Driving style recognition method, system, device and storage medium

Abstract

The present invention relates to the field of intelligent driving data processing technology, specifically a driving style identification method, system, device, and storage medium. One driving style identification method comprises: obtaining raw driving data; extracting driving style features from the raw driving data according to a first preset rule; extracting time-series statistical features from the raw driving data according to a second preset rule; fusing the driving style features with the time-series statistical features to obtain a hybrid feature; and inputting the hybrid feature into a driving style classification model, which processes the hybrid feature using a dynamic routing mechanism and outputs a driving style classification result. The present invention can comprehensively reflect a driver's behavioral patterns and improve the accuracy of the driving style classification results.

Description

Technical Field

[0001] The present invention relates to the field of intelligent driving data processing technology, and in particular to a driving style recognition method, system, device and storage medium. Background Art

[0002] Current driving style recognition technology primarily relies on time series data collected by vehicle sensors (such as accelerometers, throttle position sensors, and brake signal sensors). Typical technical solutions typically follow the following processing flow:

[0003] First, the original time series data is denoised and segmented according to a fixed time window (for example, a 40-second time window). Statistical features reflecting driving style, such as sudden acceleration frequency and average braking intensity, are then extracted from the segmented data. To eliminate feature redundancy, principal component analysis (PCA) is used for dimensionality reduction. Furthermore, a clustering algorithm (such as K-means) is used to generate pseudo-labels of driving styles, classifying driving behaviors into categories such as aggressive, stable, and conservative. Finally, based on the pseudo-labels, a fully connected neural network (Multi-Layer Perceptron, MLP) or a time series model (LSTM / TCN) is used for supervised training and classification recognition. LSTM stands for Long Short-Term Memory, and TCN stands for Temporal Convolutional Network.

[0004] Although existing solutions have achieved basic recognition functions, there are still significant technical bottlenecks:

[0005] One-sided feature expression: Traditional methods rely on a single feature source—aggregated statistical features (such as mean and extreme values)—which ignore the dynamic temporal changes of driving behavior. Directly modeling raw time series data is susceptible to noise and has low computational efficiency.

[0006] Insufficient time series modeling capabilities: Recursive models such as LSTM suffer from the vanishing gradient problem when processing long sequences, while convolutional models such as TCN have difficulty capturing global driving pattern correlations due to limited local receptive fields.

[0007] Rigid feature fusion mechanism: Existing technologies use feature concatenation or weighted summation methods, which cannot adaptively learn the complementary relationship between aggregated features and time series statistical features, nor can they distinguish the contribution weights of different feature dimensions to classification results.

[0008] The model has weak generalization ability: In complex driving scenarios (such as urban congestion and sudden lane changes on highways), the recognition accuracy is significantly reduced due to insufficient feature representation. The measured F1 score is generally lower than 90%, which seriously restricts its practical application value.

[0009] The above defects collectively make it difficult for existing technologies to meet the core requirements of intelligent driving systems for accurate and robust driving style recognition. Summary of the Invention

[0010] The present application aims to at least solve the technical problems existing in the prior art and provide a driving style recognition method, system, device and storage medium.

[0011] In a first aspect, the present invention provides a driving style recognition method, comprising:

[0012] Obtaining raw driving data;

[0013] extracting driving style features of the original driving data according to a first preset rule;

[0014] extracting time series statistical features of the original driving data according to a second preset rule;

[0015] Fusion of driving style features and time series statistical features to obtain hybrid features;

[0016] The mixed features are input into the driving style classification model, and the driving style classification model processes the mixed features using a dynamic routing mechanism and outputs a driving style classification result.

[0017] In a second aspect, the present invention provides a driving style recognition system, the system comprising:

[0018] Acquisition module, used to obtain raw driving data;

[0019] a first feature extraction module, configured to extract driving style features of the original driving data according to a first preset rule;

[0020] A second feature extraction module is used to extract time series statistical features of the original driving data according to a second preset rule;

[0021] Fusion module, used to fuse driving style features and time series statistical features to obtain hybrid features;

[0022] The output module is used to input the mixed features into the driving style classification model. The driving style classification model uses a dynamic routing mechanism to process the mixed features and output a driving style classification result.

[0023] In a third aspect, the present invention provides an electronic device, comprising:

[0024] at least one processor; and,

[0025] a memory communicatively connected to the at least one processor; wherein,

[0026] The memory stores a computer program executable by the at least one processor. The computer program is executed by the at least one processor to enable the at least one processor to perform the driving style recognition method described above.

[0027] In a fourth aspect, the present invention further provides a computer-readable storage medium, wherein at least one computer program is stored in the computer-readable storage medium, and the at least one computer program is executed by a processor in an electronic device to implement the driving style recognition method described above.

[0028] In summary, this application has the following beneficial technical effects:

[0029] Driving style features and time-series statistical features are extracted from raw driving data. This approach simultaneously captures global trends in driver operation (e.g., features derived from throttle, acceleration, and braking signals) and local dynamic change information (e.g., statistical indicators such as mean, standard deviation, median, and quantile). This fusion strategy enables the extracted hybrid features to more comprehensively reflect driver behavior patterns, providing richer information for subsequent driving style classification model recognition and improving the accuracy of driving style classification results.

[0030] The encoder extracts a global representation of the time series statistical features, which is then concatenated with the aggregated features. The concatenated joint representation is then fed into the capsule network module. The feature vectors are hierarchically modeled and weightedly fused through a dynamic routing mechanism, enabling adaptive learning of the importance of each feature in the final classification task, significantly improving the discriminative ability and robustness of driving style recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 A schematic flow chart of a driving style identification method provided by one embodiment of the present invention;

[0032] Figure 2 An illustrative diagram of an embodiment of a driving style identification method provided by another embodiment of the present invention;

[0033] Figure 3 A schematic diagram of the structure of an electronic device for implementing the driving style recognition method provided by one embodiment of the present invention.

[0034] Reference numerals: 10, processor; 11, memory; 12, communication bus; 13, communication interface.

[0035] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings and in conjunction with the embodiments. DETAILED DESCRIPTION

[0036] The following describes embodiments of the present invention in detail. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended only to explain the present invention and are not to be construed as limiting the present invention.

[0037] In the description of the present invention, it should be understood that the terms "longitudinal", "transverse", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", etc., indicating the orientation or position relationship, are based on the orientation or position relationship shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention.

[0038] In the description of the present invention, unless otherwise specified and limited, it should be noted that the terms "installed", "connected" and "connected" should be understood in a broad sense. For example, it can be a mechanical connection or an electrical connection, or it can be the internal communication between two components. It can be a direct connection or an indirect connection through an intermediate medium. For ordinary technicians in this field, the specific meanings of the above terms can be understood according to the specific circumstances.

[0039] Reference Figure 1 FIG. 1 is a flow chart of a driving style recognition method according to an embodiment of the present invention. In this embodiment, the driving style recognition method includes:

[0040] S1. Obtaining original driving data.

[0041] Specifically, the original driving data is time series data, which can be collected during the vehicle driving process using vehicle sensors (such as GPS, vehicle controller, etc.). The original driving data includes information such as the driving vehicle's speed, acceleration, throttle and regenerative braking signal.

[0042] After obtaining the data collected by the vehicle sensors, the data collected by the vehicle sensors is cleaned and standardized to obtain the final original driving data.

[0043] S2. Extract driving style features of the original driving data according to a first preset rule.

[0044] The driving style characteristics are information that can reflect the driving style of the driving vehicle. The driving style includes at least one of the maximum jerk, average jerk, jerk standard deviation, maximum throttle opening, average throttle opening, throttle opening standard deviation, throttle increase rate, throttle decrease rate, braking signal ratio, braking signal maximum value, average braking duration, brake release rate and brake trigger frequency of the driving vehicle; among which jerk, also called force change rate, is the rate of change of acceleration.

[0045] Specifically, the step of extracting the driving style features of the original driving data according to the first preset rule includes:

[0046] S21 . Segment the original driving data according to a preset time length to obtain a plurality of segmented data.

[0047] S22 . Perform feature extraction on the segmented data to obtain first aggregated features, where the first aggregated features are used to reflect information about the driving style of the segmented data.

[0048] S23. Perform principal component analysis and dimensionality reduction processing on the first aggregated features to obtain second aggregated features.

[0049] S24. Construct a feature matrix based on the second aggregated features of all segmented data to obtain driving style features of the original driving data.

[0050] Specifically, 13 aggregated features reflecting driving style are extracted from each segmented data. Then, principal component analysis (PCA) is performed on the first aggregated features of each segmented data, and the main four principal components that can reflect driving style are retained, thereby reducing the dimensionality and obtaining the second aggregated features.

[0051] PCA is a commonly used data dimensionality reduction technique that converts high-dimensional data into low-dimensional data through linear transformation while preserving the data's key features as much as possible. Here, a value of ∑ = 4 is typically set, compressing the original driving style-related features into a 4-dimensional space for easier processing and analysis.

[0052] S3. Extracting time series statistical features of the original driving data according to a second preset rule.

[0053] Specifically, extracting the time series statistical features of the original driving data according to the second preset rule includes:

[0054] S31. Analyze the segmented data and extract the time series expression of the segmented data.

[0055] S32. Determine local statistical features of the segmented data according to the time series expression.

[0056] The local statistical features include at least one of the mean, standard deviation, maximum value, minimum value, median, 25% quantile, and 75% quantile of the time series expression of the segmented data;

[0057] S33. Integrate the local statistical features of all segmented data to obtain time series statistical features.

[0058] After extracting the required time series expressions (such as speed "Velocity[km / h]", longitudinal acceleration "LongitudinalAcceleration[m / s^2]", throttle "Throttle[%]", regenerative braking signal "Regenerative Braking Signal", etc.) from the original segmented data, the local statistical features of each time series expression (including mean, standard deviation, maximum, minimum, median, 25% quantile, 75% quantile) are calculated respectively to form an extended statistical feature vector, which can capture the dynamic change information within the data and complement the driving style characteristics.

[0059] S4. Fusion of driving style features and time series statistical features to obtain hybrid features.

[0060] Specifically, the steps of fusing driving style features and time series statistical features to obtain hybrid features include:

[0061] S41. Extract global temporal features of local statistical features through an encoder.

[0062] Specifically, linear mapping is performed on the time series statistical features and global average pooling is performed to obtain global time series features.

[0063] The expression of global average pooling is:

[0064]

[0065] Among them, h represents the global temporal feature, T represents the total length of the time step of the original driving data, t represents the time step index corresponding to the segmented data, and U t Represents the time series statistical features after linear mapping corresponding to the segmented data of time step t, d model Feature dimension representing global temporal features;

[0066] Linear mapping and pooling combine the global memory capacity of the state space, allowing the global temporal features to simultaneously integrate information from all moments of the original driving data, retain long-range dependencies, and be naturally robust to segments of different lengths.

[0067] In a preferred implementation of this embodiment, an S4 layer encoder is used to process local statistical features. The full name of the S4 layer is Structured State Space. The time series signals such as the driving speed, acceleration, and throttle of the driving vehicle during driving often contain dynamic changes in long time windows (such as the characteristic patterns before and after a sudden acceleration). Traditional recursive neural networks or long short-term memory networks are prone to gradient vanishing / explosion on very long sequences.

[0068] The S4 layer uses the State Space Model (SSM) to describe the evolution of the system state over time using linear ordinary differential equations:

[0069]

[0070] The S4 layer discretizes this continuous model and learns matrices A, B, C, and D, which can efficiently capture long-term dependencies within the neural network framework.

[0071] Specifically, linear mapping is performed on the time series statistical features and global average pooling is performed to obtain the global time series features. The processing steps are as follows:

[0072] Given a time series input Xseq∈RT×F, it is transformed through a full connection and then globally pooled:

[0073]

[0074] Among them, X seq This represents the corresponding raw time series matrix after the raw driving data is segmented. T represents the time step. For example, if 40 seconds of data is collected at a 10 Hz frequency, there will be 40 × 10 = 400 time points. F is the number of channels. In this example, there are four channels, including speed, acceleration, throttle, and brake signals, meaning that each time point has four dimensions of feature data.

[0075] The "state matrix" of the S4 state space layer maps the F-dimensional input data at each moment (that is, the four channel features at each time point) to a feature space of dimensions. Through matrix multiplication, the data is converted from the original feature dimensions to the feature dimensions required by the model.

[0076] is a bias vector with dimension d model . When input data is passed through W ssm After the linear transformation, the bias vector is added to make the model have stronger fitting ability.

[0077] mean(·) represents the global average pooling operation, which averages all rows of the matrix according to the time dimension. For the matrix obtained by the previous transformation, the dimension is the time step T multiplied by d model , aggregate the information in the time dimension and get a d model dimensional vector h.

[0078] After the preceding operations, the entire time period of the original driving data (originally T × F-dimensional time series data) is compressed into a one-dimensional global representation h. This preserves long-range dependencies, capturing the correlation between features at different points in time throughout the entire time period. This facilitates subsequent model processing based on this global representation, such as classification and prediction.

[0079] S42: Perform weighted fusion on the second aggregated features corresponding to each segmented data and the global temporal features to obtain a hybrid feature.

[0080] Specifically, first, the second aggregate feature and the time series statistical feature are concatenated to obtain a new vector, which is recorded as an auxiliary feature. The expression of the auxiliary feature is:

[0081]

[0082] Among them, aux represents auxiliary features, style represents the second aggregation feature, and stats represents time series statistical features. Dimensions representing mixed features;

[0083] In this embodiment, the dimension of the mixed feature after splicing is d agg +7F, when =4 and the number of channels F=4, the dimension of the mixed feature is 4+28=32 dimensions.

[0084] Global statistical features (including driving style aggregation features and time series statistical features) are combined to form an aux vector, which serves as an auxiliary input for the driving style classification model. In this embodiment, the driving style classification model uses a capsule network, a special neural network structure. This auxiliary input can provide the network with more dimensional information, helping the network to better learn and capture patterns and features in the data, thereby improving the model's performance on related tasks (such as driving behavior analysis and driving style classification).

[0085] After that, the global temporal feature h is concatenated with the auxiliary feature aux, and after layer normalization and Dropout, the final expression of the hybrid feature is:

[0086]

[0087] This is a concatenation operation of the feature vector h output by the S4 layer and the auxiliary feature vector aux obtained previously. h represents the global temporal feature, and aux represents the auxiliary feature. After concatenation, a new joint representation vector is obtained, which is the mixed feature z. The dimension of the mixed feature z is d. model +d agg +7F.

[0088] LayerNorm(·) stands for layer normalization, a normalization technique that normalizes the inputs of all neurons in a neural network layer, ensuring that the input data in all dimensions have the same distribution. This can accelerate model convergence and help stabilize the training process, avoiding problems such as training difficulties caused by large differences in data distribution.

[0089] Dropout (·) is a technique used to prevent overfitting in neural networks. During training, a subset of neuron activations is randomly dropped with a certain probability (0.2 in this example). This prevents the model from becoming overly dependent on certain neurons, enhancing its generalization ability and avoiding overfitting, where the model performs well on the training set but significantly degrades on the test set.

[0090] After layer normalization and Dropout operations, the final feature vector z is obtained, and its dimension is consistent with the concatenated vector dimension, which is still d model +d agg +7F; The overall process generates a normalized (via layer normalization) and overfitting-resistant (via dropout) hybrid feature vector z through feature concatenation, normalization, and overfitting prevention. This feature vector integrates multiple types of feature information, providing higher-quality and more generalizable input for subsequent model tasks (such as classification and regression).

[0091] S5. Input the mixed features into the driving style classification model. The driving style classification model processes the mixed features using a dynamic routing mechanism and outputs a driving style classification result.

[0092] In a preferred implementation of this embodiment, the driving style classification model adopts a capsule network. Traditional neural networks use scalar neurons to represent "whether a certain feature exists", while capsule networks use a vector (or matrix) to represent "the existence of a certain concept and its attributes" (such as direction, amplitude, etc.); the length of the vector represents the probability of the existence of the feature or category; the direction of the vector encodes the feature attribute.

[0093] The capsule network uses a dynamic routing mechanism to process mixed features and obtain the driving style classification results of the original driving data. Specifically, the dynamic routing algorithm processes mixed features through the following steps: primary capsule mapping, prediction capsule generation, and classification output:

[0094] Specifically, the primary capsule mapping is performed first. The expression of the primary capsule mapping is:

[0095]

[0096] W prim Is a weight matrix whose dimension is determined by two parts, where d model +d agg +7F is the dimension of the input mixed feature vector z; I = num_primary represents the number of primary capsules, and D = primary_dim represents the dimension of each primary capsule vector. prim This weight matrix is ​​used to map the input features to the primary capsule space.

[0097] Is a bias matrix with dimension IxD, which is the same as the weight matrix W prim The output dimensions after the operation match. When the fusion features are passed through the weight matrix W prim After the linear transformation, adding this bias term makes the model have stronger fitting ability.

[0098] Represents the primary capsule set, the mixed feature vector z is mixed with the weight matrix W prim Multiply and add the bias term b prim Then, through the Reshape operation, we can get Each primary capsule is a D-dimensional vector, and I such vectors constitute a primary capsule set, where the length D of each capsule vector represents the attribute of the sub-feature.

[0099] The purpose of the primary capsule mapping is to transform the previously generated mixed feature vector (dimension d model +d agg +7F), through the weight matrix W prim and the bias term b prim The data is linearly transformed and reshaped before being projected into multiple "primary capsule" spaces. This is a key step in the capsule network, organizing and representing the fused features in the form of capsules, making it easier for the subsequent capsule network to further learn and process these features and explore the inherent relationships between them.

[0100] Afterwards, the primary capsule collection To compress:

[0101]

[0102] squash(.) represents the compression operation, ||s|| represents the Euclidean norm of vector s, also called L2 norm; for vector s=(s1,s2,…,s n ), its Euclidean norm calculation formula is ||s|| is used to measure the length of a vector.

[0103] Function: In capsule networks, the length of a capsule vector represents its "existence probability." By compressing the vector modulus to [0, 1], when ||s|| approaches 1 when large, the probability of the entity represented by the capsule is high; when the modulus approaches 0, the probability of the entity existing is low. This helps differentiate the activation strength of capsules, allowing the model to better identify and process features of varying importance, enhancing the model's ability to represent and learn data features.

[0104] Then, a prediction capsule is generated, and the expression of the prediction capsule is:

[0105]

[0106] is the capsule mapping matrix dedicated to the jth category, Used to map primary capsules to the category subspace.

[0107] is the “prediction capsule” vector of the i-th primary capsule for the j-th class; For each category j, a dedicated matrix Voting primary capsules into the category space.

[0108] Then perform routing iteration (r=1...R):

[0109]

[0110] is the routing logits from the i-th primary capsule to the j-th category capsule (initialized to 0); Perform softmax operation to obtain the coupling coefficient It represents the allocation ratio of the i-th primary capsule to category j, which is used to measure the degree of association between the primary capsule and the category capsule.

[0111] represents the weighted sum of the j-th capsule, is the prediction vector of the primary capsule after linear mapping, is the coupling coefficient; the prediction vector after linear mapping of I primary capsules Using the coupling coefficient Weighted summation to obtain the weighted sum of the j-th capsule

[0112] Squash(·) represents the compression function, The modulus length is compressed to [0,1]; is the j-th capsule vector, The length of represents the probability of the category existing.

[0113] Is the dot product of the primary capsule prediction vector and the category capsule vector, indicating the "fit" between the two. Add this fit and update the route This enhances the coupling scores of more fitting paths, allowing the model to better capture the association between primary capsules and category capsules.

[0114] Finally, the classification output:

[0115] Finally, each category capsule vector v j The length of is obtained by softmax to obtain the probability distribution:

[0116]

[0117] In the formula, ||v j || represents the length of the j-th capsule vector, which characterizes the probability strength of the existence of this category. K represents the number of driving categories (e.g., K = 3 represents 3 categories).

[0118] and Satisfying the normalization of probability distribution, each represents the predicted probability of the jth class.

[0119] Through the softmax mechanism, the capsule vector length ||v j ||Convert to class probability Exponential function exp(||v j ||) amplifies the differences between different categories, and the denominator ∑ k exp(||v k ||) to achieve normalization and finally complete the classification task (such as selecting The largest category is taken as the prediction result); the dynamic routing mechanism automatically gives greater weight to primary capsules with high "fit", suppresses noisy capsules, and is more robust, thus making the predicted driving style classification results more accurate.

[0120] This application first extracts a global representation of the time series statistical features through the S4 encoder, then splices it with the second aggregated features, and then sends the spliced ​​joint representation (mixed features) to the capsule network module. The feature vectors are hierarchically modeled and weighted fused through a dynamic routing mechanism to achieve adaptive importance learning of each feature in the final classification task, greatly improving the discriminative ability and robustness of driving style recognition. This method can adaptively learn the complex associations between different features, highlight key features, and suppress redundant noise, thereby improving the discriminative ability of feature representation. This is something that traditional MLP, LSTM or TCN models cannot directly achieve.

[0121] Reference Figure 2 In another embodiment of the present application, the driving style recognition method further includes:

[0122] Cluster driving style features and generate driving style pseudo labels;

[0123] The driving style pseudo labels and driving style classification results are used to optimize the network parameters of the driving style classification model, and the driving style classification model is updated in real time.

[0124] Using driving style pseudo-labels and driving style classification results to timely optimize the network parameters of the driving style classification model can further improve the accuracy of the driving style classification results output by the driving style classification model and enhance the robustness of the model.

[0125] Based on the same inventive concept, an embodiment of the present invention provides a driving style recognition system.

[0126] The driving style recognition system of the present invention can be installed in an electronic device. According to the functions to be implemented, the driving style recognition system includes an acquisition module, a first feature extraction module, a second feature extraction module, a fusion module and an output module.

[0127] The acquisition module can acquire raw driving data; the first feature extraction module can extract driving style features of the raw driving data according to a first preset rule; the second feature extraction module can extract time series statistical features of the raw driving data according to a second preset rule; the fusion module can fuse the driving style features and the time series statistical features to obtain hybrid features; the output module can input the hybrid features into a driving style classification model, and the driving style classification model uses a dynamic routing mechanism to process the hybrid features and output a driving style classification result.

[0128] The module described in the present invention may also be referred to as a unit, which refers to a series of computer program segments that can be executed by a processor of an electronic device and can complete fixed functions, and is stored in a memory of the electronic device.

[0129] The various variations and specific examples of the driving style recognition method provided in the above embodiments are also applicable to the driving style recognition system of the present embodiment. Based on the above detailed description of the driving style recognition method, those skilled in the art will clearly understand the implementation method of the driving style recognition system of the present embodiment. For the sake of brevity, these details will not be repeated here.

[0130] This application also discloses an electronic device, such as Figure 3 Figure 2 is a schematic diagram of the structure of an electronic device implementing a driving style recognition method according to an embodiment of the present invention. The electronic device may include at least one processor 10, a memory 11 communicatively coupled to the at least one processor, a communication bus 12, and a communication interface 13. It may also include a computer program stored in the memory 11 and executable on the processor 10, such as a driving style recognition method program.

[0131] In some embodiments, the processor 10 may be comprised of an integrated circuit, such as a single packaged integrated circuit or a plurality of packaged integrated circuits with the same or different functions, including one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and a combination of various control chips. The processor 10 is the control core (Control Unit) of the electronic device, connecting the various components of the entire electronic device using various interfaces and circuits. It executes or runs programs or modules stored in the memory 11 (e.g., a method for executing driving style recognition) and calls data stored in the memory 11 to perform various functions of the electronic device and process data.

[0132] The memory 11 includes at least one type of readable storage medium, and the readable storage medium includes a flash memory, a mobile hard disk, a multimedia card, a card-type memory (for example, an SD or DX memory, etc.), a magnetic memory, a disk, an optical disk, etc. In some embodiments, the memory 11 may be an internal storage unit of an electronic device, such as a mobile hard disk of the electronic device. In other embodiments, the memory 11 may also be an external storage device of an electronic device, such as a plug-in mobile hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped on the electronic device. Furthermore, the memory 11 may also include both an internal storage unit and an external storage device of the electronic device. The memory 11 can be used not only to store application software and various types of data installed in the electronic device, such as the code of the driving style recognition method program, but also to temporarily store data that has been output or is to be output.

[0133] The communication bus 12 may be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus. The bus may be divided into an address bus, a data bus, a control bus, etc. The bus is configured to enable communication between the memory 11 and at least one processor 10, etc.

[0134] The communication interface 13 is used for communication between the above-mentioned electronic device and other devices, including a network interface and a user interface. Optionally, the network interface may include a wired interface and / or a wireless interface (such as a WI-FI interface, a Bluetooth interface, etc.), which is generally used to establish a communication connection between the electronic device and other electronic devices. The user interface may be a display (Display), an input unit (such as a keyboard (Keyboard)), optionally, the user interface may also be a standard wired interface, a wireless interface. Optionally, in some embodiments, the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, and an OLED (Organic Light-Emitting Diode, organic light-emitting diode) touch device, etc. Among them, the display may also be appropriately referred to as a display screen or a display unit, for displaying information processed in the electronic device and for displaying a visual user interface.

[0135] Figure 3 Only the electronic device with components is shown, and it can be understood by those skilled in the art that Figure 3 The structure shown does not constitute a limitation of the electronic device, and may include fewer or more components than shown, or combine certain components, or arrange the components differently. For example, although not shown, the electronic device may also include a power supply (such as a battery) to power each component. Preferably, the power supply can be logically connected to at least one processor 10 through a power management device, so that functions such as charging management, discharging management, and power consumption management are implemented through the power management device. The power supply may also include one or more DC or AC power supplies, recharging devices, power failure detection circuits, power converters or inverters, power status indicators, and other arbitrary components. The electronic device may also include a variety of sensors, Bluetooth modules, Wi-Fi modules, etc., which will not be repeated here.

[0136] It should be understood that the embodiment is for illustration only and the scope of the patent application is not limited to this structure.

[0137] Furthermore, if the module / unit integrated into the electronic device is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. The computer-readable storage medium can be volatile or non-volatile.

[0138] The present application provides a computer-readable storage medium, including, for example, any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a mobile hard drive, a magnetic disk, an optical disk, a computer memory, or a read-only memory (ROM). The computer-readable storage medium stores a computer program capable of being loaded by a processor and executing the driving style recognition method of the above-described embodiment.

[0139] Throughout this specification, reference to terms such as "one embodiment," "some embodiments," "example," "specific example," "one implementation," "a preferred implementation," or "some examples" means that the specific features, structures, materials, or characteristics described in conjunction with that embodiment or example are included in at least one embodiment or example of the present invention. In this specification, schematic representations of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

[0140] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to the embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the claims and their equivalents.

Claims

1. A driving style recognition method, characterized in that: The method comprises: Obtaining raw driving data; extracting driving style features of the original driving data according to a first preset rule; extracting time series statistical features of the original driving data according to a second preset rule; Fusion of driving style features and time series statistical features to obtain hybrid features; The mixed features are input into the driving style classification model, which processes the mixed features using a dynamic routing mechanism and outputs a driving style classification result; The original driving data is time series data, and the extracting of driving style features of the original driving data according to the first preset rule includes: Segmenting the original driving data according to a preset time length to obtain multiple segmented data; Performing feature extraction on the segmented data respectively to obtain first aggregated features, where the first aggregated features are used to reflect information about the driving style of the segmented data; Performing principal component analysis and dimensionality reduction on the first aggregated features to obtain the second aggregated features; Constructing a feature matrix based on the second aggregated features of all segmented data to obtain the driving style features of the original driving data; Extracting the time series statistical features of the original driving data according to the second preset rule includes: Parse segmented data and extract the time series expression of segmented data; Determine local statistical features of the segmented data according to the time series expression, where the local statistical features include at least one of a mean, a standard deviation, a maximum value, a minimum value, a median, a 25% quantile, and a 75% quantile of the time series expression of the segmented data; Integrate the local statistical features of all segmented data to obtain time series statistical features; The hybrid features obtained by fusing driving style features and time series statistical features include: Extract global temporal features of local statistical features through the encoder; The second aggregated features corresponding to each segmented data and the global temporal features are weightedly fused to obtain hybrid features.

2. The driving style recognition method according to claim 1, wherein: The driving style feature is information that can reflect the driving style of the driving vehicle. The driving style includes at least one of the maximum jerk, average jerk, jerk standard deviation, maximum throttle opening, average throttle opening, throttle opening standard deviation, throttle increase rate, throttle decrease rate, brake signal ratio, brake signal maximum value, average brake duration, brake release rate and brake trigger frequency.

3. The driving style recognition method according to claim 1, wherein: The global temporal features of the local statistical features extracted by the encoder include: Perform linear mapping on the time series statistical features and perform global average pooling to obtain global time series features; The expression of global average pooling is: in, Represents the global timing characteristics, represents the total length of time steps of the original driving data, Indicates the time step index corresponding to the segmented data, Represents the time step The time series statistical characteristics after linear mapping corresponding to the segmented data, The feature dimension representing the global time series characteristics.

4. The driving style recognition method according to any one of claims 1 to 3, characterized in that: The method further comprises: Cluster driving style features and generate driving style pseudo labels; The driving style pseudo labels and driving style classification results are used to optimize the network parameters of the driving style classification model, and the driving style classification model is updated in real time.

5. A driving style recognition system, used to implement the driving style recognition method according to any one of claims 1 to 4, characterized in that: include: Acquisition module, used to obtain raw driving data; a first feature extraction module, configured to extract driving style features of the original driving data according to a first preset rule; A second feature extraction module is used to extract time series statistical features of the original driving data according to a second preset rule; Fusion module, used to fuse driving style features and time series statistical features to obtain hybrid features; An output module is used to input the mixed features into the driving style classification model, and the driving style classification model uses a dynamic routing mechanism to process the mixed features and output the driving style classification results; The original driving data is time series data, and the extracting of driving style features of the original driving data according to the first preset rule includes: Segmenting the original driving data according to a preset time length to obtain multiple segmented data; Performing feature extraction on the segmented data respectively to obtain first aggregated features, where the first aggregated features are used to reflect information about the driving style of the segmented data; Performing principal component analysis and dimensionality reduction on the first aggregated features to obtain the second aggregated features; Constructing a feature matrix based on the second aggregated features of all segmented data to obtain the driving style features of the original driving data; Extracting the time series statistical features of the original driving data according to the second preset rule includes: Parse segmented data and extract the time series expression of segmented data; Determine local statistical features of the segmented data according to the time series expression, where the local statistical features include at least one of a mean, a standard deviation, a maximum value, a minimum value, a median, a 25% quantile, and a 75% quantile of the time series expression of the segmented data; Integrate the local statistical features of all segmented data to obtain time series statistical features; The hybrid features obtained by fusing driving style features and time series statistical features include: Extract global temporal features of local statistical features through the encoder; The second aggregated features corresponding to each segmented data and the global temporal features are weightedly fused to obtain hybrid features.

6. An electronic device, characterized in that: The electronic device comprises: at least one processor (10); and, a memory (11) communicatively coupled to the at least one processor (10); The memory (11) stores a computer program executable by the at least one processor (10), and the computer program is executed by the at least one processor (10) so as to enable the at least one processor (10) to execute the driving style recognition method according to any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program; when the computer program is executed by a processor, the driving style recognition method according to any one of claims 1 to 4 is implemented.

Images