Intelligent driver distraction degree detection method, device, equipment, medium and product

Abstract

The invention discloses an intelligent driver distraction degree detection method and device, equipment, a medium and a product, and relates to the field of distraction driving behavior recognition, and the method comprises the steps: obtaining an original driving data set containing visual behavior parameters, manual operation parameters and vehicle dynamic parameters of a driver; screening key features in the original driving data set by using an SF2SOG method to construct a training data set; assigning a label of distraction degree to the training data through a condensation clustering algorithm; constructing a complete training data set comprising a training data set and a label of the distraction degree corresponding to each training sample data, and training a stacking integration model comprising a random forest classifier, an extreme gradient lifting tree classifier, an adaptive enhancement classifier and a plurality of basic classifiers and a meta classifier based on the complete training data set; and accurate prediction of the distraction degree is realized. According to the invention, accurate and real-time identification of the distraction state of the driver is realized, traffic accidents can be prevented, and the road safety is improved.

Description

Technical Field

[0001] The present application relates to the field of distracted driving behavior recognition, and in particular to an intelligent driver distraction degree detection method, device, equipment, medium and product. Background Art

[0002] With the rapid development of intelligent driving technology, vehicle cockpit design is undergoing unprecedented changes, especially the widespread application of In-Vehicle Information System (IVIS), which has greatly enriched the information interaction during driving. These systems provide drivers with a more convenient and rich driving experience through large screens and multi-functional touch interfaces.

[0003] However, while enjoying the convenience brought by intelligent driving technology, the problem of driver distraction has become increasingly prominent. Complex visual and manual tasks not only easily interfere with the driver's driving operations, but also significantly increase driving risks. Visual and manual distraction have been confirmed by authoritative organizations as important factors leading to traffic accidents. Therefore, how to effectively identify and classify the degree of driver distraction has become the key to improving driving safety.

[0004] In the related art, the driver distraction recognition system is mainly based on simple visual or manual task monitoring, and infers the driver's distraction state by detecting parameters such as the driver's eye movement behavior, driving operation and vehicle dynamics. Most of these methods use a single classification model, such as support vector machine (SVM), random forest (RF), etc. Although they can identify the degree of distraction to a certain extent, they have the problems of insufficient recognition accuracy and poor generalization ability in the face of the complexity and diversity of driving situations. In addition, the existing systems often only consider some driving behavior factors, and fail to make full use of multi-dimensional data for comprehensive analysis and classification, resulting in limited model performance. More importantly, related distraction recognition methods, such as facial features and eye movement data analysis based on convolutional neural network (CNN), although they improve the recognition accuracy to a certain extent, lack the interpretability of the classification process and cannot clearly reveal the impact of each input feature on the final classification result. This lack of interpretability of the classification method brings safety risks in practical applications, because it is impossible to accurately understand why the model makes a specific classification decision, making it difficult to classify and adjust potential errors. Summary of the invention

[0005] The purpose of this application is to provide an intelligent driver distraction detection method, device, equipment, medium and product, which can accurately and effectively identify and interpret the driver's visual-manual distraction level, thereby improving the safety and reliability of the intelligent driving system.

[0006] To achieve the above objectives, this application provides the following solutions:

[0007] In a first aspect, the present application provides an intelligent driver distraction degree detection method, comprising:

[0008] Acquire an original driving data set; the original driving data set includes state values ​​of different driving parameters at different historical moments, and the different driving parameters include visual behavior parameters of the driver, manual operation parameters, and dynamic parameters of the driving vehicle;

[0009] Adopt SF 2 The SOG method determines a preset number of driving parameters with high discrimination in the original driving data set as feature parameters, and constructs a training data set; the training data set includes a plurality of training sample data, and the training sample data is composed of state values ​​of each feature parameter in the same historical time period of the same driver;

[0010] An agglomerative clustering algorithm is used to perform unsupervised clustering on the training data set to obtain a label of the degree of distraction of each training sample data in the training data set; the label of the degree of distraction includes: high distraction, medium distraction and low distraction;

[0011] The training data set and the labels of the degree of distraction corresponding to each training sample data in the training data set constitute a complete training data set;

[0012] Based on the complete training data set, a stacked integration model is trained to obtain a trained stacked integration model; the stacked integration model includes a plurality of different basic classifiers and a meta-classifier; the plurality of different basic classifiers are all connected to the meta-classifier; the plurality of different basic classifiers include a random forest classifier, an extreme gradient boosting tree classifier and an adaptive boosting classifier, the basic classifier is used to predict the degree of distraction, and the meta-classifier is used to fuse the prediction results of the degree of distraction output by each basic classifier to obtain a fusion prediction result of the degree of distraction;

[0013] The state values ​​of each characteristic parameter of the driver to be detected within a preset time period are input into the trained stacked integration model to obtain a fusion prediction result of the distraction degree of the driver to be detected.

[0014] In a second aspect, the present application provides an intelligent driver distraction degree detection device, comprising:

[0015] A data acquisition module is used to obtain an original driving data set; the original driving data set includes state values ​​of different driving parameters at different historical moments, and the different driving parameters include visual behavior parameters of the driver, manual operation parameters, and dynamic parameters of the driving vehicle;

[0016] Feature importance evaluation module, used to adopt SF 2 The SOG method determines a preset number of driving parameters with high discrimination in the original driving data set as feature parameters, and constructs a training data set; the training data set includes a plurality of training sample data, and the training sample data is composed of state values ​​of each feature parameter in the same historical time period of the same driver;

[0017] An unsupervised clustering module is used to perform unsupervised clustering on the training data set using an agglomerative clustering algorithm to obtain a label of the degree of distraction of each training sample data in the training data set; the labels of the degree of distraction include: high distraction, medium distraction and low distraction;

[0018] A training data integration module, used for forming a complete training data set by integrating the training data set and the labels of the degree of distraction corresponding to each training sample data in the training data set;

[0019] A model training module, used for training a stacked ensemble model based on the complete training data set to obtain a trained stacked ensemble model; the stacked ensemble model includes a plurality of different basic classifiers and a meta-classifier; the plurality of different basic classifiers are all connected to the meta-classifier; the plurality of different basic classifiers include a random forest classifier, an extreme gradient boosting tree classifier and an adaptive boosting classifier, the basic classifier is used for predicting the degree of distraction, and the meta-classifier is used for fusing the prediction results of the degree of distraction output by each basic classifier to obtain a fused prediction result of the degree of distraction;

[0020] The distraction detection module is used to input the state values ​​of each characteristic parameter of the driver to be detected within a preset time period into the trained stacked integration model to obtain a fusion prediction result of the distraction degree of the driver to be detected.

[0021] In a third aspect, the present application provides a computer device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of any one of the above-described intelligent driver distraction degree detection methods.

[0022] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the above-mentioned intelligent driver distraction degree detection methods.

[0023] In a fifth aspect, the present application provides a computer program product, including a computer program, which, when executed by a processor, implements the steps of any of the above-mentioned intelligent driver distraction degree detection methods.

[0024] According to the specific embodiments provided in this application, this application has the following technical effects:

[0025] The present application provides an intelligent driver distraction degree detection method, device, equipment, medium and product, which obtains the original driving data set containing the driver's visual behavior parameters, manual operation parameters and driving vehicle dynamic parameters, and uses SF 2 The SOG method selects characteristic parameters with high discrimination to construct a training data set, which solves the problem of inaccurate selection of driving parameters and the impact on the accuracy of distraction detection in traditional methods, and realizes the accurate capture and utilization of key driving parameters. Furthermore, the present application uses an agglomerative clustering algorithm to perform unsupervised clustering on the training data set, and labels the degree of distraction for the training sample data, including high distraction, medium distraction and low distraction. This step solves the problem of high cost and low efficiency of manual labeling of distraction degree labels, and realizes the automatic and efficient classification of the driver's distraction state. Based on the complete training data set, the present application trains a stacked ensemble model containing multiple basic classifiers and a meta-classifier. Among them, basic classifiers such as random forest classifiers, extreme gradient boosting tree classifiers and adaptive boosting classifiers are used for preliminary prediction of distraction degree, while meta-classifiers are used to fuse the prediction results of each basic classifier to obtain a more accurate and stable distraction degree fusion prediction result. This design solves the problem of insufficient generalization ability and limited prediction accuracy of a single classifier model, and realizes accurate and real-time recognition of the driver's distraction degree. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0027] Figure 1 A flowchart of an intelligent driver distraction degree detection method provided in one embodiment of the present application;

[0028] Figure 2 A flowchart of an intelligent driver distraction degree detection method provided by another embodiment of the present application;

[0029] Figure 3A schematic diagram of a process for selecting feature parameters using an L2,0 norm sparsity learning method provided in an embodiment of the present application;

[0030] Figure 4 A schematic diagram of a flow chart of the stacked integration model training steps provided in one embodiment of the present application;

[0031] Figure 5 A schematic diagram of functional modules of an intelligent driver distraction degree detection device provided in one embodiment of the present application;

[0032] Figure 6 A schematic diagram of the structure of a computer device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0033] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0034] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0035] In an exemplary embodiment, Figure 1 and Figure 2 As shown, an intelligent driver distraction degree detection method is provided, which is executed by a computer device, and can be executed by a computer device such as a terminal or a server alone, or by a terminal and a server together. In the embodiment of the present application, the method includes the following steps 101 to 106. Among them:

[0036] Step 101, obtaining an original driving data set; the original driving data set includes state values ​​of different driving parameters at different historical moments, and the different driving parameters include visual behavior parameters, manual operation parameters and dynamic parameters of the driving vehicle of the driver.

[0037] Step 102, using SF 2 The SOG method determines a preset number of driving parameters with high discrimination in the original driving data set as feature parameters to construct a training data set; the training data set includes a plurality of training sample data, and the training sample data consists of state values ​​of each feature parameter of the same driver in the same historical time period.

[0038] Step 103, using an agglomerative clustering algorithm to perform unsupervised clustering on the training data set to obtain a label of the distraction level of each training sample data in the training data set; the labels of the distraction level include: high distraction, medium distraction and low distraction.

[0039] Step 104 , the training data set and the labels of the distraction levels corresponding to each training sample data in the training data set form a complete training data set.

[0040] Step 105: Based on the complete training data set, a stacked integration model is trained to obtain a trained stacked integration model; the stacked integration model includes multiple different basic classifiers and a meta-classifier; the multiple different basic classifiers are all connected to the meta-classifier; the multiple different basic classifiers include a random forest classifier, an extreme gradient boosting tree classifier and an adaptive boosting classifier, the basic classifiers are used to predict the degree of distraction, and the meta-classifier is used to fuse the prediction results of the degree of distraction output by each basic classifier to obtain a fused prediction result of the degree of distraction.

[0041] Step 106 , inputting the state values ​​of each characteristic parameter of the driver to be detected within a preset time period into the trained stacked integration model to obtain a fusion prediction result of the distraction degree of the driver to be detected.

[0042] By implementing the above steps 101 to 106, the present application integrates an efficient feature selection method, a clustering algorithm and a stacked ensemble learning model, which can improve the accuracy of identifying the degree of driver distraction and is suitable for the complex environment of intelligent driving.

[0043] In another exemplary embodiment of the present application, in order to collect multi-dimensional driving parameter data of the driver, step 101 specifically includes:

[0044] Through driving simulation experiments, the driving parameters of 35 participants were collected using a driving simulator and a Tobii eye tracking device to form the original driving data set. The driving parameters include:

[0045] Driver’s visual behavior parameters: such as average pupil diameter (APD), total duration of glances (TDG), and gaze hit rate (GH).

[0046] Driver's manual operation parameters: such as number of manual operations, operation time, hit rate, etc.

[0047] Dynamic parameters of driving the vehicle: such as steering wheel operating angle, vehicle speed, vehicle lateral position, etc.

[0048] To ensure the accuracy and completeness of the data, the original driving data set is preprocessed. The preprocessing steps include:

[0049] Fill missing values: Use reasonable interpolation methods or default values ​​to fill in the missing parts of the data set to avoid the impact of data loss on subsequent analysis.

[0050] Delete blank values: Completely remove meaningless blank records in the data set to ensure data purity and usability.

[0051] Digital encoding of text features: Use the LabelEncoder tool in the sklearn library to digitally encode the text features in the dataset (such as driver number, vehicle type, etc.) to facilitate subsequent data processing and analysis.

[0052] After the above preprocessing steps, a driving dataset with clear structure, complete data and easy analysis is obtained, which lays a solid foundation for subsequent research and analysis.

[0053] In another exemplary embodiment of the present application, in order to reduce data redundancy and noise and improve model training efficiency, step 102 specifically includes:

[0054] Since the collected driving parameters have high dimensions, direct use may lead to model overfitting, so this application uses SF 2 The SOG (Sparse and Flexible Projection for Unsupervised Feature Selection) method is used to select feature parameters.

[0055] SF 2 SOG is an unsupervised embedded feature selection method that can select the most discriminative features through adaptive graphs and sparsity constraints while maintaining the data manifold structure. Its formula is as follows:

[0056]

[0057] Wherein, F represents the manifold embedding matrix composed of the state values ​​of a preset number of selected driving parameters, each column represents a driving parameter, each row represents the state values ​​of each driving parameter of the same driver at the same historical moment, W represents the dimensionality reduction transformation matrix used to map the manifold embedding matrix to a low-dimensional space, G represents the adaptive similarity matrix composed of the similarities between the driving parameters in F, Tr() represents the sum of the elements on the diagonal of the matrix, and L Grepresents the Laplacian matrix of G, X represents the original state matrix constructed based on the original driving dataset, S represents the initial similarity matrix composed of the similarities between the driving parameters in X, λ and γ represent the first hyperparameter and the second hyperparameter respectively, Indicates calculation of F norm and square, W T W=I m represents the orthogonality constraint of W, I m represents the m-dimensional identity matrix, ||W|| 2,0 = h represents the sparsity constraint of W, h represents the number of selected features; g i is the i-th row vector of the matrix G, indicating the similarity between the i-th driving parameter and all driving parameters, J represents the all-1 matrix, g ij is the i-th row and j-th column element in G, indicating the similarity between the i-th driving parameter and the j-th driving parameter, and T represents the transposed matrix.

[0058] like Figure 3 As shown in the figure, the feature parameter selection process is given, in which the driving parameters from the original driving data set are directly selected as feature parameters. Six different types of driving parameters are used as examples. Each row represents a type of driving parameter, and the numbers 1-6 represent the six types of driving parameters. Each column represents a training sample data. The color of the circle represents the importance of the driving parameter to the sample, and the numbers 1, 2, 4, and 6 represent the numbers of the selected feature parameters. After applying the L2,0 norm for sparsity calculation, a certain type of feature will be judged as unimportant and thus not selected.

[0059] In the application of this application, SF 2 The SOG method is used to select 15 most discriminative features from the original driving dataset. These features include pupil diameter, gaze duration, vehicle lateral position change, etc. The selected feature parameters provide more effective input for subsequent clustering and classification.

[0060] In another exemplary embodiment of the present application, step 103 specifically includes:

[0061] Agglomerative clustering algorithm is used to perform unsupervised clustering on the training data set. Agglomerative clustering is a bottom-up hierarchical clustering algorithm that can merge the most similar training sample data into one cluster by calculating the distance between training sample data until the predetermined number of clusters is reached.

[0062] In this application, the agglomerative clustering algorithm is used to classify the driver's distraction behavior into three categories: high distraction, medium distraction, and low distraction. The clustering process is as follows:

[0063] Initialization: Treat each training sample data in the training data set as a separate cluster.

[0064] Calculate distance: Calculate the distance between all clusters, using Euclidean distance as the distance metric.

[0065] Merge clusters: Merge the two closest clusters each time until the predetermined number of clusters (three categories) is reached to obtain the clustering result.

[0066] Generate labels: According to the clustering results, each training sample in the training data set is labeled with a corresponding distraction level (high, medium, low). These clustering labels are used for subsequent stacking ensemble model training.

[0067] In another exemplary embodiment of the present application, step 105 specifically includes:

[0068] Use the Heuristic Stacking Ensemble Model, which is a common ensemble learning method that combines multiple layers of models to improve classification results. In this application, the stacking ensemble model includes the following parts:

[0069] Basic classifiers: Random Forest (RF), eXtreme Gradient Boosting (XGBoost), Adaptive Boosting (Adaboost). Each basic classifier classifies the data separately.

[0070] Meta-classifier: Logistic Regression (LR), which is used to integrate the prediction results of the basic classifiers to generate a fusion prediction result of the degree of distraction.

[0071] The base classifier is trained by 5-fold cross validation to prevent overfitting and improve the generalization ability of the model. In order to further optimize the model performance, the present application introduces the particle swarm optimization (PSO) algorithm to automatically adjust the hyperparameters of the base classifier.

[0072] The core idea of ​​PSO is to search for the optimal solution by simulating the foraging behavior of bird flocks. It updates the position of particles (candidate solutions) by moving them in the search space, thereby finding the parameter combination that maximizes the objective function (classification accuracy).

[0073] In this application, PSO is used to optimize the hyperparameters of random forest and XGBoost, such as the number of trees and maximum depth, and the optimal parameter settings are obtained through iterative search, ultimately improving the prediction accuracy of the stacked ensemble model.

[0074] Among them, implementing this embodiment, such as Figure 4As shown in the figure, the specific training process of the stacked model is as follows:

[0075] Divide the training dataset into multiple subsets.

[0076] For each subset, different base classifiers are used for training and prediction to obtain the prediction results of the distraction degree of each base classifier.

[0077] These prediction results are used as new features and combined into a new training dataset.

[0078] Use this new training dataset to train a meta-classifier.

[0079] Finally, the trained meta-classifier is used to make a fusion prediction of the distraction level. The input of the stacked ensemble model is SF 2 The 15 most discriminative feature parameters selected by the SOG algorithm, the output is the result of the stacked ensemble model to predict the degree of distraction. The label is the label of the degree of distraction generated by the cluster analysis.

[0080] After training, the stacked ensemble model predicts the driver's distraction level and evaluates the model through indicators such as accuracy (ACC), precision, recall, and F1 score. Experiments show that the stacked ensemble model of this application shows high accuracy in distraction tasks of different difficulty levels, especially in high-distraction scenarios, the classification accuracy is significantly better than the traditional single model, and finally reaches 96.25% accuracy.

[0081] In another exemplary embodiment of the present application, in order to improve the transparency and interpretability of the model, after step 105, the following steps are further included:

[0082] Based on the complete training data set, the SHAP value of each feature parameter is calculated; the SHAP value represents the influence of the feature parameter on the fusion prediction result.

[0083] This application uses SHAP (SHapley Additive exPlanations) analysis. SHAP is based on cooperative game theory and explains the impact of each feature on the classification result by calculating the marginal contribution of the feature. The specific analysis steps are as follows:

[0084] 1. For each input sample, calculate the SHAP value of each feature parameter to indicate the influence of the feature parameter on the fusion prediction result.

[0085] SHAP value for each feature The calculation formula is as follows:

[0086]

[0087] in, represents the SHAP value of the i′th feature parameter, {x M} represents a set of state values ​​containing M characteristic parameters, M = 1, 2, 3, ..., N; M ≠ i′, N represents the total number of characteristic parameters, and P represents {x M}, x i′ represents the state value of the i′th feature parameter, f() represents the stacked ensemble model, P∪{x i′} means that the eigenvalue x of the i'th feature i′ Add to the feature subset P.

[0088] 2. Visualize the SHAP value to show the contribution of each feature to different prediction results.

[0089] 3. Through global and local analysis, the most important feature parameters are identified, such as the fluctuation of the pupil diameter (APD) and the lateral position of the vehicle, which have a significant impact on the prediction results of the distraction level.

[0090] SHAP explains the prediction target using an additive feature attribution approach, in which the input feature parameters are combined to account for the contribution of each input feature parameter to the prediction of the stacked ensemble model. A given sample x i′ , represents the state value of the kth feature parameter in the ith sample, where k = 1, 2, ..., n, and the model's prediction for this sample is denoted by y i′ , and the average prediction of all samples is expressed as The formula is as follows:

[0091]

[0092] The meaning of this formula is to accumulate the effects of each feature on the prediction result to produce a prediction value. For a given sample, the prediction value y i′ , can be decomposed into the sum of the SHAP values ​​of all feature parameters plus the average prediction value of all samples Indicates the contribution of the kth feature parameter in the i-th sample to the prediction result (i.e., SHAP value), and explains y by calculating the SHAP value of each feature parameter i′ How is it obtained by the contribution of each feature? If it is greater than 0, it means that the feature improves the prediction value and has a positive effect; if If it is less than 0, it means that the feature reduces the predicted value and has a reverse effect.

[0093] The present application also provides an application scenario, which applies the above-mentioned intelligent driver distraction degree detection method. Specifically: The intelligent driver distraction degree detection method provided in this embodiment can be applied in the driver status monitoring scenario of the intelligent driving assistance system. The driver status monitoring scenario of the intelligent driving assistance system includes a data collection link, a data analysis and processing link, and a state feedback and intervention link. The driver's driving data enters the data analysis and processing link from the data collection link. The intelligent driver distraction degree detection method provided in the embodiment of the present application belongs to the data analysis and processing link. In this link, the predicted result of the driver's distraction degree is obtained. According to the prediction results, a warning is issued to the driver in various ways such as sound, vision or touch in the state feedback and intervention link to remind him to pay attention to driving safety. If the driver's distraction level continues to be high, the system can further take intervention measures, such as automatically adjusting the vehicle speed, lane keeping assistance, etc., to ensure driving safety.

[0094] Based on the same inventive concept, the embodiment of the present application also provides an intelligent driver distraction degree detection device for implementing the intelligent driver distraction degree detection method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more embodiments of the intelligent driver distraction degree detection device provided below can refer to the limitations of the intelligent driver distraction degree detection method above, and will not be repeated here.

[0095] In an exemplary embodiment, Figure 5 As shown, an intelligent driver distraction degree detection device is provided, comprising:

[0096] The data acquisition module 201 is used to obtain an original driving data set; the original driving data set includes state values ​​of different driving parameters at different historical moments, and the different driving parameters include the driver's visual behavior parameters, manual operation parameters and dynamic parameters of the driving vehicle.

[0097] Feature importance evaluation module 202 is used to use SF 2 The SOG method determines a preset number of driving parameters with high discrimination in the original driving data set as feature parameters to construct a training data set; the training data set includes a plurality of training sample data, and the training sample data consists of state values ​​of each feature parameter of the same driver in the same historical time period.

[0098] The unsupervised clustering module 203 is used to perform unsupervised clustering on the training data set using an agglomerative clustering algorithm to obtain a distraction level label for each training sample data in the training data set; the distraction level label includes: high distraction, medium distraction and low distraction.

[0099] The training data integration module 204 is used to form a complete training data set by combining the training data set and the label of the distraction level corresponding to each training sample data in the training data set.

[0100] The model training module 205 is used to train the stacked integration model based on the complete training data set to obtain a trained stacked integration model; the stacked integration model includes multiple different basic classifiers and a meta-classifier; the multiple different basic classifiers are all connected to the meta-classifier; the multiple different basic classifiers include a random forest classifier, an extreme gradient boosting tree classifier and an adaptive boosting classifier, the basic classifiers are used to predict the degree of distraction, and the meta-classifier is used to fuse the prediction results of the degree of distraction output by each basic classifier to obtain a fused prediction result of the degree of distraction.

[0101] The distraction detection module 206 is used to input the state values ​​of each characteristic parameter of the driver to be detected within a preset time period into the trained stacked integration model to obtain a fusion prediction result of the distraction degree of the driver to be detected.

[0102] As an optional implementation, the intelligent driver distraction degree detection device of the present application further includes: an explanatory analysis module 207 .

[0103] The explanatory analysis module 207 is used to calculate the SHAP value of each feature parameter based on the complete training data set; wherein the SHAP value represents the degree of influence of the feature parameter on the fusion prediction result;

[0104] The SHAP (Shapley Additive Explanatory Model) theory is used to explain the prediction results of the distraction degree, quantify the impact of each input feature parameter on the fusion prediction results of the distraction degree, and provide global and local explanatory analysis.

[0105] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Figure 6As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store data for detecting the degree of driver distraction. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, an intelligent driver distraction detection method is implemented.

[0106] Those skilled in the art will understand that Figure 6 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components. In an exemplary embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program.

[0107] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0108] In an exemplary embodiment, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0109] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0110] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0111] The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. The non-relational database may include a distributed database based on blockchain, etc., but is not limited thereto. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but is not limited thereto.

[0112] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0113] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this 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 this application.

Claims

1. An intelligent driver distraction degree detection method, characterized in that: The intelligent driver distraction degree detection method comprises: Acquire an original driving data set; the original driving data set includes state values ​​of different driving parameters at different historical moments, and the different driving parameters include visual behavior parameters of the driver, manual operation parameters, and dynamic parameters of the driving vehicle; Adopt SF 2 The SOG method determines a preset number of driving parameters with high discrimination in the original driving data set as feature parameters, and constructs a training data set; the training data set includes a plurality of training sample data, and the training sample data is composed of state values ​​of each feature parameter in the same historical time period of the same driver; An agglomerative clustering algorithm is used to perform unsupervised clustering on the training data set to obtain a label of the degree of distraction of each training sample data in the training data set; the label of the degree of distraction includes: high distraction, medium distraction and low distraction; The training data set and the labels of the degree of distraction corresponding to each training sample data in the training data set constitute a complete training data set; Based on the complete training data set, a stacked integration model is trained to obtain a trained stacked integration model; the stacked integration model includes a plurality of different basic classifiers and a meta-classifier; the plurality of different basic classifiers are all connected to the meta-classifier; the plurality of different basic classifiers include a random forest classifier, an extreme gradient boosting tree classifier and an adaptive boosting classifier, the basic classifier is used to predict the degree of distraction, and the meta-classifier is used to fuse the prediction results of the degree of distraction output by each basic classifier to obtain a fusion prediction result of the degree of distraction; The state values ​​of each characteristic parameter of the driver to be detected within a preset time period are input into the trained stacked integration model to obtain a fusion prediction result of the distraction degree of the driver to be detected.

2. The intelligent driver distraction degree detection method according to claim 1, characterized in that: The formula for calculating the discriminability of driving parameters in the original driving dataset is: Wherein, F represents the manifold embedding matrix composed of the state values ​​of a preset number of selected driving parameters, each column represents a driving parameter, each row represents the state values ​​of each driving parameter of the same driver at the same historical moment, W represents the dimensionality reduction transformation matrix used to map the manifold embedding matrix to a low-dimensional space, G represents the adaptive similarity matrix composed of the similarities between the driving parameters in F, Tr() represents the sum of the elements on the diagonal of the matrix, and L G represents the Laplacian matrix of G, X represents the original state matrix constructed based on the original driving dataset, S represents the initial similarity matrix composed of the similarities between the driving parameters in X, λ and γ represent the first hyperparameter and the second hyperparameter respectively, Indicates calculation of F norm and square, W T W=I m represents the orthogonality constraint of W, I m represents the m-dimensional identity matrix, ||W|| 2,0 = h represents the sparsity constraint of W, h represents the number of selected features; g i is the i-th row vector of the matrix G, indicating the similarity between the i-th driving parameter and all driving parameters, J represents the all-1 matrix, g ij is the i-th row and j-th column element in G, indicating the similarity between the i-th driving parameter and the j-th driving parameter, and T represents the transposed matrix.

3. The intelligent driver distraction degree detection method according to claim 1, characterized in that: The agglomerative clustering algorithm is used to perform unsupervised clustering on the training data set to obtain the distraction level label of each training sample data in the training data set, including: Each training sample data in the training data set is regarded as a separate cluster; Calculate the distance between all clusters, using Euclidean distance as the distance metric; Each time, the two closest clusters are merged until the preset number of clusters is reached to obtain the clustering result; According to the clustering results, each training sample data in the training data set is labeled with the corresponding degree of distraction.

4. The intelligent driver distraction degree detection method according to claim 1, characterized in that: Get the trained stacked ensemble model, followed by: Based on the complete training data set, the SHAP value of each feature parameter is calculated; the SHAP value represents the influence of the feature parameter on the fusion prediction result.

5. The intelligent driver distraction degree detection method according to claim 4, characterized in that: The calculation formula of the SHAP value of the feature parameter is: in, represents the SHAP value of the i'th feature parameter, {x M } represents a set of state values ​​containing M characteristic parameters, M = 1, 2, 3, ..., N; M ≠ i′, N represents the total number of characteristic parameters, and P represents {x M }, x i′ represents the state value of the i′th feature parameter, f() represents the stacked ensemble model, P∪{x i′ } means that the eigenvalue x of the i'th feature i′ Add to the feature subset P.

6. The intelligent driver distraction degree detection method according to claim 1, characterized in that: The base classifier was trained using the 5-fold cross-validation method, and the hyperparameters were optimized using the particle swarm optimization algorithm.

7. An intelligent driver distraction detection device, characterized in that: The intelligent driver distraction degree detection device comprises: A data acquisition module is used to obtain an original driving data set; the original driving data set includes state values ​​of different driving parameters at different historical moments, and the different driving parameters include visual behavior parameters of the driver, manual operation parameters, and dynamic parameters of the driving vehicle; Feature importance evaluation module, used to adopt SF 2 The SOG method determines a preset number of driving parameters with high discrimination in the original driving data set as feature parameters, and constructs a training data set; the training data set includes a plurality of training sample data, and the training sample data is composed of state values ​​of each feature parameter in the same historical time period of the same driver; An unsupervised clustering module is used to perform unsupervised clustering on the training data set using an agglomerative clustering algorithm to obtain a label of the degree of distraction of each training sample data in the training data set; the labels of the degree of distraction include: high distraction, medium distraction and low distraction; A training data integration module, used for forming a complete training data set by integrating the training data set and the labels of the degree of distraction corresponding to each training sample data in the training data set; A model training module, used for training a stacked ensemble model based on the complete training data set to obtain a trained stacked ensemble model; the stacked ensemble model includes a plurality of different basic classifiers and a meta-classifier; the plurality of different basic classifiers are all connected to the meta-classifier; the plurality of different basic classifiers include a random forest classifier, an extreme gradient boosting tree classifier and an adaptive boosting classifier, the basic classifier is used for predicting the degree of distraction, and the meta-classifier is used for fusing the prediction results of the degree of distraction output by each basic classifier to obtain a fused prediction result of the degree of distraction; The distraction detection module is used to input the state values ​​of each characteristic parameter of the driver to be detected within a preset time period into the trained stacked integration model to obtain a fusion prediction result of the distraction degree of the driver to be detected.

8. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the intelligent driver distraction degree detection method according to any one of claims 1 to 6.

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 intelligent driver distraction degree detection method according to any one of claims 1 to 6 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the intelligent driver distraction degree detection method according to any one of claims 1 to 6 is implemented.

Images