computer to execute any of the anti-collision warning methods described in the first aspect above.

[0051] Thirdly, this application provides a vehicle, including a processor and a memory, wherein the processor is configured to execute a collision avoidance warning program for the vehicle stored in the memory to implement the collision avoidance warning method for the vehicle described in any one of the first aspects.

[0052] Fourthly, this application provides a collision avoidance warning device for a vehicle, the device comprising:

[0053] The acquisition module is used to acquire a set of parameter values ​​corresponding to each of the multiple operating parameters of the vehicle during a preset time period; the set of parameter values ​​includes the parameter values ​​of the corresponding operating parameter at different time points.

[0054] The preprocessing module is used to preprocess the multiple sets of parameter values ​​to obtain a time parameter matrix;

[0055] The input module is used to input the time parameter matrix into a pre-trained collision avoidance warning model to obtain multiple warning strategies output by the collision avoidance warning model, and the probability value corresponding to each warning strategy.

[0056] The determination module is used to determine the current target warning strategy for the vehicle from multiple warning strategies based on the probability value, and to issue a warning prompt to the vehicle according to the target warning strategy.

[0057] Fifthly, this application provides a storage medium storing one or more programs that can be executed by one or more processors to implement the vehicle collision avoidance warning method described in any of the first aspects.

[0058] The technical solution provided in this application obtains a set of parameter values ​​for each of the multiple operating parameters of a vehicle traveling within a preset time period. These parameter value sets include the parameter values ​​of the corresponding operating parameters at different time points. The multiple parameter value sets are preprocessed to obtain a time parameter matrix. This time parameter matrix is ​​then input into a pre-trained collision avoidance warning model to obtain multiple warning strategies output by the model, along with a probability value corresponding to each warning strategy. Based on the probability values, the target warning strategy for the vehicle is determined from the multiple warning strategies, and a warning is issued to the vehicle according to the target warning strategy. This technical solution integrates multiple parameter value sets involved in vehicle operation into a time parameter matrix, processes this time parameter matrix using a pre-trained collision avoidance warning model, and determines the warning strategy that the vehicle needs to acquire based on the model's output. Through the training of the collision avoidance warning model and the integration of multiple different parameter value sets, it can quickly and accurately predict collision events, achieving more accurate and rapid prediction of collision events during vehicle operation. This provides users with early warnings, prevents collisions between vehicles, improves driving safety, and enhances the user experience. Attached Figure Description

[0059] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0060] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the accompanying drawings used in the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0061] One or more embodiments are illustrated by way of example with reference numerals in the accompanying drawings. These illustrations do not constitute a limitation on the embodiments. Elements with the same reference numerals in the drawings are denoted as similar elements. Unless otherwise stated, the figures in the drawings are not to be limited by scale.

[0062] Figure 1 A schematic diagram illustrating an application scenario for a vehicle collision avoidance warning method provided in this application embodiment;

[0063] Figure 2 A flowchart illustrating an embodiment of a collision avoidance warning method for a vehicle provided in this application;

[0064] Figure 3 A flowchart illustrating an embodiment of another vehicle collision avoidance warning method provided in this application;

[0065] Figure 4 A flowchart illustrating an embodiment of another collision avoidance warning method for vehicles provided in this application;

[0066] Figure 5 This is a schematic diagram of the structure of a collision avoidance warning model provided in an embodiment of this application;

[0067] Figure 6 A flowchart illustrating another embodiment of a vehicle collision avoidance warning method provided in this application;

[0068] Figure 7 A flowchart illustrating an embodiment of a vehicle collision avoidance warning method provided in this application;

[0069] Figure 8 A block diagram illustrating an embodiment of a vehicle collision avoidance warning device provided in this application;

[0070] Figure 9 This is a structural schematic diagram of a vehicle provided in an embodiment of this application. Detailed Implementation

[0071] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described below are only some, not all, of the embodiments of this application. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0072] The following disclosure provides numerous different embodiments or examples for implementing various structures of this application. To simplify the disclosure, specific examples of components and arrangements are described below. These are merely examples and are not intended to limit the scope of this application. Furthermore, reference numerals and / or letters may be repeated in different examples. Such repetition is for simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or arrangements discussed.

[0073] To address the technical problem that existing technologies using single sensors and simple algorithms to detect potential collisions between vehicles are prone to inaccurate detection, resulting in missed or false detections and failing to provide accurate warnings to users, thus impacting user experience, this application provides a vehicle collision avoidance warning method, computer program product, and vehicle. This method integrates multiple parameter values ​​involved in vehicle operation into a time parameter matrix, processes this matrix using a pre-trained collision avoidance warning model, and determines the warning strategy to be acquired based on the model's output. Through training the collision avoidance warning model and integrating multiple sets of different parameter values, it can quickly and accurately predict collision events, achieving more accurate and rapid prediction of collisions during vehicle operation. This provides users with early warnings, prevents collisions with other vehicles, improves driving safety, and enhances user experience.

[0074] To facilitate understanding of the vehicle collision avoidance warning method provided in this application embodiment, the following will first illustrate the application scenarios involved in the vehicle collision avoidance warning method provided in this application.

[0075] join Figure 1 This is a schematic diagram illustrating an application scenario corresponding to a vehicle collision avoidance warning method provided in an embodiment of this application. For example... Figure 1 As shown, application scenario 10 may include vehicle 11 and vehicle 12.

[0076] During the driving process, vehicle 11 may be located behind or in front of vehicle 12, and this embodiment does not limit this.

[0077] This application does not impose any restrictions on the type, size, model, etc. of vehicles 11 and 12.

[0078] In practical applications, when the driver of vehicle 11 is driving vehicle 11, if there is at least one vehicle, such as vehicle 12, in the vicinity, and if the distance between vehicle 11 and vehicle 12 is close, and if vehicle 11 is traveling too fast or too slow, or has a large turning angle, it is easy to collide with vehicle 12.

[0079] In this regard, in related technologies, the controller of vehicle 11 can detect the distance between itself and vehicle 12 through pre-installed sensors, and when it is determined that the distance is greater than a preset distance threshold, it predicts that vehicle 11 may collide with vehicle 12, and promptly warns the driver of vehicle 11.

[0080] However, because sensors are easily affected by environmental factors (severe weather, electromagnetic interference, etc.), the above methods are prone to inaccurate detection, resulting in missed or false detections. This can lead to an inability to provide accurate warnings to users, safety incidents, and a negative impact on user experience.

[0081] To address the aforementioned technical issues, this application provides a vehicle collision avoidance warning method that can accurately and quickly predict collision events during vehicle operation, thereby providing users with accurate early warnings and improving user experience.

[0082] The collision avoidance warning method for vehicles provided in this application will be further explained and described below with reference to the accompanying drawings and specific embodiments. The embodiments do not constitute a limitation on the embodiments of this application.

[0083] join Figure 2 This is a flowchart illustrating an embodiment of a vehicle collision avoidance warning method provided in this application. As an embodiment, the vehicle collision avoidance warning method provided in this application can be applied to vehicles, for example... Figure 1 The vehicle 11 shown can also be used in vehicle controllers. For example... Figure 2 As shown, the process may include the following steps:

[0084] Step 201: Obtain the set of parameter values ​​corresponding to each operating parameter among multiple operating parameters of the vehicle during a preset time period. The set of parameter values ​​includes the parameter values ​​of the corresponding operating parameter at different time points.

[0085] The aforementioned preset time period refers to a pre-set time period from the current time point, such as within 1 minute or 2 minutes from the current time.

[0086] The aforementioned operating parameters refer to the parameters involved in the vehicle's operation, which may include, but are not limited to: the vehicle's speed, acceleration, steering angle, relative distance and relative speed to other vehicles, and the acceleration of other vehicles.

[0087] The aforementioned set of parameter values ​​refers to the parameter values ​​of the vehicle's operating parameters at different points in time. This set of parameter values ​​may include the same or different parameter values; that is, the operating parameter may correspond to the same parameter value or different parameter values ​​at different points in time. This application embodiment does not impose any restrictions on this. For example, when the operating parameter is vehicle speed, the set of parameter values ​​may include multiple vehicle speed values ​​within a preset time period. These multiple speed values ​​may be the same or different. This application embodiment does not impose any restrictions on this.

[0088] In some embodiments of this application, the executing entity of this application can obtain the parameter value of each operating parameter at different time points from multiple operating parameters of the vehicle driving within a preset time period, and obtain the parameter value set corresponding to each operating parameter.

[0089] As an optional implementation, the execution entity of this application embodiment can obtain a set of multiple parameter values ​​of the vehicle driving within a preset time period in real time.

[0090] As another optional implementation, the execution subject of this application embodiment can periodically obtain a set of multiple parameter values ​​of the vehicle during a preset time period.

[0091] As another optional implementation, the execution entity of this application embodiment can obtain a set of multiple parameter values ​​of the vehicle's travel within a preset time period when other vehicles are detected within a preset range by pre-installed sensors. The preset range can be within a preset distance from the vehicle in different directions.

[0092] As for the specifics of how each running parameter sets its values, this will be explained below. Figure 3 The process shown will be explained in detail here.

[0093] Step 202: Preprocess the multiple parameter value sets to obtain the time parameter matrix.

[0094] The aforementioned preprocessing refers to the preprocessing of the obtained set of parameter values. The processing methods may include, but are not limited to, outlier handling, missing value handling, and normalization.

[0095] The aforementioned time parameter matrix refers to the parameter values ​​of multiple operating parameters at different points in time during the vehicle's operation. In other words, the time parameter matrix can include the correspondence between time points and the parameter values ​​of multiple operating parameters, and it can be a two-dimensional matrix of time points and multiple operating parameters.

[0096] In some embodiments of this application, after obtaining the above-mentioned set of multiple parameter values ​​during vehicle operation, the set of multiple parameter values ​​can be preprocessed to obtain a more accurate correspondence between time points and multiple operating parameters, thereby obtaining a time parameter matrix of the vehicle driving within a preset time period.

[0097] The specific process of preprocessing multiple parameter value sets to obtain the aforementioned time parameter matrix will be explained below. Figure 3 The process shown will be explained in detail here.

[0098] Step 203: Input the above time parameter matrix into the pre-trained collision avoidance warning model to obtain multiple warning strategies output by the above collision avoidance warning model, as well as the probability value corresponding to each warning strategy.

[0099] Step 204: Based on the above probability values, determine the current target warning strategy for the vehicle from among the multiple warning strategies, and issue a warning prompt to the vehicle according to the target warning strategy.

[0100] The following provides a unified explanation of steps 203 and 204:

[0101] The aforementioned collision avoidance warning model refers to a pre-trained model used to predict whether a collision is possible in the current operating state of a vehicle and to provide a response warning strategy.

[0102] The aforementioned warning strategy refers to a warning strategy used to warn the driver of the vehicle based on the vehicle's current operating status. For example, if the current operating status may indicate a collision with other vehicles, the corresponding warning strategy could be to slow down, thereby causing the driver to reduce the vehicle's speed and avoid a collision with other vehicles.

[0103] The aforementioned warning prompt refers to a warning prompt issued to the user. This warning prompt is used to prompt the user to take collision avoidance strategies, such as slowing down or sudden braking. The prompting method may be an audio prompt, a symbol prompt of a preset color output through a visual interface, etc. This application embodiment does not limit this.

[0104] In some embodiments of this application, after determining the time parameter matrix of the vehicle within a preset time period, the executing entity of this application can input the time parameter matrix into a pre-trained collision avoidance warning model. The collision avoidance warning model can predict whether the vehicle will collide with other vehicles in the current operating state based on the received time parameter matrix, and output multiple warning strategies and the probability value corresponding to each warning strategy.

[0105] Then, based on this probability value, the current target warning strategy for the vehicle can be determined from the above multiple warning strategies, and a warning prompt can be issued to the vehicle according to the target warning strategy.

[0106] As an optional implementation approach, the early warning strategy with the highest probability value can be determined as the target early warning strategy.

[0107] As another optional implementation method, each of the above-mentioned early warning strategies can correspond to a priority. Based on this, when there are multiple early warning strategies with the same probability value, the early warning strategy with the highest priority can be determined as the target early warning strategy.

[0108] Finally, the executing entity of this application embodiment can issue a warning to the vehicle according to the above-described target warning strategy.

[0109] As an optional implementation, the aforementioned warning strategy may include, but is not limited to: normal driving, deceleration driving, and emergency braking. Here, normal driving indicates driving at the current speed, deceleration driving indicates reducing the current speed by a preset step size, and emergency braking indicates immediately stopping the vehicle.

[0110] Based on this, in an exemplary embodiment, when the target warning strategy is slow driving or sudden braking, a warning prompt sound corresponding to the target warning strategy can be played through a preset sound playback module (e.g., a speaker, an audio output device), such as "Please slow down!";

[0111] In another exemplary implementation, when the current target warning strategy is slow driving or sudden braking, the warning symbol corresponding to the target warning strategy can be displayed through a preset display interface. For example, the warning symbol corresponding to slow driving is a yellow exclamation mark, and the warning symbol corresponding to sudden braking is a red exclamation mark.

[0112] In another exemplary implementation, when the target warning strategy is slow driving or sudden braking, a warning prompt sound corresponding to the target warning strategy can be played through a preset sound playback module, and a warning prompt symbol corresponding to the target warning strategy can be displayed through a preset display interface.

[0113] The technical solution provided in this application obtains a set of parameter values ​​for each of the multiple operating parameters of a vehicle traveling within a preset time period. These parameter value sets include the parameter values ​​of the corresponding operating parameters at different time points. The multiple parameter value sets are preprocessed to obtain a time parameter matrix. This time parameter matrix is ​​then input into a pre-trained collision avoidance warning model to obtain multiple warning strategies output by the model, along with a probability value corresponding to each warning strategy. Based on the probability values, the target warning strategy for the vehicle is determined from the multiple warning strategies, and a warning is issued to the vehicle according to the target warning strategy. This technical solution integrates multiple parameter value sets involved in vehicle operation into a time parameter matrix, processes this time parameter matrix using a pre-trained collision avoidance warning model, and determines the warning strategy that the vehicle needs to acquire based on the model's output. Through the training of the collision avoidance warning model and the integration of multiple different parameter value sets, it can quickly and accurately predict collision events, achieving more accurate and rapid prediction of collision events during vehicle operation. This provides users with early warnings, prevents collisions between vehicles, improves driving safety, and enhances the user experience.

[0114] See Figure 3 This is a flowchart illustrating an embodiment of another vehicle collision avoidance warning method provided in this application. Figure 3 The process shown is in Figure 2 Based on the illustrated process, this paper describes how, when the aforementioned operating parameters include the vehicle's operating parameters, the target vehicle's operating parameters, and the operating parameters between the vehicle and the target vehicle, the specific steps are taken to obtain the set of parameter values ​​for each operating parameter within a preset time period, and how to preprocess these multiple parameter value sets to obtain a time parameter matrix. For example... Figure 3 As shown, the process may include the following steps:

[0115] Step 301: Measure the set of speeds, the set of accelerations, and the set of steering angles of the vehicle within a preset time period.

[0116] In some embodiments of this application, the vehicle's operating parameters may include the vehicle's speed, acceleration, and steering angle. Based on this, the executing entity of this application can use multiple pre-installed different parameter sensors to measure the vehicle's speed set, acceleration set, and steering angle set over a preset time period. These parameter sensors are sensors pre-installed on the vehicle for measuring its operating parameters, and may include, but are not limited to, wheel speed sensors, acceleration sensors, and steering angle sensors.

[0117] As an optional implementation, the aforementioned parameter sensors may include wheel speed sensors, acceleration sensors, and steering angle sensors.

[0118] The aforementioned wheel speed sensor can be installed on the wheel of the vehicle to collect the wheel speed of the vehicle and determine the wheel speed as the vehicle speed, thereby obtaining the speed set corresponding to a preset time period.

[0119] The aforementioned acceleration sensor can be based on microelectromechanical systems (MEMS) technology. Through an internally installed movable mass block, when the vehicle accelerates or decelerates, the displacement generated by the inertia of the mass block is converted into an electrical signal to determine the magnitude and direction of the vehicle's acceleration, thereby obtaining the set of accelerations of the vehicle within a preset time period.

[0120] The aforementioned steering angle sensor can be installed on the vehicle's steering column to measure the vehicle's steering angle and obtain a set of steering angles for the vehicle within a preset time period.

[0121] Step 302: Determine the set of relative distances and relative velocities between the vehicle and the target vehicle, as well as the set of collision avoidance accelerations of the target vehicle.

[0122] Step 303: Determine the above-mentioned sets of velocity, acceleration, steering angle, relative distance, relative velocity, and anti-collision acceleration as parameter value sets.

[0123] The following provides a unified explanation of steps 302 and 303:

[0124] The aforementioned target vehicle refers to the vehicle that is currently closest to the vehicle (the vehicle where the execution subject of this application embodiment is located), that is, the vehicle that may collide with the vehicle. It can be a vehicle located in front of, behind, or to the sides of the vehicle. This application embodiment does not limit this.

[0125] In some embodiments of this application, the operating parameters may include not only the operating parameters of the vehicle, but also the operating parameters between the vehicle and the target vehicle (such as the relative distance and relative speed between the vehicle and the target vehicle), and the operating parameters of the target vehicle (such as the acceleration of the target vehicle (hereinafter referred to as "anti-collision acceleration")).

[0126] Based on this, the executing entity of this application embodiment can determine the set of relative distances and relative speeds between the vehicle and the target vehicle within a preset time period, as well as the set of anti-collision accelerations of the target vehicle.

[0127] As one implementation method, a set of relative distances, a set of relative speeds, and a set of collision avoidance accelerations of the target vehicle within a preset time period can be determined based on a preset radar sensor. Here, the aforementioned radar sensor refers to a radar sensor pre-installed on the vehicle, which can be a lidar sensor or an infrared radar sensor; this application embodiment does not impose any limitation on this.

[0128] As an optional implementation, the execution entity of this application embodiment can send a laser beam to the target vehicle through a lidar sensor, receive the reflected light from the target vehicle, and calculate the relative distance and relative speed between the vehicle and the target vehicle based on the time difference between the emitted light and the reflected light, thereby obtaining a set of relative distances and a set of relative speeds.

[0129] In the measurement of the acceleration of the target vehicle, the execution subject of this application embodiment can continuously measure the relative distance and speed of the target vehicle using a lidar, and determine the acceleration of the target vehicle based on the speed change and time interval before and after the time interval, thereby obtaining the set of vehicle acceleration within a preset time period.

[0130] Based on steps 301 and 302, the vehicle's speed set, acceleration set, steering angle set, relative distance set between the vehicle and the target vehicle, relative speed set, and anti-collision acceleration set of the target vehicle can be obtained. The execution subject of this application embodiment can determine the above-mentioned speed set, acceleration set, steering angle set, relative distance set between the vehicle and the target vehicle, relative speed set, and anti-collision acceleration set of the target vehicle as the parameter value set of the vehicle within a preset time period.

[0131] Step 304: For each set of parameter values, perform outlier processing, missing value processing, and normalization processing on the parameter values ​​in the above set of parameter values ​​to obtain a new set of parameter values ​​containing multiple new parameter values.

[0132] Step 305: Merge multiple sets of new parameter values ​​to obtain a time parameter matrix. The time parameter matrix includes new parameter values ​​for different operating parameters corresponding to each time point.

[0133] The following provides a unified explanation of steps 304 and 305:

[0134] The outlier handling described above can be achieved using the Z-score analysis method. This method identifies outliers by calculating the standard deviation distance (Z-score) between each data point and the mean. For example, data points with a Z-score greater than 3 or less than -3 are considered outliers and are deleted.

[0135] The missing value handling mentioned above refers to the use of linear interpolation to fill in missing data at a certain point in time. Linear interpolation is a data interpolation method that can be used to estimate unknown data values ​​between known discrete data points.

[0136] The above normalization process can be achieved by using a minimum-to-maximum normalization method to scale the data to the range [0, 1]. This method first determines the minimum value (Xmin) and maximum value (Xmax) of a set of data. For each value X in the dataset, the normalized new value X' can be calculated using the following formula (i), thereby scaling each value in the dataset to the range [0, 1].

[0137] X' = ​​(X - Xmin) / (Xmax - Xmin) (Equation 1)

[0138] Where X' is the new value after normalization, X is the value to be normalized, Xmin is the minimum value in the dataset, and Xmax is the maximum value in the dataset.

[0139] In some embodiments of this application, after obtaining the above-mentioned parameter value set, outlier processing, missing value processing, and normalization processing can be performed on the parameter values ​​in each of the above parameter value sets to obtain a new parameter value set containing multiple new parameter values.

[0140] Then, multiple sets of the above new parameter values ​​can be merged to obtain a time parameter matrix. The time parameter matrix can include new parameter values ​​of different operating parameters corresponding to each time point within a preset time period.

[0141] For example, suppose the new set of parameter values ​​may include the following three: {a1, a2, a3, a4}, {b1, b2, b3, b4}, and {c1, c2, c3, c4}. Further suppose the time points corresponding to these parameter value sets are t1, t2, t3, and t4, respectively. After merging the new set of parameter values, we can obtain the following time parameter matrix:

[0142]

[0143] The technical solution provided in this application embodiment measures the speed, acceleration, and steering angle of a vehicle within a preset time period using multiple pre-installed different parameter sensors. It also determines the relative distance, relative speed, and anti-collision acceleration of the target vehicle between the vehicle and a target vehicle using a pre-installed radar sensor. These speed, acceleration, steering angle, relative distance, relative speed, and anti-collision acceleration sets are all defined as parameter value sets. For each parameter value set, outlier processing, missing value processing, and normalization are performed on the parameter values ​​to obtain a new parameter value set containing multiple new parameter values. These new parameter value sets are then merged to obtain a time parameter matrix, which includes new parameter values ​​for different operating parameters corresponding to each time point. This technical solution avoids the situation where the failure of one sensor would prevent the prediction of collision avoidance events. By performing outlier processing, missing value processing, and normalization processing on multiple parameter value sets, the error of the measured parameter values ​​can be reduced, and the time parameter matrix of the vehicle's travel within a preset time period can be more accurately determined, thereby improving the accuracy of collision avoidance event prediction.

[0144] See Figure 4 This is a flowchart illustrating another embodiment of a vehicle collision avoidance warning method provided in this application. Figure 4 The process shown is in Figure 1 Based on the illustrated process, this paper describes how, in a collision avoidance warning model that includes an input layer, convolutional layer, pooling layer, fully connected layer, and output layer, multiple warning strategies and their corresponding probability values ​​are determined through the collision avoidance warning model. Figure 4 As shown, the process may include the following steps:

[0145] Step 401: Obtain the time parameter matrix through the input layer, and perform convolution calculation on the time parameter matrix through the convolution layer to obtain multiple convolution features.

[0146] In some embodiments of this application, the above-mentioned collision avoidance warning model can be as follows: Figure 5 As shown, see Figure 5 This is a structural schematic diagram of a collision avoidance warning model provided in an embodiment of this application. Figure 5 As shown, the collision avoidance warning model may include an input layer, a convolutional layer, a pooling layer, a fully connected layer, and an output layer.

[0147] Based on this, after obtaining the time parameter matrix, the execution subject of this application embodiment can input the time parameter matrix into the collision avoidance warning model. The collision avoidance warning model can obtain the time parameter matrix through the input layer and perform convolution calculation on the time parameter matrix through the convolution layer to obtain multiple convolution features.

[0148] As an optional implementation, the convolutional layer may include N convolutional kernels, where N is a positive integer; for example, the convolutional layer may include 32 convolutional kernels. Based on this, when performing convolution calculations on the time parameter matrix through the convolutional layer, for each convolutional kernel, the kernel can be slid across the time parameter matrix with a preset stride. The preset stride refers to a pre-defined distance the convolutional kernel slides across the time parameter matrix each time.

[0149] Then, for each sliding submatrix of the convolution kernel, convolution calculation can be performed on that submatrix to obtain the sub-features corresponding to that submatrix. The aforementioned submatrix is ​​the matrix that maps the convolution kernel in the time parameter matrix.

[0150] For example, assuming the convolution kernel is a 3x3 structure, and the time parameter matrix is ​​a 5x5 matrix, and the preset stride is 1, the first submatrix mapped by the convolution kernel in the time parameter matrix is ​​a 3x3 submatrix starting from the top left corner. After moving according to the preset stride, the submatrix corresponding to the first three rows and columns 2 to 4 of the time parameter matrix can be obtained. Similarly, after the convolution kernel has moved horizontally, it can move downwards until it reaches the bottom right corner of the time parameter matrix, thus dividing the time parameter matrix into multiple submatrices.

[0151] As an exemplary implementation, when performing convolution calculations on a submatrix to obtain the corresponding sub-features, the weight matrix and the first bias term corresponding to the convolution kernel can be obtained. Then, the elements in the submatrix can be weighted and summed according to the weight matrix to obtain the initial convolution value. Finally, the initial convolution value can be added to the first bias term to obtain the sub-features corresponding to the submatrix. Here, the weight matrix and the first bias term are values ​​obtained by pre-training the convolutional layer, and the weights in the weight matrix correspond one-to-one with the elements in the submatrix.

[0152] For example, assuming the convolution kernel is a 3x3 matrix, the submatrix obtained through the above convolution kernel is also a 3x3 matrix. Correspondingly, the above weight matrix is ​​also a 3x3 matrix. Based on this, the corresponding sub-features can be obtained for this submatrix using the following equation (ii):

[0153]

[0154] Wherein, y is the aforementioned sub-feature, and w is the aforementioned sub-feature.ij Let x be the weight in the i-th row and j-th column of the weight matrix. ij Let be the element in the i-th row and j-th column of the submatrix, and b is the first bias term mentioned above.

[0155] Finally, the matrix composed of multiple sub-features can be determined as the convolutional feature after the convolution kernel convolves the time parameter matrix.

[0156] Following the steps above, each convolution kernel can obtain a convolutional feature corresponding to a time parameter matrix, thus yielding N convolutional features.

[0157] Step 402: Extract features from multiple convolutional features using pooling layers to obtain a pooling feature matrix.

[0158] The pooling layer described above is used to reduce data dimensionality, decrease computational load, and retain key features. Based on this, feature extraction can be performed on the multiple convolutional features obtained above using the pooling layer to obtain a pooled feature matrix with reduced dimensionality.

[0159] The pooling feature matrix described above is the feature matrix obtained after extracting the main features from multiple convolutional features.

[0160] In this embodiment of the application, in order to extract the main features in the convolutional features, the collision avoidance warning model in this embodiment of the application can extract the convolutional features through a pooling layer after convolving the time parameter matrix to obtain the corresponding pooling feature matrix.

[0161] As an optional implementation, a pooling window for the pooling layer can be determined, and for each convolutional feature, the convolutional feature can be divided into multiple sub-window matrices according to the pooling window. Here, the pooling window is a pre-set identity matrix used for feature extraction; for example, the pooling window can be 2*2. Based on this, the time parameter matrix can be divided according to the pooling window to obtain multiple 2*2 sub-window matrices.

[0162] Next, the maximum value of each sub-window matrix can be obtained, and the matrix formed by these maximum values ​​can be used as the sub-pooling feature matrix corresponding to that convolutional feature. Finally, multiple sub-pooling feature matrices can be combined to obtain a pooling feature matrix with the number of channels equal to the number of convolutional features (e.g., N as mentioned above).

[0163] For example, assuming the pooling window is 2x2 and the convolutional features are 4x4, the pooling window can divide the aforementioned time parameter matrix into four sub-window matrices. For each window matrix, the maximum value can be extracted, resulting in a 2x2 sub-pooling feature matrix. Continuing to assume the number of convolutional features is N, combining the N sub-pooling feature matrices yields a pooling feature matrix of shape 2x2xN.

[0164] Step 403: Perform fully connected calculations on the above pooling feature matrix through a fully connected layer to obtain vector features.

[0165] The fully connected layer described above is used to further integrate the pooling feature matrix output by the pooling layer.

[0166] In some embodiments of this application, the pooling feature matrix output by the pooling layer can be further integrated through a fully connected layer and mapped to the output layer.

[0167] As an optional implementation, the fully connected layer described above may include M neurons, where M is a positive integer, such as 64. Based on this, the pooling feature matrix can be fully computed using M neurons to obtain the corresponding vector features.

[0168] As an example implementation, the above pooling feature matrix can be converted into a one-dimensional vector. For example, if the pooling feature matrix is ​​(H, W, C), where H is the number of rows, W is the number of columns, and C is the number of channels, then converting it into a one-dimensional vector will give a vector of length H*W*C.

[0169] Based on this, the aforementioned one-dimensional vector can be input into the fully connected layer, and for each neuron in the fully connected layer, the first weight vector and the second bias term corresponding to that neuron can be obtained. Then, the one-dimensional vector can be weighted and summed according to the first weight vector to obtain the initial sub-vector features, and the aforementioned sub-vector features can be added to the second bias term to obtain the sub-vector features.

[0170] For example, suppose the above one-dimensional vector is (z1, z2, ..., z...). n The first weight vector mentioned above is (w 11 w 12 ,…,w 1n If the second bias term is b1, then the initial sub-vector features can be calculated using the following formula (iii):

[0171]

[0172] Wherein, d1 is the initial sub-vector feature, and w is the initial sub-vector feature. 1i For the i-th weight vector, the above z 1i Let be the i-th vector element in the one-dimensional vector, and b1 is the second bias term mentioned above.

[0173] Finally, the sub-vector features corresponding to each of the M neurons can be combined to obtain the aforementioned vector features, and these vector features can be fully connected to the neurons in the output layer. For example, the initial sub-vector feature obtained by the first neuron is d1, and the initial sub-vector feature obtained by the Mth neuron is d2. MThen the feature vector can be (d1, d2, ..., d). M ).

[0174] Step 404: Perform probability calculation on the vector features through the output layer to obtain the probability value corresponding to each of the pre-set multiple early warning strategies, and output multiple early warning strategies and the probability value corresponding to each early warning strategy.

[0175] The aforementioned early warning strategies are pre-set early warning strategies that may occur at the output layer.

[0176] In some embodiments of this application, multiple warning strategies can be pre-set in the output layer, and the probability value of each warning strategy corresponding to the current operating condition of the vehicle can be determined through the output layer. The multiple warning strategies corresponding to the current operating condition of the vehicle and the probability value corresponding to each warning strategy can be output.

[0177] As an optional implementation, the output layer may include T neurons, each of which may correspond to a warning strategy, where T is a positive integer.

[0178] Based on this, when calculating the probability of the vector features output by the fully connected layer, a second weight vector and a third bias term can be determined for each neuron. Both the aforementioned weight vector and third bias term are obtained beforehand by training the output layer.

[0179] Then, based on the second weight vector, the vector features can be weighted and summed to obtain the first value, and the first value can be added to the third bias term to obtain the second value.

[0180] For example, suppose the vector features of the output of a fully connected layer are D = (d1, d2, ..., dn). 64 ), the second weight vector of the first neuron in the output layer is k1 = (k 11 k 12 , ..., k 1，64 The second value m1 of the first neuron in the output layer can be obtained by the following equation (iv):

[0181]

[0182] Where m1 is the second value, and k is the second value. 1i For the i-th weight vector in the weight vector, the above d i Let b2 be the i-th vector feature in the vector features, and b2 is the third bias term mentioned above.

[0183] Finally, a preset activation function can be used to calculate the probability value corresponding to each of the T second values, which is then the probability value of the warning strategy corresponding to the second value.

[0184] As an exemplary implementation, the probability value of the early warning strategy corresponding to each second value can be determined by the following formula (v):

[0185]

[0186] The above p j Let mj be the probability value corresponding to the j-th neuron, mn be the second value corresponding to the j-th neuron, and mn be the probability value corresponding to the n-th neuron. In this equation (v), the total number of neurons is 3 for example.

[0187] The technical solution provided in this application, in a collision avoidance warning model comprising an input layer, convolutional layer, pooling layer, fully connected layer, and output layer, obtains a time parameter matrix through the input layer, performs convolution calculations on the time parameter matrix through the convolutional layer to obtain multiple convolutional features, extracts features from the multiple convolutional features through the pooling layer to obtain a pooled feature matrix, performs fully connected calculations on the pooled feature matrix through the fully connected layer to obtain vector features, and performs probability calculations on the vector features through the output layer to obtain the probability value corresponding to each of the pre-set multiple warning strategies, and outputs multiple warning strategies and the probability value corresponding to each warning strategy. This technical solution, through operations such as convolution, feature extraction, feature integration, feature full connection, and probability value calculation of each warning strategy on the time parameter matrix of the collision avoidance warning model, can more quickly and accurately output multiple warning strategies for the vehicle under the current operating conditions, as well as the probability value corresponding to each warning strategy.

[0188] See Figure 6 This is a flowchart illustrating another embodiment of a vehicle collision avoidance warning method provided in this application. Figure 6 The process shown is in Figure 4 Based on the illustrated process, the specific training method for the collision avoidance warning model is described. For example... Figure 6 As shown, the process may include the following steps:

[0189] Step 601: Obtain a predetermined training sample set, which includes multiple time parameter matrix samples and a probability value sample set corresponding to each time parameter matrix sample. The probability value sample set includes the probability value corresponding to each of the multiple preset early warning strategies.

[0190] Step 602: Select a target sample set with a preset amount of data from the above training sample set.

[0191] Step 603: Train the initial model using the above target sample set to obtain the initial probability value set corresponding to each time parameter matrix sample output by the initial model. The above initial probability value set includes the initial probability value corresponding to each of the preset multiple early warning strategies.

[0192] The following provides a unified explanation of steps 601 to 603:

[0193] The aforementioned time parameter matrix sample refers to the correspondence between the parameter values ​​and time of multiple operating parameters of a vehicle traveling within a preset time period.

[0194] The aforementioned probability value sample set consists of a predetermined set of multiple warning strategies that a vehicle may potentially encounter, corresponding to the aforementioned time parameter matrix sample, and the probability value corresponding to each warning strategy.

[0195] In some embodiments of this application, a training sample set for training a collision avoidance warning model can be obtained in advance. Then, in order to improve the training efficiency of the model, a target sample set with a preset amount of data (e.g., 64) can be selected from the above training sample set, and the initial model can be trained using the target sample set to obtain the initial probability value set corresponding to each time parameter matrix sample output by the initial model. The above initial probability value set includes the initial probability value corresponding to each of the preset multiple warning strategies.

[0196] As an optional implementation, the execution entity of this application embodiment can obtain the training sample set input by the user through a preset visual interface.

[0197] As another optional implementation, the execution entity of this application embodiment can obtain a pre-stored training sample set from a preset storage medium.

[0198] In addition, the initial model can also be trained directly using the acquired training sample set, and this application embodiment does not limit this.

[0199] The specific details of how the target sample set is used to train the initial model will be explained below. Figure 7 The process shown will be explained in detail here.

[0200] Step 604: Based on the probability value sample set and the corresponding initial probability value set, determine whether the initial model meets the preset stopping training condition. If yes, proceed to step 605; otherwise, proceed to step 606.

[0201] Step 605: Determine the initial model as the collision avoidance warning model.

[0202] Step 606: Optimize the parameters in the initial model using a preset optimization algorithm according to a preset step size to obtain a new initial model.

[0203] Step 607: Determine the new initial model as the initial model and return to step 603.

[0204] The following provides a unified explanation of steps 604 to 607:

[0205] The aforementioned conditions for stopping training refer to the pre-set conditions that meet the requirements for stopping training. These conditions may include, but are not limited to: the number of training iterations reaching a preset value (e.g., 200), and the accuracy, recall, F1 score (the average of accuracy and recall) of the initial probability value set output by the initial model exceeding a preset threshold.

[0206] In some embodiments of this application, after training the initial model using the target sample set to obtain the initial probability value set corresponding to each time parameter matrix sample output by the initial model, it can be determined whether the initial model meets the preset training stop condition based on the probability value sample set corresponding to the target sample set and the corresponding initial probability value set. That is, it can be determined whether the initial model has met the preset training stop condition based on the probability value sample set and the initial probability value set corresponding to the same time parameter matrix sample.

[0207] As an optional implementation, it can be determined whether the current number of training iterations is greater than or equal to a preset threshold. Optionally, if it is determined that the number of training iterations is greater than the threshold, it can be determined that the initial model meets a preset stopping training condition.

[0208] As an alternative implementation, the current accuracy, recall, and F1 score of the initial model can be determined based on the aforementioned probability value sample set and the initial probability value set. Optionally, if the accuracy, recall, and F1 score are all greater than their respective preset thresholds, the initial model is determined to meet the preset stopping training condition.

[0209] Optionally, if it is determined that the preset stop training conditions are met, the initial model can be determined to have completed training, and the initial model can be used to determine the collision avoidance warning model obtained from the training.

[0210] Optionally, if it is determined that the preset training stop condition is not met, a preset optimization algorithm can be used to optimize the parameters in the initial model (e.g., the weight parameters of each layer) according to a preset step size to obtain a new optimized initial model. Then, the new initial model can be used as the initial model to execute step 603 above, that is, to retrain the new initial model using the target sample set. The optimization algorithm can be the Adam optimization algorithm or other algorithms; this embodiment does not limit this. The preset step size can be a pre-set parameter adjustment step size, such as 0.001, or other values; this embodiment does not limit this.

[0211] The technical solution provided in this application involves obtaining a predetermined training sample set, selecting a target sample set with a predetermined amount of data from the training sample set, and training an initial model using the target sample set. This yields a set of initial probability values ​​corresponding to each time parameter matrix sample output by the initial model. The initial probability value set includes the initial probability value corresponding to each of a plurality of predetermined warning strategies. Based on the probability value sample set and the corresponding initial probability value set, it is determined whether the initial model meets a predetermined stopping training condition. If yes, the initial model is determined as a collision avoidance warning model; otherwise, a predetermined optimization algorithm is used to optimize the parameters in the initial model according to a predetermined step size to obtain a new initial model. This new initial model is then determined as the initial model and trained again using the target sample set. This technical solution, by pre-setting the amount of sample data for the training model and continuously optimizing the initial model using a predetermined optimization algorithm during the training process based on the target sample set, and adjusting the parameters of the initial model according to a predetermined step size, achieves more efficient and accurate training of the collision avoidance warning model.

[0212] See Figure 7 The following is a flowchart illustrating an embodiment of a vehicle collision avoidance warning method provided in this application. Figure 7 The process shown is in Figure 6 Based on the illustrated process, this section describes how the initial model is trained using the target sample set, assuming the initial model includes an initial input layer, initial convolutional layer, initial pooling layer, initial fully connected layer, and initial output layer. For example... Figure 7 As shown, the process may include the following steps:

[0213] Step 701: For each time parameter matrix sample in the target sample set, obtain the time parameter matrix sample through the initial input layer, and perform convolution calculation on the time parameter matrix sample through the initial convolution layer to obtain multiple initial convolution features.

[0214] In some embodiments of this application, the initial model may include an initial input layer, an initial convolutional layer, an initial pooling layer, an initial fully connected layer, and an initial output layer.

[0215] Based on this, when the execution subject of this application training the initial model using the time parameter matrix samples in the target sample set, the time parameter matrix samples can be input into the initial model. The initial model can obtain the time parameter matrix samples through the initial input layer and perform convolution calculation on the time parameter matrix samples through the initial convolution layer to obtain multiple initial convolution features.

[0216] As an optional implementation, the initial convolutional layer may include N initial convolutional kernels, where N is a positive integer; for example, the initial convolutional layer may include 32 initial convolutional kernels. Based on this, when performing convolution calculations on the temporal parameter matrix samples through the initial convolutional layer, for each initial convolutional kernel, the kernel can be slid across the temporal parameter matrix samples according to a preset stride. The preset stride refers to the distance the initial convolutional kernel slides across the temporal parameter matrix samples in each iteration.

[0217] Then, for each sliding sub-initial matrix of the initial convolution kernel, convolution calculation can be performed on the sub-initial matrix to obtain the sub-initial features corresponding to the sub-initial matrix. The aforementioned sub-initial matrix is ​​the matrix mapped by the initial convolution kernel in the temporal parameter matrix samples.

[0218] For example, assuming the initial convolutional kernel is a 3x3 structure, and the temporal parameter matrix sample is a 5x5 matrix, and the preset stride is 1, then the sub-initial matrix first mapped by the initial convolutional kernel in the temporal parameter matrix sample is the 3x3 sub-initial matrix starting from the top left corner of the temporal parameter matrix sample. After moving according to the preset stride, the sub-initial matrices corresponding to the first three rows and the 2nd to 4th columns of the temporal parameter matrix sample can be obtained. Similarly, after the initial convolutional kernel has moved horizontally, it can move downwards until it reaches the bottom right corner of the temporal parameter matrix sample, thereby dividing the temporal parameter matrix sample into multiple sub-initial matrices.

[0219] As an exemplary implementation, when performing convolution calculations on the sub-initial matrix to obtain the corresponding sub-initial features, the initial weight matrix and the first initial bias term corresponding to the initial convolution kernel can be obtained. Then, the elements in the sub-initial matrix can be weighted and summed according to the initial weight matrix to obtain the initial convolution value. Finally, the initial convolution value can be added to the first initial bias term to obtain the sub-initial features corresponding to the sub-initial matrix. Here, the initial weight matrix and the first initial bias term are values ​​obtained by pre-training the initial convolutional layer, and the weights in the initial weight matrix correspond one-to-one with the elements in the sub-initial matrix.

[0220] Finally, the matrix composed of multiple initial sub-features can be determined as the initial convolutional feature after the initial convolution kernel convolves the temporal parameter matrix samples.

[0221] Following the steps above, each initial convolution kernel can obtain an initial convolution feature corresponding to a time parameter matrix sample, thus yielding N initial convolution features.

[0222] Step 702: Extract features from multiple initial convolutional features through the initial pooling layer to obtain the initial pooling feature matrix.

[0223] The initial pooling layer described above is used to reduce data dimensionality, decrease computational load, and retain key features. Based on this, feature extraction can be performed on the multiple initial convolutional features obtained above through the initial pooling layer to obtain the initial pooling feature matrix after dimensionality reduction.

[0224] The initial pooling feature matrix mentioned above is the feature matrix after extracting the main features from multiple initial convolutional features.

[0225] In this embodiment of the application, in order to extract the main features in the initial convolution features, the initial model in this embodiment of the application can extract the initial convolution features through the initial pooling layer after convolving the time parameter matrix samples to obtain the corresponding initial pooling feature matrix.

[0226] As an optional implementation, an initial pooling window for the initial pooling layer can be determined, and for each initial convolutional feature, the initial convolutional feature can be divided into multiple sub-initial window matrices according to the initial pooling window. Here, the initial pooling window is a pre-set identity matrix used for feature extraction; for example, the initial pooling window can be 2*2. Based on this, the temporal parameter matrix samples can be divided according to the initial pooling window to obtain multiple 2*2 sub-initial window matrices.

[0227] Next, the maximum value of each sub-initial window matrix can be obtained, and the matrix formed by these maximum values ​​can be used as the sub-initial pooling feature matrix corresponding to that initial convolutional feature. Finally, multiple sub-initial pooling feature matrices can be combined to obtain an initial pooling feature matrix with the number of channels equal to the number of initial convolutional features (e.g., N as mentioned above).

[0228] For example, assuming the initial pooling window is 2x2 and the initial convolutional features are 4x4, this initial pooling window can divide the aforementioned temporal parameter matrix samples into four sub-initial window matrices. For each initial window matrix, the maximum value can be extracted, ultimately resulting in a 2x2 sub-initial pooling feature matrix. Continuing to assume the number of initial convolutional features is N, combining the N sub-initial pooling feature matrices yields an initial pooling feature matrix of shape 2x2xN.

[0229] Step 703: Perform fully connected calculations on the initial pooling feature matrix through an initial fully connected layer to obtain the initial vector features.

[0230] The aforementioned initial fully connected layer is used to further integrate the initial pooling feature matrix output by the initial pooling layer.

[0231] In some embodiments of this application, the initial pooling feature matrix output by the initial pooling layer can be further integrated through the initial fully connected layer and mapped to the initial output layer.

[0232] As an optional implementation, the initial fully connected layer can include M initial neurons, where M is a positive integer, such as 64. Based on this, the initial pooling feature matrix can be fully connected using the M initial neurons to obtain the corresponding initial vector features.

[0233] As an example implementation, the above initial pooling feature matrix can be converted into an initial one-dimensional vector. For example, if the above initial pooling feature matrix is ​​(H, W, C), where H is the number of rows, W is the number of columns, and C is the number of channels, then converting it into an initial one-dimensional vector will yield an initial vector of length H*W*C.

[0234] Based on this, the aforementioned initial one-dimensional vector can be input into the initial fully connected layer, and for each initial neuron in the initial fully connected layer, the first initial weight vector and the second initial bias term corresponding to that initial neuron can be obtained. Then, based on the first initial weight vector, the initial one-dimensional vector can be weighted and summed to obtain the first initial sub-vector feature, and the aforementioned first sub-vector feature can be added to the second initial bias term to obtain the initial sub-vector feature.

[0235] Finally, the initial sub-vector features corresponding to each of the M initial neurons can be combined to obtain the initial vector features mentioned above, and these initial vector features can be fully connected to the initial neurons in the initial output layer. For example, the initial sub-vector feature obtained by the first initial neuron is d1, and the initial sub-vector feature obtained by the Mth initial neuron is d2. M Then the initial vector features can be (d1, d2, ..., d... M ).

[0236] Step 704: Perform probability calculation on the initial vector features through the initial output layer to obtain the initial probability value corresponding to each of the pre-set multiple early warning strategies, and output multiple early warning strategies and the initial probability value corresponding to each early warning strategy.

[0237] The aforementioned early warning strategies are possible early warning strategies that are pre-set in the initial output layer.

[0238] In some embodiments of this application, multiple warning strategies can be pre-set in the initial output layer, and the initial probability value of each warning strategy corresponding to the current operating condition of the vehicle can be determined through the initial output layer. The multiple warning strategies corresponding to the current operating condition of the vehicle and the initial probability value corresponding to each warning strategy can be output.

[0239] As an optional implementation, the initial output layer may include T initial neurons, each of which may correspond to a warning strategy, where T is a positive integer.

[0240] Based on this, when calculating the probability of the initial vector features output by the initial fully connected layer, a second initial weight vector and a third initial bias term can be determined for each initial neuron. Both the second initial weight vector and the third initial bias term are obtained by pre-training the initial output layer.

[0241] Then, based on the second initial weight vector, the features of the initial vector can be weighted and summed to obtain the first initial value, and the first initial value can be added to the third initial bias term to obtain the second initial value.

[0242] Finally, a preset activation function can be used to calculate the T second initial values ​​to obtain the initial probability value corresponding to each second initial value. This initial probability value is the initial probability value of the warning strategy corresponding to the second initial value.

[0243] As an exemplary implementation, the total value of all current second initial values ​​can be determined, and then each second initial value can be divided by the total value to obtain the initial probability value corresponding to each second initial value.

[0244] The technical solution provided in this application, in a collision avoidance warning model comprising an initial input layer, an initial convolutional layer, an initial pooling layer, an initial fully connected layer, and an initial output layer, obtains a time parameter matrix sample through the initial input layer, performs convolution calculations on the time parameter matrix sample through the initial convolutional layer to obtain multiple initial convolutional features, extracts features from the multiple initial convolutional features through the initial pooling layer to obtain an initial pooling feature matrix, performs fully connected calculations on the aforementioned initial pooling feature matrix through the initial fully connected layer to obtain initial vector features, and performs probability calculations on the initial vector features through the initial output layer to obtain the initial probability value corresponding to each of the pre-set multiple warning strategies, and outputs multiple warning strategies and the initial probability value corresponding to each warning strategy. This technical solution, through operations such as convolution, feature extraction, feature integration, feature full connection, and probability value calculation for each warning strategy on the time parameter matrix sample by the initial model, can train the initial model more quickly and accurately, thereby obtaining a more accurate collision avoidance warning model.

[0245] See Figure 8 This is a block diagram illustrating an embodiment of a vehicle collision avoidance warning device provided in this application. Figure 8 As shown, the device may include:

[0246] The acquisition module 81 is used to acquire a set of parameter values ​​corresponding to each of the multiple operating parameters of the vehicle during a preset time period; the set of parameter values ​​includes the parameter values ​​of the corresponding operating parameter at different time points;

[0247] Preprocessing module 82 is used to preprocess the multiple sets of parameter values ​​to obtain a time parameter matrix;

[0248] The input module 83 is used to input the time parameter matrix into a pre-trained collision avoidance warning model to obtain multiple warning strategies output by the collision avoidance warning model, and the probability value corresponding to each warning strategy.

[0249] The determination module 84 is used to determine the current target warning strategy for the vehicle from multiple warning strategies based on the probability value, and to issue a warning prompt to the vehicle according to the target warning strategy.

[0250] like Figure 9 The diagram shown is a structural schematic of a vehicle according to an embodiment of this application, including a processor 91, a communication interface 92, a memory 93, and a communication bus 94. The processor 91, communication interface 92, and memory 93 communicate with each other via the communication bus 94.

[0251] Memory 93 is used to store computer programs;

[0252] In one embodiment of this application, when the processor 91 executes the program stored in the memory 93, it implements the vehicle collision avoidance warning method provided in any of the foregoing method embodiments, including:

[0253] The system obtains a set of parameter values ​​for each of the multiple operating parameters of a vehicle during a preset time period; the set of parameter values ​​includes the parameter values ​​of the corresponding operating parameter at different time points.

[0254] Preprocessing the multiple sets of parameter values ​​yields a time parameter matrix;

[0255] The time parameter matrix is ​​input into a pre-trained collision avoidance warning model to obtain multiple warning strategies output by the collision avoidance warning model, and the probability value corresponding to each warning strategy.

[0256] Based on the probability value, a target warning strategy for the vehicle is determined from multiple warning strategies, and a warning prompt is issued to the vehicle according to the target warning strategy.

[0257] This application also provides a computer program product that, when run on a computer, causes the computer to execute any of the above-described anti-collision warning methods.

[0258] This application also provides a storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the vehicle collision avoidance warning method provided in any of the foregoing method embodiments.

[0259] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0260] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented using software and a general-purpose hardware platform, or of course, using hardware. Based on this understanding, the above technical solutions, in essence or the parts that contribute to the related technology, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0261] It should be understood that the terminology used herein is for the purpose of illustrating specific embodiments of the text only and is not intended to be limiting. Unless the context clearly indicates otherwise, the singular forms “a,” “an,” and “described” as used herein may also include the plural forms. The terms “comprising,” “including,” “containing,” and “having” are inclusive and therefore indicate the presence of the stated features, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, elements, components, and / or combinations thereof. The method steps, processes, and operations described herein are not construed as requiring them to be performed in the specific order described or illustrated unless the order is explicitly indicated. It should also be understood that additional or alternative steps may be used.

[0262] The above description is merely a specific embodiment of this application, enabling those skilled in the art to understand or implement this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features claimed herein.

Claims

1. A collision avoidance warning method for vehicles, characterized in that, The method includes: The system obtains a set of parameter values ​​for each of the multiple operating parameters of a vehicle during a preset time period; the set of parameter values ​​includes the parameter values ​​of the corresponding operating parameter at different time points. Preprocessing the multiple sets of parameter values ​​yields a time parameter matrix; The time parameter matrix is ​​input into a pre-trained collision avoidance warning model to obtain multiple warning strategies output by the collision avoidance warning model, as well as the probability value corresponding to each warning strategy. Based on the probability value, determine the current target warning strategy for the vehicle from multiple warning strategies, and issue a warning prompt to the vehicle according to the target warning strategy; The collision avoidance warning model includes an input layer, a convolutional layer, a pooling layer, a fully connected layer, and an output layer. The model outputs multiple warning strategies and a probability value corresponding to each strategy in the following manner: The input layer obtains the time parameter matrix, and the convolutional layer performs convolution calculations on the time parameter matrix to obtain multiple convolutional features; the pooling layer extracts features from the multiple convolutional features to obtain a pooled feature matrix; the fully connected layer performs fully connected calculations on the pooled feature matrix to obtain vector features; the output layer performs probability calculations on the vector features to obtain the probability value corresponding to each of the pre-set multiple warning strategies, and outputs the multiple warning strategies and the probability value corresponding to each strategy. The output layer comprises T neurons, each corresponding to a warning strategy, where T is a positive integer. The step of calculating the probability of the vector features through the output layer to obtain the probability value corresponding to each of the pre-defined warning strategies includes: for each neuron, determining a second weight vector and a third bias term; performing a weighted summation of the vector features based on the second weight vector to obtain a first value; adding the first value to the third bias term to obtain a second value; and using a preset activation function to calculate the probability value corresponding to each of the T second values.

2. The method according to claim 1, characterized in that, The operating parameters include the operating parameters of the vehicle, the operating parameters of the target vehicle, and the operating parameters between the vehicle and the target vehicle, wherein the target vehicle is the vehicle closest to the vehicle. The set of parameter values ​​corresponding to each of the multiple operating parameters obtained from the vehicle's operation within a preset time period includes: The speed set, acceleration set, and steering angle set of the vehicle are measured respectively during the preset time period; Determine the set of relative distances and relative velocities between the vehicle and the target vehicle, as well as the set of anti-collision accelerations of the target vehicle; The set of velocities, the set of accelerations, the set of steering angles, the set of relative distances, the set of relative velocities, and the set of anti-collision accelerations are all determined as the set of parameter values.

3. The method according to claim 1, characterized in that, The convolutional layer includes N convolutional kernels, where N is a positive integer. The convolutional layer performs convolution calculations on the time parameter matrix to obtain multiple convolutional features, including: For each convolution kernel, the convolution kernel is slid in the time parameter matrix according to a preset stride. For each sliding submatrix of the convolution kernel, a convolution calculation is performed on the submatrix to obtain the sub-features corresponding to the submatrix; A matrix composed of multiple sub-features is determined as the convolutional feature after the convolution kernel convolves the time parameter matrix.

4. The method according to claim 3, characterized in that, The step of performing convolution calculations on the sub-matrix to obtain the sub-features corresponding to the sub-matrix includes: Obtain the weight matrix and the first bias term corresponding to the convolution kernel; The elements in the submatrix are weighted and summed according to the weight matrix to obtain the initial convolution value; The initial convolution value is added to the first bias term to obtain the sub-feature corresponding to the sub-matrix.

5. The method according to claim 1, characterized in that, The step of extracting features from the multiple convolutional features through the pooling layer to obtain a pooled feature matrix includes: Determine the pooling window of the pooling layer; For each of the convolutional features, the convolutional features are divided into multiple sub-window matrices according to the pooling window; Obtain the maximum value in each of the sub-window matrices, and use the matrix composed of the maximum values ​​as the sub-pooling feature matrix corresponding to the convolutional feature; The multiple sub-pooling feature matrices are combined to obtain a pooling feature matrix with the number of channels equal to the number of convolutional features.

6. The method according to claim 1, characterized in that, The fully connected layer comprises M neurons, where M is a positive integer. The process of performing fully connected computation on the pooling feature matrix through the fully connected layer to obtain vector features includes: Convert the pooling feature matrix into a one-dimensional vector; For each neuron, obtain the first weight vector and the second bias term corresponding to the neuron; Based on the first weight vector, the one-dimensional vector is weighted and summed to obtain the initial sub-vector features; The initial sub-vector features are added to the second bias term to obtain the sub-vector features; The sub-vector features corresponding to each of the M neurons are combined to obtain the vector features.

7. A computer program product, characterized in that, When it is run on a computer, it causes the computer to perform the anti-collision warning method according to any one of claims 1 to 6.

8. A vehicle, characterized in that, include: A processor and a memory, the processor being configured to execute a vehicle collision avoidance warning program stored in the memory to implement the vehicle collision avoidance warning method according to any one of claims 1 to 6.

Images