Vehicle collision detection method, medium and vehicle

Abstract

The invention provides a vehicle collision detection method, a medium and a vehicle, and relates to the technical field of vehicle collision detection. The method comprises the following steps: firstly, based on various state data of a target vehicle, obtaining standard time sequence data corresponding to each type of state data, and carrying out Grubby angle field conversion on the standard time sequence data to obtain a two-dimensional image corresponding to each type of state data, so as to convert one-dimensional vehicle time sequence data into the two-dimensional image, thereby retaining a time sequence relationship of original data; and the data integrity is improved. And then based on the two-dimensional images corresponding to the various state data, a multi-dimensional feature matrix corresponding to the target vehicle is generated so as to capture correlation between different time sequence data, and the state of the vehicle is reflected more comprehensively. And finally, inputting the multi-dimensional feature matrix into a preset collision detection model for collision prediction to obtain a collision detection result of the target vehicle so as to improve the accuracy of vehicle collision identification.

Description

Technical Field

[0001] The present application relates to the technical field of vehicle collision detection, and in particular to a vehicle collision detection method, medium, and vehicle. Background Art

[0002] Vehicle collision recognition has always been a key research area for all automotive OEMs and suppliers. Current vehicle collision recognition methods mostly rely on adding sensors to the vehicle, collecting sensor data through sensor modules, and then using an onboard central processing unit to perform vehicle collision recognition on this data. More specifically, these methods utilize methods such as fixed thresholds for collision recognition. However, rule thresholds are often set based on expert experience, which can easily lead to bias and false positives. Additionally, some methods rely on computer vision algorithms to identify vehicle collisions, using external cameras to capture real-time driving images for collision prediction. However, the quality of these images is significantly affected by weather and environmental factors, making it difficult to guarantee the accuracy of collision detection results. Summary of the Invention

[0003] The present application provides a vehicle collision detection method, medium and vehicle, which can improve the accuracy of vehicle collision identification.

[0004] A first aspect of an embodiment of the present application provides a vehicle collision detection method, the method comprising: Based on multiple state data of the target vehicle, obtain standard time series data corresponding to each state data; Performing Gram angle field conversion on the standard time series data to obtain a two-dimensional image corresponding to each state data; generating a multidimensional feature matrix corresponding to the target vehicle based on the two-dimensional images corresponding to the plurality of state data; The multi-dimensional feature matrix is ​​input into a preset collision detection model to perform collision prediction, and a collision detection result of the target vehicle is obtained.

[0005] Optionally, generating a multidimensional feature matrix corresponding to the target vehicle based on the two-dimensional images corresponding to the plurality of state data includes: The two-dimensional images corresponding to all the state data are stacked to obtain a multi-dimensional feature matrix corresponding to the target vehicle.

[0006] Optionally, based on multiple state data of the target vehicle, standard time series data corresponding to each state data is obtained, including: For any of the state data, dividing the state data into a plurality of time segments; For any of the time segments, performing standardization processing on each sampling point in the time segment to obtain standardized data corresponding to each sampling point; Based on the standardized data corresponding to all the sampling points, standard time series data corresponding to the state data is generated.

[0007] Optionally, performing Gram angle field conversion on the standard time series data to obtain a two-dimensional image corresponding to each state data includes: For the standard time series data corresponding to any one of the state data, performing polar coordinate conversion on each sampling point in the standard time series data to obtain an angle value of each sampling point in polar coordinates; Based on the angle values ​​corresponding to the sampling points, a two-dimensional image corresponding to the state data is obtained.

[0008] Optionally, obtaining a two-dimensional image corresponding to the state data based on the angle value corresponding to the sampling point includes: Determining a correlation value between any two of the sampling points based on the angle values ​​corresponding to the sampling points; the correlation value is used to reflect the degree of temporal similarity between the two sampling points; generating a Gram matrix corresponding to the standard time series data based on the correlation values ​​between the sampling points; Based on the Gram matrix corresponding to the standard time series data, a two-dimensional image corresponding to the state data is obtained.

[0009] Optionally, the preset collision detection model includes a convolution module, a block attention module, a flattening module and a classification module; Inputting the multidimensional feature matrix into a preset collision detection model to perform collision prediction, and obtaining a collision detection result of the target vehicle, including: Inputting the multidimensional feature matrix into the convolution module for feature extraction to obtain a first feature map; Performing feature adjustment on the first feature map based on the block attention weights by the block attention module to obtain a second feature map; Performing dimensionality reduction on the second feature map by the flattening module to obtain a one-dimensional feature vector; The one-dimensional feature vector is subjected to collision prediction by the classification module to obtain a collision detection result of the target vehicle.

[0010] Optionally, the convolution module includes at least one extraction layer, and any of the extraction layers includes a first convolution layer and a second convolution layer, the first convolution layer and the second convolution layer have different convolution kernel sizes and the same number of channels; Inputting the multidimensional feature matrix into the convolution module for feature extraction to obtain a first feature map, including: For any of the extraction layers, perform feature extraction, dimensionality reduction, and activation operations on the input features of the extraction layer through the first convolutional layer and the second convolutional layer, respectively, to obtain an initial feature map corresponding to the first convolutional layer and an intermediate feature map corresponding to the second convolutional layer; Perform channel fusion on the initial feature map and the intermediate feature map to obtain output features corresponding to the extraction layer; The input feature of the first extraction layer is the multidimensional feature matrix, and the output feature of the last extraction layer is the first feature map.

[0011] Optionally, performing feature adjustment on the first feature map based on the block attention weight to obtain a second feature map includes: Performing region division on the first feature map to obtain a plurality of initial region sub-maps of the same size; Determining a block attention weight of each of the initial region sub-graphs, and performing feature weighting on the initial region sub-graphs based on the block attention weights to obtain a region feature graph; The plurality of regional feature maps are spliced ​​together to obtain the second feature map.

[0012] Based on the same inventive concept, a second aspect of an embodiment of the present application provides a vehicle collision detection device, the device comprising: A time series data acquisition module is used to obtain standard time series data corresponding to each state data based on multiple state data of the target vehicle; A time series data conversion module is used to perform Gram angle field conversion on the standard time series data to obtain a two-dimensional image corresponding to each state data; A multi-dimensional feature construction module, configured to generate a multi-dimensional feature matrix corresponding to the target vehicle based on the two-dimensional images corresponding to the plurality of state data; The collision prediction module is used to input the multi-dimensional feature matrix into a preset collision detection model to perform collision prediction and obtain a collision detection result of the target vehicle.

[0013] Based on the same inventive concept, the third aspect of the embodiment of the present application provides a readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the vehicle collision detection method proposed in the first aspect of the present application is implemented.

[0014] Based on the same inventive concept, the fourth aspect of an embodiment of the present application provides a vehicle, comprising a processor and a memory; the memory stores machine-executable instructions that can be executed by the processor, and the processor is used to execute the machine-executable instructions to implement the vehicle collision detection method proposed in the first aspect of the present application.

[0015] Compared with the prior art, this application has the following advantages: The present application provides a vehicle collision detection method. First, based on multiple state data of a target vehicle, standard time series data corresponding to each state data is obtained. Gram angle field conversion is performed on the standard time series data to obtain a two-dimensional image corresponding to each state data. Then, based on the two-dimensional images corresponding to the multiple state data, a multidimensional feature matrix corresponding to the target vehicle is generated. Finally, the multidimensional feature matrix is ​​input into a preset collision detection model for collision prediction, thereby obtaining a collision detection result for the target vehicle. Thus, through Gram angle field conversion, the one-dimensional vehicle time series data is converted into a two-dimensional image to preserve the temporal relationship of the original data and improve data integrity. A multidimensional feature matrix for the target vehicle is constructed based on the two-dimensional images converted from the multiple time series data to capture the correlation between different time series data and more comprehensively reflect the vehicle's state. Finally, the multidimensional feature matrix is ​​processed by the collision detection model to improve the accuracy of vehicle collision recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments of the present invention. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0017] Figure 1 This is a flow chart of a vehicle collision detection method in one embodiment of the present application; Figure 2 This is a schematic diagram of the structure of a collision detection model in one embodiment of the present application; Figure 3 This is a schematic diagram of the functional modules of a vehicle collision detection device according to an embodiment of the present application; Figure 4 It is a structural schematic diagram of a vehicle in one embodiment of the present application. DETAILED DESCRIPTION

[0018] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0019] Please refer to Figure 1 , Figure 1 This is a flow chart of a vehicle collision detection method proposed in one embodiment of the present application. Figure 1As shown, the method includes the following steps: S101: Based on multiple state data of the target vehicle, obtain standard time series data corresponding to each state data.

[0020] In this embodiment, the target vehicle refers to a vehicle that requires collision detection, and the status data includes vehicle data such as vehicle speed data, acceleration data, yaw rate data, accelerator pedal position data, low-voltage battery voltage data, and battery pack relay status data.

[0021] In specific implementation, vehicle speed data can be obtained through sensors such as speed sensors, and the vehicle's acceleration data, yaw rate data, etc. can be calculated based on this; accelerator pedal position data can be obtained through sensors such as accelerator pedal position sensors, low-voltage battery voltage data can be obtained through sensors such as voltage sensors, and battery pack relay status data can be obtained through relay status sensors.

[0022] In some embodiments, the time interval for collecting vehicle status data is 2 seconds. In other embodiments, the time interval for collecting vehicle status data is 5 seconds. The time interval for collecting vehicle status data is not limited thereto and may vary depending on factors such as the performance of onboard sensors and the speed at which vehicle data is transmitted.

[0023] In this embodiment, standard time series data refers to high-quality data obtained after preprocessing the state data. For example, data cleaning is performed on the state data to delete or fill missing data, or the state data is normalized using a sliding window to smooth the data and reduce data noise.

[0024] After obtaining various status data of the target vehicle, this embodiment preprocesses the status data to obtain standard time series data corresponding to each status data, so as to improve the data quality and facilitate subsequent analysis.

[0025] S102: Perform Gram angle field conversion on the standard time series data to obtain a two-dimensional image corresponding to each state data.

[0026] The Gram angular field transform maps time series data into a high-dimensional space and uses angle information to capture the correlation between data.

[0027] In this embodiment, for any type of status data, by performing Gram angle field conversion on its standard time series data, a two-dimensional matrix describing the data change trend can be generated, thereby converting the one-dimensional vehicle time series data into a two-dimensional image to retain the time series relationship of the original data and converting it into frequency domain features. It can more intuitively represent the status characteristics of the target vehicle in a graphical manner, which is conducive to the subsequent use of convolutional neural network models to better extract the status characteristics of the target vehicle and promote the detection of vehicle collision situations.

[0028] Exemplarily, after performing Gram angle field conversion on standard time series data containing N (eg, 50) sampling points, an N×N two-dimensional image will be obtained.

[0029] S103: Generate a multi-dimensional feature matrix corresponding to the target vehicle based on the two-dimensional images corresponding to the various state data.

[0030] In this embodiment, after obtaining two-dimensional images corresponding to multiple state data, these two-dimensional images can be combined to obtain a multi-dimensional feature tensor representation, that is, a multi-dimensional feature matrix, to retain the correlation between different state data and more comprehensively reflect the state of the vehicle.

[0031] Furthermore, by stacking the two-dimensional images corresponding to all state data, a multi-dimensional feature matrix corresponding to the target vehicle is obtained.

[0032] In this embodiment, two-dimensional images corresponding to status data such as vehicle speed data, acceleration data, yaw angular velocity data, accelerator pedal position data, and low-voltage battery voltage data are stacked according to status categories to generate a multidimensional array, that is, a multidimensional feature matrix that reflects the various status characteristics of the target vehicle is obtained. The multidimensional feature matrix can be used for subsequent data analysis, pattern recognition, or machine learning tasks to detect vehicle collision situations.

[0033] For example, assuming there are C types (for example, 30 types) of state data, and each type of state data generates an N×N two-dimensional image, the shape of the stacked multi-dimensional feature matrix is: C×N×N.

[0034] S104: Inputting the multi-dimensional feature matrix into a preset collision detection model to perform collision prediction and obtain a collision detection result of the target vehicle.

[0035] In this embodiment, the preset collision detection model is built based on a convolutional neural network and is obtained through collision detection training using the historical status data of the target vehicle. The historical status data includes historical data of the target vehicle in different historical time periods when it has and has not collided.

[0036] By extracting and processing the multi-dimensional feature matrix constructed above through a pre-set collision detection model, deeper features of the vehicle status data can be learned, thereby improving the accuracy of vehicle collision detection results. The collision detection results include collision and non-collision.

[0037] In addition, when a collision is detected, the collision results will be shared with the vehicle after-sales service provider, so that vehicle repair, rescue, insurance and other services can be obtained in a timely manner, improving the user's car experience.

[0038] Optionally, based on multiple state data of the target vehicle, standard time series data corresponding to each state data is obtained, including: S201: For any type of state data, divide the state data into multiple time segments.

[0039] In this implementation, a sliding window approach can be used to obtain multiple time slices for each type of state data. Specifically, a fixed-size window is defined and slid across the state data (time series) to extract statistical features of the state data within different time series windows. In other words, a window represents a time slice. Furthermore, the size of the sliding window can be dynamically adjusted based on actual needs. This allows the selection of the most appropriate window size when processing state data with different characteristics, balancing processing efficiency and result accuracy.

[0040] For example, assuming that the length of the state data is M and the window size is w, the sliding window will sequentially calculate the 1st to wth data, the 2nd to w+1th data of the state data, until the last window.

[0041] S202: For any time segment, perform standardization processing on each sampling point in the time segment to obtain standardized data corresponding to each sampling point.

[0042] In this embodiment, for any time segment, each sampling point in the time segment is normalized using the following method to obtain the normalized data corresponding to each sampling point:

[0043] in, Is a vector representing the normalized signal value of the nth window of the i-th state data; Is a vector representing the original signal value of the nth window of the i-th state data; is the mean value calculated based on all historical data of the i-th state data; is the standard deviation calculated based on all historical data of the i-th state data.

[0044] Therefore, within each sliding window, that is, within any time segment, by performing standardization on the signal data within the window, the data can be smoothed, the noise can be reduced, and higher quality sampling data can be obtained.

[0045] S203: Generate standard time series data corresponding to the state data based on the standardized data corresponding to all sampling points.

[0046] In this embodiment, after the standardized data corresponding to all sampling points are arranged in chronological order, the standard time series data corresponding to the state data can be obtained, thereby providing more complete and accurate data for subsequent vehicle collision detection, making the collision detection results more accurate.

[0047] Optionally, Gram angle field conversion is performed on the standard time series data to obtain a two-dimensional image corresponding to each state data, including: S301: For the standard time series data corresponding to any state data, perform polar coordinate conversion on each sampling point in the standard time series data to obtain the angle value of each sampling point in the polar coordinate.

[0048] In this embodiment, the signal data in the standard time series data is first normalized, and the signal value of each sampling point is mapped to The normalized standard time series data is then converted to a polar coordinate system to obtain the angle value of each sampling point in the polar coordinate system.

[0049] Specifically, assuming that the signal value of the t-th sampling point of the i-th state data is normalized , the angle value of the t-th sampling point in the polar coordinate system can be calculated by the following formula:

[0050] in, Represents the normalized signal value of the t-th sampling point, Indicates the polar coordinate angle corresponding to the t-th sampling point.

[0051] S302: Obtain a two-dimensional image corresponding to the state data based on the angle value corresponding to the sampling point.

[0052] In this embodiment, based on the angle value corresponding to each sampling point, a Gram matrix is ​​constructed by calculating the angle between different sampling points to generate a two-dimensional image corresponding to the state data, thereby converting the one-dimensional time series data into a two-dimensional image and retaining the time information of the data.

[0053] Specifically, the process of forming a two-dimensional image mainly includes: S302-1: Determine a correlation value between any two sampling points based on the angle values ​​corresponding to the sampling points.

[0054] In this embodiment, the correlation value is used to reflect the similarity between two sampling points in time. In specific implementation, the correlation value between two sampling points can be calculated as follows:

[0055] in, Indicates the The polar coordinate angle corresponding to the sampling point, Indicates the The polar coordinate angle corresponding to the sampling point, express Sampling point and Correlation value between sampling points. The bigger, the Sampling point and The greater the correlation between the sampling points, the more similar the two signal values ​​are.

[0056] S302-2: Generate a Gram matrix corresponding to the standard time series data based on the correlation values ​​between the sampling points.

[0057] In this embodiment, the Gram matrix G corresponding to the standard time series data is as follows:

[0058] Here, n represents the number of sampling points contained in a certain state data (or standard time series data). The above formula also shows that the Gram matrix preserves the temporal dependence of the time series data. As it moves from the upper left corner to the lower right corner, the number of sampling points increases with time. Therefore, time is reflected in the geometric structure of the matrix.

[0059] S302-3: Based on the Gram matrix corresponding to the standard time series data, obtain a two-dimensional image corresponding to the state data.

[0060] In this embodiment, the value of each element in the Gram matrix is ​​used as the pixel value of the image, so as to generate a two-dimensional image corresponding to the state data.

[0061] This embodiment converts the one-dimensional time series data into a two-dimensional image by performing a Gram angle field conversion on the standard time series data. This not only preserves the temporal information of the data, but also, because pixel values ​​at different locations in the image correspond to values ​​at different moments in the time series data, allows for intuitive observation of the changing trends of the vehicle's status data over time. Furthermore, converting complex one-dimensional data into concise and easy-to-understand two-dimensional images not only simplifies the data processing process but also facilitates subsequent feature extraction using image analysis methods. For example, convolutional neural networks can be used to extract features from two-dimensional images and identify patterns in time series data.

[0062] Please refer to Figure 2 , Figure 2 This is a schematic diagram of the structure of the collision detection model in one embodiment of the present application. Figure 2 As shown in Figure 3, the preset collision detection model includes a convolution module, a block attention module, a flattening module, and a classification module.

[0063] Input the multi-dimensional feature matrix into the preset collision detection model to perform collision prediction and obtain the collision detection results of the target vehicle, including: S401: Input the multidimensional feature matrix into the convolution module for feature extraction to obtain a first feature map.

[0064] In this embodiment, the convolution module includes at least one extraction layer. Figure 2 As shown in the figure, when the convolution module includes two extraction layers, the size of the output first feature map is controlled by gradually performing convolution operations on the multidimensional feature matrix, reducing the spatial dimensions (height and width) of the multidimensional feature matrix and increasing the depth (number of channels). For example, the number of output channels of the first extraction layer is expanded to 64, and the number of output channels of the second extraction layer is expanded to 128, so as to reduce the computational complexity of subsequent modules.

[0065] Specifically, any extraction layer includes a first convolutional layer and a second convolutional layer, and the convolution kernel sizes of the first convolutional layer and the second convolutional layer are different, but the number of channels is the same.

[0066] For example, the convolution kernel size of the first convolutional layer is 3×3 with 64 channels, while the convolution kernel size of the second convolutional layer is 7×7 with 64 channels. Small convolution kernels (such as 3×3) can extract local image details such as edges and corners, while large convolution kernels (such as 7×7) can integrate information from a wider range and extract more abstract and global features such as texture. Thus, convolution kernels of different sizes capture features of different scales in the multidimensional feature matrix, facilitating vehicle collision detection.

[0067] Furthermore, for any extraction layer, the input features of the extraction layer are subjected to feature extraction, dimensionality reduction, and activation operations through the first and second convolutional layers, respectively, to obtain the initial feature map corresponding to the first convolutional layer and the intermediate feature map corresponding to the second convolutional layer. The initial feature map and the intermediate feature map are then channel-fused to obtain the output features corresponding to the extraction layer. The input features of the first extraction layer are a multidimensional feature matrix, and the output features of the last extraction layer are the first feature map.

[0068] like Figure 2As shown in the figure, in the first extraction layer, the convolution kernel size of the first convolution layer is 3×3 and the number of channels is 64. The first convolution layer extracts the initial features of the multidimensional feature matrix and obtains the initial feature map. At the same time, the convolution kernel size of the second convolution layer is 1×1 and the number of channels is 64. The second convolution layer processes the multidimensional feature matrix, while maintaining the original features of the multidimensional feature matrix and adjusting the number of channels of the multidimensional feature matrix to 64, thus obtaining the intermediate feature map.

[0069] Then, by fusing the initial feature map after preliminary feature extraction with the intermediate feature map that retains the original features of the multi-dimensional feature matrix on the channel, the input is directly added to the output of the convolutional layer to form a residual connection, which makes it easier for the network to transmit gradient information during back propagation, greatly reducing the attenuation of the gradient during the propagation process and avoiding gradient disappearance.

[0070] After extracting features by convolving the multidimensional feature matrix with convolution kernels of varying sizes, the first and second convolutional layers further perform pooling and activation operations. The purpose of pooling is to reduce the spatial size of the data and computational complexity while preserving important feature information. For example, suppose the first and second convolutional layers produce a 64×N×N feature map after convolution. After pooling, its size is reduced to 64×(N / 2)×(N / 2), thereby reducing the feature dimensionality. The purpose of activation is to introduce nonlinear factors and enhance the network's nonlinear capabilities, enabling the network to learn more complex feature representations and improving the model's generalization ability.

[0071] The feature map output by the first extraction layer continues to be further processed as the input feature of the second extraction layer, and finally outputs the first feature map (the final output of the entire convolution module).

[0072] The processing of the second extraction layer is the same as that of the first extraction layer. For example, if the first extraction layer outputs a feature map of size 64 × (N / 2) × (N / 2), after processing it in the second extraction layer, the resulting feature map will be 128 × (N / 4) × (N / 4).

[0073] S402: Using the block attention module, feature adjustment is performed on the first feature map based on the block attention weight to obtain a second feature map.

[0074] In this implementation, after the convolutional module of the neural network, the feature map typically contains rich information in both spatial and channel dimensions. However, not every region within this information is equally important for the vehicle collision detection task. Therefore, this embodiment calculates the block attention weights for each region and adjusts the features of the first feature map. This allows the collision detection model to focus more on the most useful regions in the first feature map, thereby improving model performance.

[0075] Specifically, this step mainly includes: S402-1: Divide the first feature map into regions to obtain a plurality of initial region sub-maps of the same size.

[0076] In this implementation, it is assumed that the size of the first feature map output by the convolution module is C×H×W, where C represents the number of channels of the first feature map, H represents the height of the first feature map (i.e., the spatial dimension), and W represents the width of the first feature map (i.e., the spatial dimension).

[0077] When C=64, H=32, W=32, and it is divided into regions of size 4×4, 8×8=64 4×4 regions will be obtained, thereby dividing the first feature map into 64 initial region sub-maps of the same size.

[0078] S402-2: Determine the block attention weight of each initial region sub-image, and perform feature weighting on the initial region sub-image based on the block attention weight to obtain a region feature map.

[0079] In this implementation, a block-wise attention weight is calculated for each initial region subgraph, allowing the collision detection model to automatically focus on the most important regions. In this step, the attention mechanism generates a weighted coefficient for each region, which represents the importance of the region.

[0080] Specifically, for each initial region sub-graph, the block attention weight corresponding to the region can be calculated as follows:

[0081] in, Indicates the first feature map The block attention weights of the initial region sub-graphs, Indicates the first feature map The initial features of the initial region subgraph.

[0082] In the above formula, the denominator is a normalization operation for all regions, ensuring that the sum of the weights of all initial region sub-images is 1. Then, by calculating the feature information of each small area, the block attention weight corresponding to each initial region sub-image is generated. The larger the block attention weight, the more important the area.

[0083] After obtaining the block attention weights corresponding to each initial region sub-image, these weights can be used to adjust the initial region sub-image to highlight more important features. Specifically, feature adjustment can be performed in the following ways:

[0084] in, Indicates the first feature map The weighted features of the initial region subgraphs, Indicates the The block attention weights of the initial region sub-graphs, Indicates the The weighted features of the initial region subgraphs.

[0085] This embodiment uses the block attention weight to perform a weighted operation on the initial region sub-graph, multiplying each initial region sub-graph by its corresponding block attention weight, thereby amplifying or reducing the features of the region, so that the weighted region feature map can more prominently highlight those small regions that are useful for the collision detection task.

[0086] S402-3: Concatenate multiple regional feature maps to obtain a second feature map.

[0087] In this implementation, after applying attention weighting to all small regions, these weighted regional feature maps are reassembled into a new feature map to obtain a second feature map. This second feature map contains the structural information of the original feature map, and the weighted regions are enhanced, which can better help the subsequent network dynamically focus on important areas, reduce interference from irrelevant areas, and thus improve the accuracy of collision detection results.

[0088] S403: Reduce the dimension of the second feature map through a flattening module to obtain a one-dimensional feature vector.

[0089] In this embodiment, after the multi-dimensional feature matrix of the target vehicle is extracted through the above-mentioned convolution module and block attention module to obtain the second feature map, the second feature map is further input into the flattening module for dimensionality reduction processing to obtain a one-dimensional feature vector.

[0090] Specifically, such as Figure 2 As shown, the flattening module includes a flattening layer and a fully connected layer. The purpose of the flattening layer is to flatten the second feature map into a one-dimensional vector. For example, the second feature map of size 128×(N / 4)×(N / 4) will become 128×(N / 4) after being flattened by the flattening layer. 2 / 16). The purpose of the fully connected layer is to further reduce the dimensionality of the feature map, for example, limiting the output dimension of the feature map to 512. Thus, through the flattening layer and the fully connected layer, the complex feature space is mapped to a concise probability distribution, allowing the subsequent classifier to perform accurate classification tasks.

[0091] S404: Performing collision prediction on the one-dimensional feature vector through a classification module to obtain a collision detection result of the target vehicle.

[0092] In this embodiment, the one-dimensional feature vector output by the flattening module is further input into the classification module for collision prediction to obtain a collision detection result of the target vehicle.

[0093] like Figure 2 As shown in Figure 1, the classification module mainly includes the Softmax classification layer. The Softmax classification layer uses the Softmax function for classification, maps the network output to the probability value of whether a collision occurs or not, and finally outputs the judgment result: collision or non-collision.

[0094] Please refer to Figure 3 Based on the same inventive concept, a second aspect of an embodiment of the present application provides a vehicle collision detection device, the vehicle collision detection device 300 comprising: The time series data acquisition module 301 is used to obtain the standard time series data corresponding to each state data based on the multiple state data of the target vehicle; The time series data conversion module 302 is used to perform Gram angle field conversion on the standard time series data to obtain a two-dimensional image corresponding to each state data; A multi-dimensional feature construction module 303 is used to generate a multi-dimensional feature matrix corresponding to the target vehicle based on the two-dimensional images corresponding to the multiple state data; The collision prediction module 304 is used to input the multi-dimensional feature matrix into a preset collision detection model to perform collision prediction and obtain a collision detection result of the target vehicle.

[0095] Optionally, the multi-dimensional feature construction module 303 is specifically configured to: The two-dimensional images corresponding to all state data are stacked to obtain the multi-dimensional feature matrix corresponding to the target vehicle.

[0096] Optionally, the time series data acquisition module 301 includes: The time segmentation submodule is used to divide the state data into multiple time segments for any state data; The preprocessing submodule is used to standardize each sampling point in any time segment to obtain the standardized data corresponding to each sampling point; based on the standardized data corresponding to all sampling points, generate standard time series data corresponding to the state data.

[0097] Optionally, the time series data conversion module 302 includes: The Gram angle field conversion submodule is used to perform polar coordinate conversion on each sampling point in the standard time series data corresponding to any state data to obtain the angle value of each sampling point in the polar coordinates; The Gram matrix calculation submodule is used to obtain a two-dimensional image corresponding to the state data based on the angle values ​​corresponding to the sampling points.

[0098] Optionally, the Gram matrix calculation submodule is specifically used to: Determine the correlation value between any two sampling points based on the angle values ​​corresponding to the sampling points; the correlation value is used to reflect the degree of temporal similarity between the two sampling points; Based on the correlation values ​​between sampling points, a Gram matrix corresponding to the standard time series data is generated; Based on the Gram matrix corresponding to the standard time series data, a two-dimensional image corresponding to the state data is obtained.

[0099] Optionally, the preset collision detection model includes a convolution module, a block attention module, a flattening module and a classification module; The collision prediction module 304 includes: The extraction submodule is used to input the multidimensional feature matrix into the convolution module for feature extraction to obtain a first feature map; an adjustment submodule, configured to perform feature adjustment on the first feature map based on the block attention weights through the block attention module to obtain a second feature map; A dimensionality reduction submodule, configured to reduce the dimensionality of the second feature map by using a flattening module to obtain a one-dimensional feature vector; The classification submodule is used to perform collision prediction on the one-dimensional feature vector through the classification module to obtain the collision detection result of the target vehicle.

[0100] Optionally, the convolution module includes at least one extraction layer, and any extraction layer includes a first convolution layer and a second convolution layer, and the first convolution layer and the second convolution layer have different convolution kernel sizes and the same number of channels; The extraction submodule is specifically used for: For any extraction layer, feature extraction, dimensionality reduction, and activation operations are performed on the input features of the extraction layer through the first convolutional layer and the second convolutional layer respectively, to obtain the initial feature map corresponding to the first convolutional layer and the intermediate feature map corresponding to the second convolutional layer; Perform channel fusion on the initial feature map and the intermediate feature map to obtain the output features corresponding to the extraction layer; Among them, the input feature of the first extraction layer is the multi-dimensional feature matrix, and the output feature of the last extraction layer is the first feature map.

[0101] Optionally, the adjustment submodule includes: A region division unit, configured to divide the first feature map into regions to obtain a plurality of initial region sub-maps of the same size; An attention adjustment unit is used to determine the block attention weight of each initial region sub-image, and perform feature weighting on the initial region sub-image based on the block attention weight to obtain a regional feature map; The splicing unit is used to splice multiple regional feature maps to obtain a second feature map.

[0102] As for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0103] In a third aspect, based on the same inventive concept, an embodiment of the present application provides a readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the vehicle collision detection method proposed in the first aspect of the present application is implemented.

[0104] It should be noted that the specific implementation of the storage medium in the embodiment of the present application refers to the specific implementation of the vehicle collision detection method proposed in the first aspect of the embodiment of the present application, and will not be repeated here.

[0105] In the fourth aspect, based on the same inventive concept, refer to Figure 4 An embodiment of the present application provides a vehicle 400, including a processor 401 and a memory 402; the memory 402 stores machine-executable instructions that can be executed by the processor 401, and the processor 401 is used to execute the machine-executable instructions to implement the vehicle collision detection method proposed in the first aspect of the present application.

[0106] It should be noted that the specific implementation of the vehicle 400 in the embodiment of the present application refers to the specific implementation of the vehicle collision detection method proposed in the first aspect of the embodiment of the present application, and will not be repeated here.

[0107] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0108] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, devices, or computer program products. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0109] The embodiments of the present application are described with reference to the flowcharts and / or block diagrams of the methods, terminal devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, 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 processor, or other programmable data processing terminal device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing terminal device generate instructions for implementing the steps in the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0110] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing terminal device to operate in a specific manner, so that the instructions stored in the computer readable memory produce a manufactured product including an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0111] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal device so that a series of operating steps are executed on the computer or other programmable terminal device to produce a computer-implemented process, thereby providing instructions for executing on the computer or other programmable terminal device to implement the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0112] Although preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they become aware of the basic inventive concepts. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the embodiments of the present invention.

[0113] Finally, it should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that includes a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or terminal device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article, or terminal device that includes the element.

[0114] The above is a detailed introduction to the vehicle collision detection method, medium and vehicle provided by the present application. Specific examples are used in this article to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core ideas. At the same time, for those skilled in the art, according to the ideas of the present application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present application.

Claims

1. A vehicle collision detection method, characterized in that: The method comprises: Based on multiple state data of the target vehicle, obtain standard time series data corresponding to each state data; Performing Gram angle field conversion on the standard time series data to obtain a two-dimensional image corresponding to each state data; generating a multidimensional feature matrix corresponding to the target vehicle based on the two-dimensional images corresponding to the plurality of state data; The multi-dimensional feature matrix is ​​input into a preset collision detection model to perform collision prediction, and a collision detection result of the target vehicle is obtained.

2. The method according to claim 1, characterized in that Based on the two-dimensional images corresponding to the plurality of state data, a multi-dimensional feature matrix corresponding to the target vehicle is generated, including: The two-dimensional images corresponding to all the state data are stacked to obtain a multi-dimensional feature matrix corresponding to the target vehicle.

3. The method according to claim 1, characterized in that Based on the various state data of the target vehicle, standard time series data corresponding to each state data is obtained, including: For any of the state data, dividing the state data into a plurality of time segments; For any of the time segments, performing standardization processing on each sampling point in the time segment to obtain standardized data corresponding to each sampling point; Based on the standardized data corresponding to all the sampling points, standard time series data corresponding to the state data is generated.

4. The method according to claim 1, wherein Performing Gram angle field conversion on the standard time series data to obtain a two-dimensional image corresponding to each state data includes: For the standard time series data corresponding to any one of the state data, performing polar coordinate conversion on each sampling point in the standard time series data to obtain an angle value of each sampling point in polar coordinates; Based on the angle values ​​corresponding to the sampling points, a two-dimensional image corresponding to the state data is obtained.

5. The method according to claim 4, characterized in that Obtaining a two-dimensional image corresponding to the state data based on the angle value corresponding to the sampling point includes: Determining a correlation value between any two of the sampling points based on the angle values ​​corresponding to the sampling points; the correlation value is used to reflect the degree of temporal similarity between the two sampling points; generating a Gram matrix corresponding to the standard time series data based on the correlation values ​​between the sampling points; Based on the Gram matrix corresponding to the standard time series data, a two-dimensional image corresponding to the state data is obtained.

6. The method according to any one of claims 1 to 5, characterized in that The preset collision detection model includes a convolution module, a block attention module, a flattening module and a classification module; Inputting the multidimensional feature matrix into a preset collision detection model to perform collision prediction, and obtaining a collision detection result of the target vehicle, including: Inputting the multidimensional feature matrix into the convolution module for feature extraction to obtain a first feature map; Performing feature adjustment on the first feature map based on the block attention weights by the block attention module to obtain a second feature map; Performing dimensionality reduction on the second feature map by the flattening module to obtain a one-dimensional feature vector; The one-dimensional feature vector is subjected to collision prediction by the classification module to obtain a collision detection result of the target vehicle.

7. The method according to claim 6, characterized in that The convolution module includes at least one extraction layer, and any of the extraction layers includes a first convolution layer and a second convolution layer, wherein the first convolution layer and the second convolution layer have different convolution kernel sizes and the same number of channels; Inputting the multidimensional feature matrix into the convolution module for feature extraction to obtain a first feature map, including: For any of the extraction layers, perform feature extraction, dimensionality reduction, and activation operations on the input features of the extraction layer through the first convolutional layer and the second convolutional layer, respectively, to obtain an initial feature map corresponding to the first convolutional layer and an intermediate feature map corresponding to the second convolutional layer; Perform channel fusion on the initial feature map and the intermediate feature map to obtain output features corresponding to the extraction layer; The input feature of the first extraction layer is the multidimensional feature matrix, and the output feature of the last extraction layer is the first feature map.

8. The method according to claim 6, characterized in that Performing feature adjustment on the first feature map based on the block attention weight to obtain a second feature map, including: Dividing the first feature map into regions to obtain a plurality of initial region sub-maps of the same size; Determining a block attention weight of each of the initial region sub-graphs, and performing feature weighting on the initial region sub-graphs based on the block attention weights to obtain a region feature graph; The plurality of regional feature maps are spliced ​​together to obtain the second feature map.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the vehicle collision detection method according to any one of claims 1 to 8 is implemented.

10. A vehicle, characterized in that: It comprises a processor and a memory; the memory stores machine-executable instructions that can be executed by the processor, and the processor is used to execute the machine-executable instructions to implement the vehicle collision detection method according to any one of claims 1 to 8.

Images