Driver fatigue and distraction driving state identification method based on behavior map

Abstract

The invention discloses a driver fatigue and distraction driving state identification method based on a behavior map, and the method comprises the steps: data collection and preparation: extracting natural driving data from a vehicle-mounted global positioning system, and processing an abnormal value and a vacancy value; the method comprises the following steps of: constructing an abnormal behavior map: by taking five behavior variables in longitudinal and transverse dimensions as a data basis, firstly extracting and coding abnormal behavior data, secondly depicting abnormal behavior sheet characteristics, then constructing the abnormal behavior map, and extracting behavior map characteristic data; and fatigue and distraction driving state identification: using the behavior map characteristic data as an input variable of a machine learning model, and identifying different risk driving states of the driver through the machine learning model. The defects of a traditional driving state recognition method can be overcome, and the applicability of the method is improved on the premise that the driving state recognition precision is guaranteed.

Description

Technical Field

[0001] The present invention relates to the field of intelligent transportation technology, and in particular to a method for identifying driver fatigue and distracted driving states based on behavior graphs. Background Art

[0002] In the field of traffic safety, drivers, as participants in road traffic, have a direct impact on driving safety through their driving behavior. Drivers are easily distracted and fatigued in monotonous and tedious driving environments, which leads to a large number of accidents. However, existing methods for identifying risky driving behaviors present several problems and challenges.

[0003] Depending on the data source, risky driving state recognition methods can be categorized as video-based and driving behavior-based. The accuracy of video-based recognition methods is significantly affected by parameters such as the capture device's clarity and frame rate. Furthermore, video image recognition models are generally complex and require high-performance equipment. Behavior-based recognition methods, which mostly use aggregated data metrics such as mean and standard deviation, struggle to accurately capture dynamic changes within micro-periods of the driving process, resulting in low recognition accuracy.

[0004] Therefore, it is necessary to provide a new method for identifying the driver's risky driving state to make up for the shortcomings of traditional driving state identification methods. Summary of the Invention

[0005] In light of this, the purpose of this invention is to provide a behavioral graph-based method for identifying driver fatigue and distracted driving states. This method addresses the shortcomings of traditional driving state identification methods and improves their applicability while ensuring accurate identification. This method provides technical support for improving driving safety and accurately identifying risky driver driving states, promoting the innovation and application expansion of traffic safety technologies, and contributing to the construction of a safer and more efficient transportation ecosystem.

[0006] The present invention solves the technical problem by adopting the following technical solutions:

[0007] A method for identifying driver fatigue and distracted driving status based on behavioral graphs includes the following steps:

[0008] S1, data collection and preparation: extracting natural driving data from the vehicle's global positioning system and handling outliers and missing values;

[0009] S2, Graph Construction: Using the five behavioral variables in the vertical and horizontal dimensions as the data foundation, first extract and encode abnormal behavior data, then characterize the sheet characteristics of abnormal behavior, and then construct an abnormal behavior graph and extract the behavioral graph feature data;

[0010] S3, Fatigue and distracted driving state identification: Use behavioral graph feature data as input variables of the machine learning model to identify the driver's different risky driving states through the machine learning model.

[0011] Furthermore, in step S1, longitudinal velocity data, acceleration data, and latitude and longitude data are extracted from the vehicle-mounted global positioning system, and the acceleration data is used to calculate the jerk. The jerk, longitudinal velocity, and acceleration data reflect the driver's longitudinal control ability; the latitude and longitude data are used to calculate the yaw angular velocity and lateral acceleration to reflect the driver's lateral control ability.

[0012] Furthermore, in step S2, the method for extracting and encoding abnormal behavior data is as follows:

[0013] The quantiles of each variable were calculated. For the variable at each moment, if the variable value fell between the 85th and 90th percentiles or between the 10th and 15th percentiles, it was coded as 1, indicating a low risk level; if it fell between the 90th and 95th percentiles or between the 5th and 10th percentiles, it was coded as 2, indicating a medium risk level; if it exceeded the 95th percentile or was lower than the 5th percentile, it was coded as 3, indicating a high risk; otherwise, it was coded as 0.

[0014] Furthermore, in step S2, the method for characterizing the sheet characteristics of abnormal behavior is as follows:

[0015] Different features are distinguished by color boxes; in order to highlight the differences in the coding values ​​in the features, different grayscale filling colors are used to represent low, medium and high risks respectively; the darker the color, the higher the coding value and the higher the risk; the length of the sheet feature is used to represent the duration of the feature coding.

[0016] Furthermore, in step S2, the method for constructing the abnormal behavior map is as follows:

[0017] Merge and group behavioral data based on rules and complete graph construction.

[0018] Furthermore, the rules used in constructing the abnormal behavior graph are as follows:

[0019] The flake features within the same time interval are merged, and the flake features at different time intervals are grouped.

[0020] Furthermore, in step S2, the method for extracting behavior graph feature data is as follows:

[0021] The severity, duration and frequency of each abnormal behavior feature are extracted to obtain a graph feature dataset of the driver's abnormal driving behavior.

[0022] The present invention provides a method for identifying driver fatigue and distracted driving status based on behavior graphs, which has the following beneficial effects:

[0023] (1) Graphical representation of driving behavior and extraction of characteristic indicators based on behavior graphs:

[0024] The present invention comprehensively considers the three key dimensions of duration, severity and frequency of abnormal characteristics of driver's driving behavior, and constructs a system for characterizing driver's abnormal driving behavior. Starting from the micro level, the abnormal behavior of the driver is characterized by using graphical technology, and the behavioral data is converted into graphical information. While realizing data visualization, the data feature extraction is completed. Whether it is the behavior analysis of individual drivers or the large-scale behavior research of driver groups, the key information of abnormal behavior can be captured with the help of this method, which not only helps to identify risky driving states, but also provides a data basis and technical means for behavioral intervention, safety training and risk assessment, thereby improving the management and control capabilities and research level of abnormal driver behavior.

[0025] (2) Accurately identify fatigue and distracted driving states based on graph features:

[0026] Another breakthrough of the present invention is that it significantly improves the accuracy of identifying risky driving states based on driving behavior. Based on graph feature data and machine learning method models, it achieves accurate identification of driving states. Traditional methods are limited by data processing capabilities and analytical model limitations, and often have difficulty accurately capturing subtle changes in driving behavior and potential risk factors, resulting in low recognition accuracy. In practical applications, it cannot meet the demand for real-time and accurate monitoring of risky driving states. However, by introducing graph features and machine learning algorithms, the present invention deeply mines the implicit information in driving behavior data, can accurately distinguish between normal driving and risky driving states, and provide timely and reliable risk warnings to traffic management departments, transportation companies, etc., thereby ensuring road traffic safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 is a flow chart of the method of the present invention;

[0028] Figure 2 This is a schematic diagram of abnormal behavior data encoding according to the present invention;

[0029] Figure 3 This is a schematic diagram of the abnormal behavior flake characteristics of the present invention;

[0030] Figure 4 This is the abnormal behavior map of the driver of the present invention. DETAILED DESCRIPTION

[0031] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0032] refer to Figure 1 This invention provides a method for identifying driver fatigue and distracted driving states based on behavioral graphs. Based on real-time driving behavior data, the method uses behavioral graphs to characterize and extract key driver behavior features. This feature information is then used as input to identify corresponding risky driving states using a machine learning algorithm. The overall approach is to use driver behavior graphs and machine learning models to construct a highly accurate, easy-to-use, and scalable method for identifying driver fatigue and distracted driving states to ensure traffic safety. This method includes the following steps:

[0033] S1, Data Collection and Preparation: Extract naturalistic driving data from the vehicle's GPS and process outliers and missing values. The vehicle's GPS provides longitudinal velocity, acceleration, and latitude and longitude data. Acceleration data is used to calculate jerk, which, along with longitudinal velocity and acceleration, reflects the driver's longitudinal control ability. Latitude and longitude data are also used to calculate yaw rate and lateral acceleration, reflecting the driver's lateral control ability.

[0034] S2, map construction: Using the five behavioral variables in the vertical and horizontal dimensions as the data basis, first extract and encode the abnormal behavior data, then characterize the sheet characteristics of abnormal behavior, and then construct the abnormal behavior map and extract the behavioral map feature data.

[0035] The method for extracting and encoding abnormal behavior data is as follows:

[0036] Each variable's quantile is calculated. For each variable at each moment, if the variable value falls between the 85th and 90th percentiles or the 10th and 15th percentiles, it is coded as 1, indicating a low risk level; if it falls between the 90th and 95th percentiles or between the 5th and 10th percentiles, it is coded as 2, indicating a medium risk level; if it exceeds the 95th percentile or falls below the 5th percentile, it is coded as 3, indicating a high risk level; otherwise, it is coded as 0. During the map construction process of the present invention, only features coded 1 to 3 are displayed, indicating the degree of abnormal driving behavior.

[0037] The method for characterizing the flake characteristics of abnormal behavior is as follows:

[0038] Different features are distinguished by color boxes; in order to highlight the differences in the coding values ​​in the features, different grayscale filling colors are used to represent low, medium and high risks respectively; the darker the color, the higher the coding value and the higher the risk; the length of the sheet feature is used to represent the duration of the feature coding.

[0039] The method for constructing an abnormal behavior map is as follows:

[0040] After constructing the abnormal behavior sheet features, the behavioral data needs to be merged and grouped to complete the map construction. To make the map more reflective of the driver's actual operating behavior and emphasize the continuity of driving operations, the present invention merges sheet features within the same time interval and distinguishes sheet features from different time intervals. Based on this, the abnormal behavior map is constructed.

[0041] The method for extracting behavioral graph feature data is as follows:

[0042] The severity, duration and frequency of each abnormal behavior feature are extracted to obtain a graph feature dataset of the driver's abnormal driving behavior.

[0043] S3, Fatigue and distracted driving state identification: Utilize behavioral graph feature data as input variables of the machine learning model to identify different risk driving states such as distraction and fatigue of the driver through the machine learning model.

[0044] This invention aims to characterize and detect risky driving states. By combining graph-based methods with machine learning, it leverages driving behavior data to identify driving states corresponding to behavioral characteristics. This method boasts high robustness, adaptability, and interpretability, improving recognition accuracy and complementing existing methods for identifying risky driving states based on video images. This invention can be applied to industries such as insurance claims and driver training and education, providing strong support for traffic safety and addressing the practical needs of the transportation sector for accurately identifying risky driving states and improving traffic safety.

[0045] Example

[0046] Taking a truck fleet's natural driving data for one year as an example, this embodiment provides a method for identifying a driver's risky driving state based on real-time behavior graphs. The specific details are as follows:

[0047] 1) Extract abnormal driving behavior data. Taking acceleration as an example, different codes are assigned to indicators falling in different intervals. Features coded as 1, 2, and 3 are extracted, such as Figure 2 shown.

[0048] 2) After encoding the feature data, the data is presented in the form of sheet features. The sheet feature frames of different features are different, and the grayscale of the filling color represents the severity of the abnormal feature. The darker the color, the more serious it is. Figure 3 shown.

[0049] 3) After the flake feature extraction is completed, the flake features are merged and differentiated based on the rules to form a driver's abnormal behavior feature map, such as Figure 4 To express the characteristics more concisely in the map, English abbreviations are used to distinguish different characteristics in the map, where SP is speed, AC is longitudinal acceleration, JE is jerk, YR is yaw rate, and LA is lateral acceleration.

[0050] 4) Based on the driver's abnormal behavior feature map, the occurrence frequency and duration indicators of different severity risk levels of the driver's abnormal characteristics can be extracted, and the resulting behavior map feature index extraction table is shown in Table 1.

[0051]

[0052] 5) Based on the feature data in Table 2, machine learning models can be used to identify risky driving conditions such as distraction and fatigue. Compared with machine learning recognition models built using other feature data, those based on graph feature data have higher accuracy in identifying abnormal driver conditions. Table 2 shows the accuracy of abnormal condition recognition for different models.

[0053] index mean mean + standard deviation Atlas Accuracy 0.76 0.83 0.92 <![CDATA[R 2 ]]> 0.73 0.88 0.93 AIC 1154 1084 1054

[0054] 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 make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for identifying driver fatigue and distracted driving status based on behavior graph, characterized in that: The steps include: S1, data collection and preparation: extracting natural driving data from the vehicle's global positioning system and handling outliers and missing values; S2, map construction: using the five behavioral variables in the vertical and horizontal dimensions as the data basis, first extract and encode the abnormal behavior data, then characterize the abnormal behavior flake characteristics, and then construct the abnormal behavior map and extract the behavior map feature data; S3, fatigue and distracted driving state identification: Use behavioral graph feature data as input variables of the machine learning model to identify the driver's different risky driving states through the machine learning model.

2. The method for identifying driver fatigue and distracted driving status based on behavior graph according to claim 1, characterized in that: In step S1, longitudinal velocity data, acceleration data and latitude and longitude data are extracted from the vehicle-mounted global positioning system, and the acceleration data is used to calculate the jerk. The jerk, longitudinal velocity and acceleration data reflect the driver's longitudinal control ability; the latitude and longitude data are used to calculate the yaw angular velocity and lateral acceleration, reflecting the driver's lateral control ability.

3. The method for identifying driver fatigue and distracted driving status based on behavior graph according to claim 2, characterized in that: In step S2, the method for extracting and encoding abnormal behavior data is as follows: The quantiles of each variable are calculated. For the variable at each moment, if the variable value falls between the 85th and 90th percentiles or between the 10th and 15th percentiles, it is coded as 1, indicating a low risk level; if it falls between the 90th and 95th percentiles or between the 5th and 10th percentiles, it is coded as 2, indicating a medium risk level; if it exceeds the 95th percentile or is lower than the 5th percentile, it is coded as 3, indicating a high risk; otherwise, it is coded as 0.

4. The method for identifying driver fatigue and distracted driving status based on behavior graph according to claim 3, characterized in that: In step S2, the method for describing the flake characteristics of abnormal behavior is as follows: Different features are distinguished by color boxes; in order to highlight the differences in the coding values ​​in the features, different grayscale filling colors are used to represent low, medium and high risks respectively; the darker the color, the higher the coding value and the higher the risk; the length of the flake feature is used to represent the duration of the feature coding.

5. The method for identifying driver fatigue and distracted driving status based on behavior graph according to claim 4, characterized in that: In step S2, the method for constructing the abnormal behavior map is as follows: Merge and group behavioral data based on rules and complete graph construction.

6. The method for identifying driver fatigue and distracted driving status based on behavior graph according to claim 5, characterized in that: The rules used in constructing the abnormal behavior graph are as follows: The flake features within the same time interval are merged, and the flake features at different time intervals are grouped.

7. The method for identifying driver fatigue and distracted driving status based on behavior graph according to claim 6, characterized in that: In step S2, the method for extracting behavior graph feature data is as follows: The three indicators of severity, duration and frequency of each abnormal behavior feature are extracted to obtain a graph feature dataset of the driver's abnormal driving behavior.

Images