Construction method and recognition method of driver behavior intention recognition system

Abstract

The invention provides a construction method of a driver behavior intention recognition system, a recognition method, a terminal and a medium, and belongs to the technical field of intelligent traffic systems. The construction method comprises the following steps: acquiring a straight driving intention multi-source data set and a steering intention multi-source data set; the data preprocessing module is used for preprocessing the data acquired by the multi-source data acquisition module to obtain a straight driving and steering intention clustering analysis data set; respectively carrying out clustering analysis on the straight driving and steering intention clustering analysis data set to obtain a straight driving and steering intention clustering training set; respectively inputting the straight driving and steering intention clustering training sets into a support vector machine for training to obtain a straight driving and steering driving intention recognition model; according to the invention, the driving intention of the driver can be accurately identified in real time.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent transportation systems, and in particular to a construction method, an identification method, a terminal and a medium of a driver behavior intention identification system. Background Art

[0002] With the rapid development of intelligent transportation systems (ITS) and autonomous driving technologies, accurate identification of driving characteristics, especially identification of driver behavior intentions, has become one of the key technologies to improve driving safety and enhance the intelligence level of driver assistance systems. Existing technical solutions mainly rely on various data analysis and modeling methods, such as hidden Markov models, Bayesian filtering technology, biometrics, mean clustering algorithms, and neural network models. These methods extract features by collecting and analyzing driving data, and then predict and classify driving behaviors.

[0003] However, although these methods have improved the accuracy of driving behavior recognition to a certain extent, there are still many shortcomings and challenges. Specifically, the defects of the existing technical solutions are mainly reflected in the following aspects:

[0004] Differences in driving styles: Different drivers exhibit unique driving styles during driving, which vary significantly in speed control, steering habits, reaction time, etc. However, most current studies use a unified parameter model for driving intention recognition, ignoring the differences in driving styles, resulting in insufficient recognition accuracy and real-time performance, making it difficult to effectively avoid safety risks.

[0005] Limitations of classifier performance: Driving characteristics recognition usually relies on a single classifier for classification, which may lead to problems such as overfitting or underfitting due to unclear data distribution characteristics. In addition, the results of a single classifier are accidental, making it difficult to ensure data accuracy and repeatability.

[0006] In view of this, there is an urgent need for a scientific and effective driver behavior intention recognition system, method, terminal and medium. Summary of the invention

[0007] In view of the deficiencies in the prior art, the present invention proposes a construction method, an identification method, a terminal and a medium of a driver behavior intention identification system to solve the technical problems existing in the prior art.

[0008] The technical solution adopted by the present invention is:

[0009] A first aspect of an embodiment of the present application provides a method for constructing a driver behavior intention recognition system, wherein the driver behavior intention recognition system includes a straight driving intention recognition model and a turning intention recognition model, and the method includes:

[0010] Acquire a multi-source dataset of straight driving intention and a multi-source dataset of turning intention;

[0011] Preprocessing the multi-source dataset of straight driving intention to obtain a cluster analysis dataset of straight driving intention, and preprocessing the multi-source dataset of turning intention to obtain a cluster analysis dataset of turning intention;

[0012] Performing cluster analysis on the straight driving intention cluster analysis data set to obtain a straight driving intention cluster training set, and performing cluster analysis on the turning intention cluster analysis data set to obtain a turning intention cluster training set;

[0013] The straight driving intention clustering training set is used to train the initial model of the support vector machine to obtain the straight driving intention recognition model. At the same time, the turning intention clustering training set is used to train the initial model of the support vector machine to obtain the turning driving intention recognition model.

[0014] Further, the said straight driving intention multi-source data set includes: accelerator pedal displacement, brake pedal displacement, accelerator pedal displacement rate, brake pedal displacement rate, vehicle state signal is longitudinal acceleration and longitudinal acceleration mean square error;

[0015] The steering intention multi-source data set includes: steering wheel angle, steering wheel angle velocity, vehicle state signal of longitudinal velocity and vehicle yaw angular velocity.

[0016] Furthermore, the preprocessing of the multi-source dataset of the driving-through intention to obtain a cluster analysis dataset of the driving-through intention includes:

[0017] The straight driving intention multi-source data set is standardized by using the Min-Max standardization processing method to obtain a straight driving intention standardized data set, and a straight driving intention correlation analysis is performed on the data in the straight driving intention standardized data set to obtain a straight driving intention correlation analysis result data set, and the straight driving intention standardized data that is greater than a first preset correlation threshold in the straight driving intention correlation analysis result data set is determined as an element in a straight driving intention dimensionality reduction processing data set, and the principal component analysis method is used to perform dimensionality reduction processing on the elements in the straight driving intention dimensionality reduction processing data set to obtain a straight driving intention cluster analysis data set.

[0018] Furthermore, the multi-source dataset of turning intention is preprocessed to obtain a cluster analysis dataset of turning intention, including:

[0019] The Z-score method is used to standardize the turning intention multi-source dataset to obtain a turning intention standardized dataset. Then, a turning intention correlation analysis is performed on the data in the turning intention standardized dataset to obtain a turning intention correlation analysis result dataset. The turning intention standardized data corresponding to a value greater than a second preset correlation threshold in the turning intention correlation analysis result dataset are determined as elements in the turning intention cluster analysis dataset.

[0020] Further, cluster analysis is performed on the straight driving intention cluster analysis data set to obtain a straight driving intention cluster training set, and cluster analysis is performed on the turning intention cluster analysis data set to obtain a turning intention cluster training set, including:

[0021] Using an improved K-means algorithm to perform cluster analysis on the straight driving intention cluster analysis data set, and obtaining a turning intention cluster training set based on the cluster analysis results;

[0022] An improved K-means algorithm is used to perform cluster analysis on the turning intention cluster analysis data set, and a turning intention cluster training set is obtained based on the cluster analysis results.

[0023] Furthermore, the improved K-means algorithm is used to perform cluster analysis on the straight driving intention cluster analysis data set, and a turning intention cluster training set is obtained based on the cluster analysis results; the improved K-means algorithm is used to perform cluster analysis on the turning intention cluster analysis data set, and a turning intention cluster training set is obtained based on the cluster analysis results. The improved K-means algorithm is expressed as:

[0024]

[0025] Among them, x (u) represents the parameter or eigenvalue of the u-th element, γ (v) represents the parameter or estimated value related to the vth group, n v represents the number of elements in the vth group or category, Q w Represents the weight, which is used to adjust the importance of certain items in the optimization problem. i represents the probability value or parameter associated with the i-th element, P(P i ) is the parameter P i A function of , where k usually represents the number of categories or the number of certain groupings.

[0026] A second aspect of an embodiment of the present application provides a method for identifying a driver's behavior intention, the method comprising:

[0027] Obtain multi-source driving data of the current driver;

[0028] Preprocessing the multi-source data to obtain data to be recognized by intent;

[0029] The intention data to be recognized is input into the driver behavior intention recognition system to obtain the current driver's intention of driving straight or turning.

[0030] A third aspect of an embodiment of the present application provides a terminal, comprising a processor, an input device, an output device and a memory, wherein the processor, input device, output device and memory are interconnected, wherein the memory is used to store a computer program, the computer program comprises program instructions, and the processor is configured to call the program instructions to execute the step instructions as in the second aspect of the embodiment of the present application.

[0031] The fourth aspect of the embodiments of the present application provides a computer-readable storage medium, which stores a computer program. The computer program includes program instructions, which, when executed by a processor, enable the processor to execute the step instructions in the second aspect of the embodiments of the present application.

[0032] It can be seen from the above technical solution that the beneficial technical effects of the present invention are as follows:

[0033] 1. The present invention can accurately identify the driving intentions of each driver in real time through the steps of multi-source data acquisition, data preprocessing, cluster analysis, intention classification construction and intention recognition. This real-time intention recognition capability enables the vehicle to respond instantly when the driver's behavioral intention changes, thereby effectively avoiding potential safety risks.

[0034] 2. The present invention combines supervised learning (support vector machine) with unsupervised learning (improved K-means algorithm), making full use of the advantages of the two learning methods. Supervised learning can efficiently classify data based on known labels, while unsupervised learning can automatically discover hidden patterns in the data and assign labels to the data set. This combination improves the model's recognition accuracy for complex driving intentions. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following is a brief introduction to the drawings required for the specific embodiments or the prior art description. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn according to the actual scale.

[0036] Figure 1 A flow chart of a method for constructing a system according to an embodiment of the present invention;

[0037] Figure 2An analysis diagram of estimating yaw rate using a wheel speed difference estimation algorithm according to an embodiment of the present invention;

[0038] Figure 3 This is a diagram of acceleration and deceleration conditions of straight driving cluster analysis using a common K-means method in an embodiment of the present invention; Figure 3 (a) is the acceleration diagram of the straight driving cluster analysis based on the common K-means method; Figure 3 (b) is the deceleration diagram of straight driving cluster analysis based on the common K-means method;

[0039] Figure 4 This is a diagram of acceleration and deceleration conditions of straight driving cluster analysis using an improved K-means method in an embodiment of the present invention; Figure 4 (a) is the acceleration diagram of the straight driving cluster analysis based on the improved K-means method; Figure 4 (b) is the deceleration diagram of straight driving cluster analysis based on the improved K-means method;

[0040] Figure 5 This is a diagram showing the classification results of the driver's steering intention unsupervised learning according to an embodiment of the present invention;

[0041] Figure 6 The embodiment of the present invention supports vector linear separability and nonlinear separability model diagram; Figure 6 (a) is the support vector linear separable model diagram; Figure 6 (b) is the support vector nonlinear separable model diagram;

[0042] Figure 7 This is a flow chart of a method according to an embodiment of the present invention;

[0043] Figure 8 Schematic diagram of a terminal according to an embodiment of the present invention. DETAILED DESCRIPTION

[0044] The following embodiments of the technical solution of the present invention are described in detail in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and are therefore only used as examples, and cannot be used to limit the protection scope of the present invention.

[0045] It should be noted that, unless otherwise specified, the technical terms or scientific terms used in this application should have the common meanings understood by those skilled in the art to which the present invention belongs.

[0046] Example

[0047] This embodiment provides a method for constructing a driver behavior intention recognition system.

[0048] The working principle of the embodiment is described in detail below:

[0049] like Figure 1As shown, a method for constructing a driver behavior intention recognition system comprises:

[0050] S01: Acquire a multi-source dataset of straight driving intention and a multi-source dataset of turning intention;

[0051] S02: preprocessing the multi-source dataset of straight driving intention to obtain a cluster analysis dataset of straight driving intention, and preprocessing the multi-source dataset of turning intention to obtain a cluster analysis dataset of turning intention;

[0052] S03: performing cluster analysis on the straight driving intention cluster analysis data set to obtain a straight driving intention cluster training set, and performing cluster analysis on the turn intention cluster analysis data set to obtain a turn intention cluster training set;

[0053] S04: using the straight driving intention clustering training set to train the initial model of the support vector machine to obtain a straight driving intention recognition model, and using the turning intention clustering training set to train the initial model of the support vector machine to obtain a turning driving intention recognition model.

[0054] Specifically, the step S01 is to obtain a multi-source dataset of straight driving intention and a multi-source dataset of turning intention;

[0055] The multi-source dataset of straight driving intention collects data such as accelerator pedal displacement, brake pedal displacement, accelerator pedal displacement rate, brake pedal displacement rate, longitudinal acceleration and longitudinal acceleration mean square error in real time through vehicle sensors.

[0056] The steering intention multi-source data set: the steering wheel angle, steering wheel angle velocity, longitudinal velocity and vehicle yaw velocity obtained by wheel speed difference estimation algorithm are collected in real time by vehicle sensors. In one embodiment, Figure 2 As shown, the yaw rate is estimated by the wheel speed difference estimation algorithm, and the expression is as follows:

[0057]

[0058] In the formula, y rate is the yaw rate, which indicates the lateral angular velocity of the vehicle when turning, w 1 is the outer wheel angular velocity, which refers to the angular velocity of the outer wheel relative to the ground when the vehicle is turning, w 2 is the inner vehicle angular velocity, which refers to the angular velocity of the inner wheel relative to the ground when the vehicle is turning, A is the distance between the inner and outer wheels, R r is the radial distance between the two masses, usually referring to the wheelbase of the vehicle, that is, the distance between the front and rear wheels of the vehicle, θ is the angle, which refers to the steering angle when the vehicle turns, v 1 and v 2Represent the linear speed of the outer wheel and the inner wheel respectively.

[0059] In this embodiment, regarding the driver's turning intention data set selection method: 15 participants were recruited, aged between 18 and 29 years old. The drivers were randomly numbered 1 to 15. First, they had normal hearing and normal or corrected vision. Another criterion required that the participants had not participated in any driving simulator research in the past 6 months. Six different driving intentions about turning were defined, namely emergency left turn, emergency right turn, normal left turn, normal right turn, slow left turn, and slow right turn. Therefore, when setting the road, 6 corresponding different roads were set according to the urgency of the turning driving intention. The steering angles were respectively 135° for emergency left turn, 135° for emergency right turn, 90° for normal left turn, 90° for normal right turn, 45° for slow left turn, and 45° for slow right turn. Each driver was tested 10 times in 6 different turning roads, and each driver was tested 60 times. A total of 900 different sets of data can be obtained.

[0060] Step S02, preprocessing the multi-source dataset of straight driving intention to obtain a cluster analysis dataset of straight driving intention, and preprocessing the multi-source dataset of turning intention to obtain a cluster analysis dataset of turning intention;

[0061] The straight driving intention multi-source data set is standardized by using a Min-Max standardization processing method to obtain a straight driving intention standardized data set, the straight driving intention correlation analysis is performed on the data in the straight driving intention standardized data set to obtain a straight driving intention correlation analysis result data set, the straight driving intention standardized data greater than a first preset correlation threshold in the straight driving intention correlation analysis result data set are determined as elements in a straight driving intention dimensionality reduction processing data set, and the principal component analysis method is used to perform dimensionality reduction processing on the elements in the straight driving intention dimensionality reduction processing data set to obtain a straight driving intention cluster analysis data set;

[0062] The Min-Max standardization method is used to standardize the data in the through driving intention multi-source data set, and all the data in the through driving intention multi-source data set are standardized to the interval [0,1], and the expression is:

[0063]

[0064] Among them, y i is the standardized data; x represents the total number of elements or data points considered; P j Represents the value of the jth element in a sequence or set; In the range of j from 1 to x, P j The minimum value of In the range of j from 1 to x, P jThe maximum value of .

[0065] In one embodiment, before the Min-Max standardization processing method is used to standardize the data in the multi-source dataset of the straight driving intention, firstly, the corresponding mean value of each type of data in the obtained multi-source dataset of the straight driving intention is obtained;

[0066] In one embodiment, the standardized data in the straight driving intention standardized data set is subjected to straight driving intention correlation analysis, and the data includes: longitudinal acceleration mean, longitudinal acceleration mean square error, average accelerator pedal opening, and accelerator pedal opening change rate for correlation analysis. The correlation expression is expressed as:

[0067]

[0068] In the formula, ρ ij Represents X i and Y j The correlation coefficient between i , Y j is a pairwise combination of four related features, E represents the expected value, D represents the variance of the random variable, and X j are random variables that can represent different features or observations in the dataset.

[0069] In one embodiment, the straight driving intention standardized data greater than a first preset correlation threshold (such as 0.7) in the straight driving intention correlation analysis result data set is determined as an element in the straight driving intention dimensionality reduction processing data set;

[0070] The main idea of ​​dimensionality reduction by principal component analysis is to convert x related features (P 1 ,P 2 ,…,P x ), after linear transformation, y new features (Q 1 ,Q 2 ,…,Q y ) Dimensionality reduction method. Specific expression:

[0071]

[0072] Where α yx (i=1,2,...,r;j=1,2,...,t) are the principal component coefficients of each relevant feature.

[0073] The variance of the principal component is an indicator to measure the important information of the principal component. In principal component analysis, the larger the variance, the more relevant information it contains about the principal component. The calculation method is as follows:

[0074]

[0075] Among them, η k Indicates the proportion of the kth eigenvalue to the sum of all eigenvalues, where k is an index representing a specific feature group, k = 1, 2, ..., t; μ k represents the kth eigenvalue, and these eigenvalues ​​are arranged in descending order, and μ 1 >μ 2 >…>μ t ; α k represents the eigenvector corresponding to the kth eigenvalue, which is a multidimensional vector containing all elements related to each feature group. m Indicates the ratio of the sum of the first m eigenvalues ​​to the total sum of all eigenvalues; μ s represents the ssth eigenvalue, where s is an index used in calculating γ m The first m eigenvalues ​​are accumulated at the same time; t represents the total number of feature groups. In the above formula, t appears twice, but the meaning of each occurrence is the same, that is, it represents the total number of feature groups.

[0076] The Z-score method is used to standardize the turning intention multi-source dataset to obtain a turning intention standardized dataset. Then, a turning intention correlation analysis is performed on the data in the turning intention standardized dataset to obtain a turning intention correlation analysis result dataset. In the correlation analysis result dataset, features whose correlation is greater than a second preset correlation threshold are selected, and the features are determined as elements in the turning intention cluster analysis dataset.

[0077] In one embodiment, the Z-score method is used to standardize the multi-source data of the turning intention, remove the skewed distribution in the data, and obtain a standardized data set of the turning intention. The calculation formula of the Z-score method is as follows:

[0078]

[0079] In the formula, μ is the mean of the overall data, σ is the standard deviation of the overall data, and x is the individual observation value;

[0080] In one embodiment, before the Z-score standardization method is used to standardize the data in the turn intention multi-source data, firstly, the corresponding mean of each type of data in the obtained turn intention multi-source data is calculated;

[0081] In one embodiment, the standardized data in the steering intention standardized data set is subjected to steering intention correlation analysis, and the data includes: the mean yaw angular velocity, the mean steering wheel angular velocity, the mean longitudinal velocity, and the mean accelerator pedal opening rate change rate are subjected to the following correlation analysis, and the correlation expression is expressed as:

[0082]

[0083] In one embodiment, the turning intention standardized data corresponding to the turning intention correlation threshold value (such as 0.6) in the turning intention correlation analysis result data set is determined as an element in the turning intention cluster analysis data set.

[0084] Step S03, performing cluster analysis on the straight driving intention cluster analysis data set to obtain a straight driving intention cluster training set, and performing cluster analysis on the turn intention cluster analysis data set to obtain a turn intention cluster training set;

[0085] Cluster analysis of driver’s driving intention based on improved K-means method:

[0086] In this embodiment, we use the improved K-means method to perform cluster analysis on the driver's straight driving intention. This method subdivides the driver's intention in longitudinal driving into six categories: slow acceleration, slow deceleration, normal acceleration, normal deceleration, emergency acceleration, and emergency deceleration. These behaviors are achieved by the driver's operation of the accelerator pedal and the brake pedal, which are the key components directly controlled by the driver.

[0087] Through improved K-means clustering, we divide the straight driving intention into three stages: slow, general and urgent. Each stage corresponds to different driving behaviors, where the slow stage includes the intention of slow acceleration and slow deceleration; the general stage includes the intention of general acceleration and general deceleration; the urgent stage includes the intention of urgent acceleration and urgent deceleration. Slow acceleration and slow deceleration may correspond to lower accelerator pedal opening and / or higher brake pedal opening; general acceleration and general deceleration may correspond to medium-range pedal opening; urgent acceleration and urgent deceleration may correspond to higher accelerator pedal opening or significant brake pedal opening.

[0088] Using an improved K-means algorithm to perform cluster analysis on the data in the through driving intention cluster analysis data set, and obtaining a through driving intention cluster training set based on the cluster analysis results;

[0089] K-means algorithm is a random clustering screening algorithm for set partitioning. Due to its high clustering efficiency and its portable characteristics, it has become one of the most widely used algorithms among all random clustering screening algorithms. For the ordinary K-means mean algorithm, only the shortest distance of all samples to the cluster center is considered. The specific classification results are as follows: Figure 3 As shown, blue represents slow status, green represents general status, and red represents emergency status.

[0090] The improved algorithm proposed in this embodiment considers the probability of the driving situation, and then reversely classifies it according to the probability to determine whether a certain behavior is urgent or slow. The specific implementation form of this probability is as follows:

[0091]

[0092] Among them, x (u) is the sample u, γ (v) is the cluster center, n v is the total sample size, Qw represents the weight parameter, P(P i ): represents the probability value or parameter associated with the ith element, which can be a probability estimate of the data point belonging to a certain cluster, or other statistics related to the data point, Nv represents the set of all data points in the vth cluster, and k represents the number of clusters, that is, the number of groups or categories into which the data points are divided.

[0093] In one embodiment, the three dimensions of accelerator pedal displacement, brake pedal displacement, and vehicle longitudinal acceleration are selected for unsupervised learning of data, and the accelerator pedal displacement is regarded as a positive value, and the brake pedal displacement is regarded as a negative value. When they are stepped on at the same time, they are superimposed and regarded as equivalent accelerator / brake pedal positions. This not only plays a role in dimensionality reduction, but also takes into account the intention of acceleration and deceleration.

[0094] The specific classification results after improvement are as follows Figure 4 As shown, blue represents slow status, green represents general status, and red represents emergency status.

[0095] Figure 4 By applying the improved K-means method, the clustering effect of the driver's straight driving intention is demonstrated, and the driving behavior is clearly divided into categories such as slow acceleration, slow deceleration, normal acceleration, normal deceleration, emergency acceleration and emergency deceleration. Figure 3 and Figure 4 It can be seen that the improved algorithm is clearly different from the algorithm before the improvement. The classification on the left and right sides is reduced because the probability of close to normal distribution is considered.

[0096] The improved K-means algorithm is used to perform cluster analysis on the data in the turning intention cluster analysis data set, and a straight driving intention cluster training set is obtained based on the cluster analysis results. Specifically, in the clustering process, appropriate parameter settings, such as the number of clusters and distance measurement, are selected according to the characteristics and requirements of the data. Based on the results of the cluster analysis, the data points are assigned to different clusters, and each cluster represents a data set with similar characteristics. Then, based on these clusters, a turning intention cluster training set is constructed, which will be used for subsequent model training.

[0097] The improved K-means algorithm is expressed as:

[0098]

[0099] Among them, x (u) represents the parameter or eigenvalue of the u-th element, γ (v) represents the parameter or estimated value related to the vth group, n v represents the number of elements in the vth group or category, Q w Represents the weight, which is used to adjust the importance of certain items in the optimization problem. i represents the probability value or parameter associated with the i-th element, P(P i ) is the parameter P i A function of , where k usually represents the number of categories or the number of certain groupings.

[0100] In one embodiment, the variables of three dimensions are selected to perform unsupervised learning of data: the mean yaw angular velocity, the mean steering wheel angular velocity, and the mean vehicle longitudinal velocity.

[0101] Unlike the straight driving situation, these three dimensions cannot be reduced to two dimensions because they are independent of each other and if only two dimensions are considered, the steering information will become incomplete.

[0102] Another difference from the straight driving situation is that left turn and right turn are completely symmetrical; however, the braking and acceleration of the straight driving need to be considered separately because the pedal displacement and acceleration have two different nonlinear relationships. However, left turn and right turn do not need to be considered separately, and their properties are completely symmetrical.

[0103] Different from the unsupervised learning of the driver's intention of driving straight, driving straight takes two dimensions into consideration, while turning left and turning right take three dimensions into consideration, so it is a three-dimensional image. The specific classification results (after normalization) are as follows: Figure 5 As shown, Figure 5 In the figure, each coordinate axis is the normalized coordinate axis, the x-axis is the mean longitudinal velocity of the vehicle, the y-axis is the mean yaw angular velocity, and the z-axis is the mean steering wheel angle. The blue point is an emergency turn, the red point is a slow turn, and the green point is a normal turn. Figure 5 The three black points marked with x in the figure are the cluster centers.

[0104] In this embodiment, we use an improved K-means method to perform cluster analysis on the driver's steering intention. This method subdivides the driving intention related to steering into six categories: slow left turn, slow right turn, general left turn, general right turn, emergency left turn and emergency right turn. When constructing the recognition model of left turn and right turn behavior, the rotation of the steering wheel, the operation of the accelerator pedal and the brake pedal are considered, which are key factors directly controlled by the driver.

[0105] Through the improved K-means clustering method, we divide the driving steering intention into three different stages: slow, general and urgent. Each stage identifies the intention of turning left and right respectively, and in each working condition, the identification of left and right turns is completely symmetrical and there is no essential difference. This classification method helps to understand and predict the driver's steering behavior more accurately, thereby supporting vehicle dynamic control and safety systems.

[0106] Using the improved K-means algorithm for clustering can have the improved K-means algorithm, which provides a more flexible and robust clustering method by introducing additional optimization items and parameter adjustments. This algorithm is better able to handle unbalanced data sets because it takes into account the number and weight of samples within the cluster, thereby improving the quality of clustering. In addition, it is more robust to noise and outliers, allowing reliable clustering results to be obtained even in complex data environments. The improved K-means algorithm also provides faster convergence and better interpretability, enabling it to perform well in a variety of application scenarios. By adjusting the weights and other parameters in the algorithm, users can customize the optimization goals based on specific problems, thereby obtaining clustering results that better meet business needs.

[0107] S04: using the straight driving intention clustering training set to train the initial model of the support vector machine to obtain a straight driving intention recognition model, and using the turning intention clustering training set to train the initial model of the support vector machine to obtain a turning driving intention recognition model.

[0108] The core idea of ​​support vector machine is to find the best classification line called hyperplane to solve the problem of maximizing the interval. Support vector linear separable and nonlinear separable models such as Figure 6 As shown, in this embodiment, the training data for intent recognition are all non-linearly separable.

[0109] Configure the parameters of SVM (such as kernel function, penalty parameter C, etc.) to optimize the model performance. After the training is completed, the straight driving intention recognition model and the turning driving intention recognition model are obtained.

[0110] like Figure 7 As shown, the embodiment of the present application also provides a method for identifying driver behavior intention, the method comprising:

[0111] Obtain multi-source driving data of the current driver;

[0112] Preprocessing the multi-source data to obtain data to be recognized by intent;

[0113] The intention data to be recognized is input into the driver behavior intention recognition system to obtain the current driver's intention of driving straight or turning.

[0114] like Figure 8 As shown, an embodiment of the present application also provides a terminal, including a processor, an input device, an output device and a memory, wherein the processor, input device, output device and memory are interconnected, wherein the memory is used to store a computer program, and the computer program includes program instructions, and the processor is configured to call the program instructions to execute part or all of the steps described in a method for identifying driver behavior intentions in an embodiment of the present application.

[0115] An embodiment of the present application also provides a computer-readable storage medium, characterized in that the computer-readable storage medium stores a computer program, and the computer program includes program instructions, and when the program instructions are executed by a processor, the processor executes part or all of the steps described in a driver behavior intention identification method in an embodiment of the present application.

[0116] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein by equivalents. These modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be included in the scope of the claims and specification of the present invention.

Claims

1. A method for constructing a driver behavior intention recognition system, characterized in that: The driver behavior intention recognition system includes a straight driving intention recognition model and a turning intention recognition model, and the method includes: Acquire a multi-source dataset of straight driving intention and a multi-source dataset of turning intention; Preprocessing the multi-source dataset of straight driving intention to obtain a cluster analysis dataset of straight driving intention, and preprocessing the multi-source dataset of turning intention to obtain a cluster analysis dataset of turning intention; Performing cluster analysis on the straight driving intention cluster analysis data set to obtain a straight driving intention cluster training set, and performing cluster analysis on the turning intention cluster analysis data set to obtain a turning intention cluster training set; The straight driving intention clustering training set is used to train the initial model of the support vector machine to obtain the straight driving intention recognition model. At the same time, the turning intention clustering training set is used to train the initial model of the support vector machine to obtain the turning driving intention recognition model.

2. The method for constructing a driver behavior intention recognition system according to claim 1, characterized in that: The multi-source data set of straight driving intention includes: accelerator pedal displacement, brake pedal displacement, accelerator pedal displacement rate, brake pedal displacement rate, vehicle state signal is longitudinal acceleration and longitudinal acceleration mean square error; The steering intention multi-source data set includes: steering wheel angle, steering wheel angle velocity, vehicle state signal of longitudinal velocity and vehicle yaw angular velocity.

3. A method for constructing a driver behavior intention recognition system according to any one of claims 1-2, characterized in that: The preprocessing of the multi-source dataset of the driving-through intention to obtain the cluster analysis dataset of the driving-through intention includes: The straight driving intention multi-source data set is standardized by using the Min-Max standardization processing method to obtain a straight driving intention standardized data set, and a straight driving intention correlation analysis is performed on the data in the straight driving intention standardized data set to obtain a straight driving intention correlation analysis result data set, and the straight driving intention standardized data that is greater than a first preset correlation threshold in the straight driving intention correlation analysis result data set is determined as an element in a straight driving intention dimensionality reduction processing data set, and the principal component analysis method is used to perform dimensionality reduction processing on the elements in the straight driving intention dimensionality reduction processing data set to obtain a straight driving intention cluster analysis data set.

4. The method for constructing a driver behavior intention recognition system according to claim 3, characterized in that: The preprocessing of the turning intention multi-source data set to obtain the turning intention cluster analysis data set includes: The Z-score method is used to standardize the turning intention multi-source dataset to obtain a turning intention standardized dataset. Then, a turning intention correlation analysis is performed on the data in the turning intention standardized dataset to obtain a turning intention correlation analysis result dataset. The turning intention standardized data corresponding to a value greater than a second preset correlation threshold in the turning intention correlation analysis result dataset are determined as elements in the turning intention cluster analysis dataset.

5. The method for constructing a driver behavior intention recognition system according to claim 1, characterized in that: The cluster analysis is performed on the straight driving intention cluster analysis data set to obtain the straight driving intention cluster training set, and the cluster analysis is performed on the turning intention cluster analysis data set to obtain the turning intention cluster training set, including: Using an improved K-means algorithm to perform cluster analysis on the straight driving intention cluster analysis data set, and obtaining a turning intention cluster training set based on the cluster analysis results; An improved K-means algorithm is used to perform cluster analysis on the turning intention cluster analysis data set, and a turning intention cluster training set is obtained based on the cluster analysis results.

6. The method for constructing a driver behavior intention recognition system according to claim 5, characterized in that: The improved K-means algorithm is used to perform cluster analysis on the straight driving intention cluster analysis data set, and a turning intention cluster training set is obtained based on the cluster analysis result; The improved K-means algorithm is used to perform cluster analysis on the turning intention cluster analysis data set, and a turning intention cluster training set is obtained based on the cluster analysis results. The improved K-means algorithm is expressed as: Among them, x (u) represents the parameter or eigenvalue of the u-th element, γ (v) represents the parameter or estimated value related to the vth group, n v represents the number of elements in the vth group or category, Q w Represents the weight, which is used to adjust the importance of certain items in the optimization problem. i represents the probability value or parameter associated with the i-th element, P(P i ) is the parameter P i A function of , where k usually represents the number of categories or the number of certain groupings.

7. A method for identifying driver behavior intention, characterized in that: Applied to the driver behavior intention recognition system according to any one of claims 1 to 6, the method comprises: Obtain multi-source driving data of the current driver; Preprocessing the multi-source data to obtain data to be recognized by intent; The intention data to be recognized is input into the driver behavior intention recognition system to obtain the current driver's intention of driving straight or turning.

8. A terminal, characterized in that: The method comprises a processor, an input device, an output device and a memory, wherein the processor, the input device, the output device and the memory are interconnected, wherein the memory is used to store a computer program, the computer program comprises program instructions, and the processor is configured to call the program instructions to execute the method as claimed in claim 7.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which includes program instructions. When the program instructions are executed by a processor, the processor is caused to perform the method according to claim 7.

Images