A method for identifying dangerous driving behaviors of drivers based on text data

Abstract

The present application discloses a method for identifying dangerous driving behaviors of drivers based on text data, including: obtaining and processing voice data and text data within a historical preset time period to obtain standard text data and constructing multiple text data reference sets; screening out the characteristic text data of the standard text data in the multiple text data reference sets, generating a training data set according to the characteristic text data, and constructing a dangerous driving behavior prediction model; collecting first text data, and obtaining a danger prediction index of the first text data based on the dangerous driving behavior prediction model; judging whether the driver has dangerous driving behaviors according to the danger prediction index, if so, generating a first reminder instruction, and obtaining second text data, performing an application evaluation on the first reminder instruction according to the second text data, and judging whether to generate a second reminder instruction according to the application evaluation result; ensuring the high efficiency and accuracy of the identification of drivers' dangerous driving behaviors, and improving road safety and traffic efficiency.

Description

Technical Field

[0001] This application relates to the technical field of driver dangerous driving behavior recognition, and particularly to a method for recognizing driver dangerous driving behavior based on text data. Background Art

[0002] With the sharp increase in the number of motor vehicles, driving safety has become one of the social safety issues that attract the attention of all mankind. More than 90% of traffic accidents are related to drivers' critical driving behaviors, such as fatigue driving, distraction, sudden illness, etc.

[0003] In the prior art, the characteristics of drivers' dangerous driving behaviors are recognized by obtaining text data. However, the processing volume and analysis volume of text data are large, and dangerous driving behaviors cannot be quickly recognized, reducing the efficiency and accuracy of recognizing drivers' dangerous driving behaviors, and reducing road safety and traffic efficiency. Summary of the Invention

[0004] To solve the above technical problems, this application provides a method for recognizing driver dangerous driving behavior based on text data. By processing voice data and text data, characteristic text data of multiple categories is obtained. The characteristic text data of multiple categories during dangerous driving periods is used to generate a training data set, and a dangerous driving behavior prediction model is constructed. According to the dangerous driving behavior prediction model, real-time text data is analyzed to accurately identify whether the current driver has dangerous driving behavior. If so, a reminder instruction is issued, and whether the danger index drops is judged based on the text data after the reminder instruction, ensuring the efficiency and accuracy of recognizing drivers' dangerous driving behaviors, and improving road safety and traffic efficiency.

[0005] In some embodiments of this application, a method for recognizing driver dangerous driving behavior based on text data is provided, including:

[0006] Obtain voice data and text data within multiple historical preset periods, process the voice data and text data to obtain standard text data and construct multiple text data reference sets;

[0007] Analyze the standard text data in multiple text data reference sets and screen out characteristic text data. Generate a training data set according to the characteristic text data, and construct a dangerous driving behavior prediction model according to the training data set;

[0008] Collect first text data, and obtain the danger prediction index of the first text data based on the dangerous driving behavior prediction model;

[0009] Determine whether a driver has dangerous driving behavior according to the danger prediction index. If so, generate a first reminder instruction, obtain second text data, evaluate the application of the first reminder instruction according to the second text data, and determine whether to generate a second reminder instruction according to the application evaluation result.

[0010] In some embodiments of the present application, the voice data and text data are processed to obtain multiple standard text data and construct multiple text data reference sets, including:

[0011] Based on the speech recognition technology, convert the speech data into text data, and preprocess the text data. The preprocessing includes unifying the time unit, removing outliers, and filling in missing values;

[0012] Perform standardization processing on the preprocessed text data. The standardization processing includes classifying the preprocessed text data to obtain text data of multiple categories, and standardizing the text data of the same category;

[0013]

[0014] where x' (i) is the standard text data of the i-th text data x (i) in the same category, and n is the number of data in the same category;

[0015] Construct a text data reference set according to the standard text data of different categories within the same historical preset period.

[0016] In some embodiments of the present application, analyze and screen out the characteristic text data from the standard text data in multiple text data reference sets, including:

[0017] Construct a standard text data category matrix H for the standard text data in each text data reference set;

[0018]

[0019] where xj' (i) is the i-th standard text data in the j-th category, where j = 1, 2,... m, i = 1, 2,... n;

[0020] Analyze the standard text data of the same category in the standard text data category matrix H to obtain the fluctuation value and change trend of the standard text data of the same category, generate a characteristic coefficient corresponding to the standard text data according to the fluctuation value and change trend, and set the standard text data with the characteristic coefficient greater than the preset characteristic coefficient threshold as the characteristic text data of the corresponding category;

[0021] Construct a feature text data category matrix H based on the feature text data in multiple categories in the same text data reference set T ;

[0022]

[0023] where x1 T is the feature factor in the first category, s1' (i) is the feature coefficient corresponding to the i-th standard text data in the first category, ∝ is the screening formula, where the screening formula is used to screen out the standard text data whose feature coefficient is greater than the preset feature coefficient threshold, xj T is the feature factor in the j-th category, sj' (i) is the feature coefficient corresponding to the i-th standard text data in the j-th category, where j = 1, 2,..., m.

[0024] In some embodiments of the present application, a training data set is generated according to the feature text data, and a dangerous driving behavior prediction model is constructed according to the training data set, including:

[0025] Establish a time reference line for the historical preset period of the same feature text data category matrix H T and set multiple acquisition time nodes based on a preset time interval;

[0026] Obtain the feature text data of multiple categories at each acquisition time node of the same time reference line, compare the feature text data with the preset normal standard text data interval, mark the pending dangerous data according to the comparison result, calculate the danger coefficient of each pending dangerous data, and divide the pending dangerous data into true dangerous data and false dangerous data according to the danger coefficient;

[0027] Generate a danger index for the current acquisition time node according to the true dangerous data, danger coefficient and the weight coefficient of the corresponding category at each acquisition time node, and set the initial attention time node and the end attention time node within the current historical preset period according to the danger index;

[0028] Intercept the true dangerous data between the initial attention time node and the end attention time node in each time reference line, and set the intercepted true dangerous data as a set of training data;

[0029] Construct a training data set from multiple sets of training data corresponding to multiple time reference lines, and perform neural network training to obtain the corresponding dangerous driving behavior prediction model.

[0030] In some embodiments of the present application, generating a danger index for the current acquisition time node according to the true dangerous data, danger coefficient and the weight coefficient of the corresponding category includes:

[0031] Obtain the characteristic text data that is not within the preset normal standard text data range and whose characteristic text data difference is greater than the preset data difference threshold according to the comparison result, and set the corresponding characteristic text data as pending dangerous data;

[0032] Generate a danger coefficient corresponding to the pending dangerous data according to the deviation degree between the characteristic text data difference of the pending dangerous data and the preset data difference threshold;

[0033] If the danger coefficient at the last acquisition time node of the pending dangerous data is less than the preset danger coefficient threshold, set the corresponding pending dangerous data as false dangerous data;

[0034] If the danger coefficient at the last acquisition time node of the pending dangerous data is greater than the preset danger coefficient threshold, set the corresponding pending dangerous data as true dangerous data;

[0035] Generate a danger index A at the current acquisition time node according to the true dangerous data at each acquisition time node according to the category, danger coefficient, and weight coefficient of the corresponding category;

[0036]

[0037] Among them, m0 is the number of categories corresponding to the true dangerous data, Zv,j is the characteristic text data difference of the v-th true dangerous data in the j-th category, Z0v,j is the preset data difference threshold of the v-th true dangerous data in the j-th category, ∣Zv,j - Z0,j∣ is the danger coefficient of the v-th true dangerous data in the j-th category, bj is the weight coefficient of the j-th category, and r,j is the number of true dangerous data in the j-th category.

[0038] In some embodiments of the present application, based on the dangerous driving behavior prediction model, obtaining the danger prediction index of the first text data includes:

[0039] Collect the first text data within the preset time period, perform preprocessing and standardization processing on the first text data to obtain the first standard text data of different categories, and screen out the first characteristic text data of the first standard text data of different categories;

[0040] Establish a time reference line for the preset time period, and set multiple first preset acquisition nodes based on the recognition accuracy;

[0041] Input the first characteristic text data of multiple categories at the same first preset acquisition node into the dangerous behavior prediction model to obtain the danger prediction index of multiple first preset acquisition nodes.

[0042] In some embodiments of the present application, judging whether the driver has dangerous driving behavior according to the danger prediction index includes:

[0043] Arrange the danger prediction indices at multiple first preset acquisition nodes according to the time sequence of the first preset acquisition nodes to obtain a danger prediction index sequence corresponding to the first text data;

[0044] Obtain the danger prediction index range of the first text data and the average growth rate of the danger prediction index according to the danger prediction index sequence;

[0045] If the danger prediction index range is less than the preset danger index threshold or the average growth rate is greater than the preset growth rate threshold, it is determined that the driver has dangerous driving behavior.

[0046] In some embodiments of the present application, the application evaluation of the first reminder instruction according to the second text data includes:

[0047] Obtain the second text data, process the second text data to obtain second standard text data of different categories, and screen out second feature text data of different categories of the second standard text data;

[0048] Input the second feature text data of different categories at the same time node into the dangerous behavior prediction model to obtain the danger prediction index at the corresponding time node, and generate the decline rate of the danger prediction index of the second text data according to the danger prediction indices at adjacent time nodes;

[0049] Predict the application duration until the danger prediction index is less than the preset danger prediction index threshold according to the decline rate of the danger prediction index of the second text data;

[0050] Generate an application evaluation result of the first reminder instruction according to the application duration.

[0051] In some embodiments of the present application, judging whether to generate a second reminder instruction according to the application evaluation result includes:

[0052] Preset an application evaluation threshold;

[0053] If the application evaluation result is less than the application evaluation threshold, generate a second reminder instruction;

[0054] If the application evaluation result is greater than the application evaluation threshold, do not generate a second reminder instruction.

[0055] A method for identifying a driver's dangerous driving behavior based on text data according to an embodiment of the present application, compared with the prior art, its beneficial effects are as follows:

[0056] By processing voice data and text data, feature text data of multiple categories are obtained. The feature text data of multiple categories during dangerous driving periods are used to generate a training dataset, and a dangerous driving behavior prediction model is constructed. According to the dangerous driving behavior prediction model, real-time text data is analyzed to accurately identify whether the current driver has dangerous driving behavior. If so, a reminder instruction is issued, and it is judged whether the danger index decreases based on the text data after the reminder instruction, ensuring the efficiency and accuracy of the identification of the driver's dangerous driving behavior and improving road safety and traffic efficiency. Description of the Drawings

[0057] Figure 1 It is a schematic diagram of a method for identifying a driver's dangerous driving behavior based on text data in a preferred embodiment of an embodiment of the present application. Detailed Embodiments

[0058] The following will further describe in detail the specific embodiments of the present application in conjunction with the drawings and embodiments. The following embodiments are used to illustrate the present application, but are not used to limit the scope of the present application.

[0059] In the description of the present application, it should be understood that the orientation or positional relationship indicated by the terms "center", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc. is based on the orientation or positional relationship shown in the drawings, and is only for the convenience of describing the present application and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and thus should not be construed as a limitation to the present application.

[0060] The terms "first" and "second" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the quantity of the indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include one or more of such features. In the description of the present application, unless otherwise stated, the meaning of "a plurality" is two or more.

[0061] In the description of the present application, it should be noted that unless otherwise clearly defined and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense. For example, it may be a fixed connection, a detachable connection, or an integral connection; it may be a mechanical connection or an electrical connection; it may be directly connected or indirectly connected through an intermediate medium, and it may be the communication inside two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present application can be understood according to specific situations.

[0062] Such as Figure 1As shown in the figure, a method for identifying dangerous driving behaviors of drivers based on text data in a preferred embodiment of an embodiment of the present application includes:

[0063] Step S101: Obtain voice data and text data within multiple historical preset time periods, process the voice data and text data to obtain standard text data, and construct multiple text data reference sets;

[0064] Step S102: Analyze the standard text data in multiple text data reference sets, screen out characteristic text data, generate a training data set based on the characteristic text data, and construct a dangerous driving behavior prediction model according to the training data set;

[0065] Step S103: Collect first text data, and based on the dangerous driving behavior prediction model, obtain a dangerous prediction index of the first text data;

[0066] Step S104: Determine whether the driver has dangerous driving behaviors according to the dangerous prediction index. If so, generate a first reminder instruction, obtain second text data, evaluate the application of the first reminder instruction according to the second text data, and determine whether to generate a second reminder instruction according to the application evaluation result.

[0067] In this embodiment, the historical preset time period is 30 minutes before the driver has a traffic accident due to dangerous driving. The text data includes physiological data, steering wheel pressure data, etc. The voice data can be converted into text data through voice recognition technology.

[0068] In this embodiment, by processing the voice data and text data, multiple categories of characteristic text data are obtained. The multiple categories of characteristic text data during the dangerous driving period are used to generate a training data set, and a dangerous driving behavior prediction model is constructed. According to the dangerous driving behavior prediction model, the real-time text data is analyzed to accurately identify whether the current driver has dangerous driving behaviors. If so, a reminder instruction is issued, and it is determined whether the dangerous index drops according to the text data after the reminder instruction, ensuring the high efficiency and accuracy of the identification of the driver's dangerous driving behaviors and improving road safety and traffic efficiency.

[0069] In some embodiments of the present application, processing the voice data and text data to obtain multiple standard text data and constructing multiple text data reference sets includes:

[0070] Based on voice recognition technology, convert the voice data into text data, and preprocess the text data. The preprocessing includes unifying time units, removing outliers, and filling in missing values;

[0071] Standardize the preprocessed text data. The standardization process includes classifying the preprocessed text data to obtain multi-category text data, and standardizing the text data of the same category.

[0072]

[0073] Among them, x' (i) is the standard text data of the i-th text data x (i) in the same category, and n is the number of data in the same category.

[0074] Construct a text data reference set according to the standard text data of different categories within the same historical preset period.

[0075] In this embodiment, unification is to uniformly process the data to ensure that the data has similar scales and distributions. Missing values refer to abnormal losses and other situations existing in the data. Outliers are data in the dataset that are significantly different from other data. Data cleaning processing is a process of identifying and correcting errors, inconsistencies, or incomplete data in the dataset.

[0076] In this embodiment, according to the data categories determined by classifying the text data, the data categories include physiological categories, stress categories, temperature categories, etc.

[0077] In this embodiment, through the preprocessing and standardization unit, multiple text data are uniformly processed, reducing the data differences between the text data, improving the quality and credibility of the text data, laying a data foundation for subsequent identification of driver dangerous driving behavior factors based on the text data, ensuring the accuracy of driver dangerous driving behavior identification, and thus reducing the occurrence of traffic accidents.

[0078] In some embodiments of the present application, analyze and screen out the feature text data from the standard text data in multiple text data reference sets, including:

[0079] Construct a standard text data category matrix H for the standard text data in each text data reference set;

[0080]

[0081] Among them, xj' (i) is the i-th standard text data in the j-th category, where j = 1, 2, … m, and i = 1, 2, … n;

[0082] Analyze the standard text data of the same category in the standard text data category matrix H to obtain the fluctuation magnitude value and the change trend of the standard text data of the same category. Generate the characteristic coefficient corresponding to the standard text data according to the fluctuation magnitude value and the change trend, and set the standard text data with the characteristic coefficient greater than the preset characteristic coefficient threshold as the characteristic text data in the corresponding category;

[0083] Construct a characteristic text data category matrix H according to the characteristic text data in multiple categories in the same text data reference set T ;

[0084]

[0085] where, x1 T is the characteristic factor in the first category, s1' (i) is the characteristic coefficient corresponding to the i-th standard text data in the first category, ∝ is the screening formula, where the screening formula is used to screen out the standard text data with the characteristic coefficient greater than the preset characteristic coefficient threshold, xj T is the characteristic factor in the j-th category, sj' (i) is the characteristic coefficient corresponding to the i-th standard text data in the j-th category, where j = 1, 2,..., m.

[0086] In this embodiment, the fluctuation magnitude value refers to the change amount of each standard text data within a historical preset time period, and the change trend refers to whether the change amount of each standard text data tends to be normal or abnormal. Compare each standard text data with the preset normal standard text data interval. If, over time, the standard text data is closer to the preset normal standard text data interval, the change trend is a normal trend and the change trend is set to -1. If, over time, the standard text data is farther from the preset normal standard text data interval, the change trend is an abnormal trend and the change trend is set to 1. The characteristic coefficient = fluctuation magnitude value * change trend. If the characteristic coefficient is larger than the preset characteristic coefficient threshold, the corresponding standard text data is the characteristic text data in the corresponding category.

[0087] In this embodiment, the standard text data is screened to obtain the characteristic text data of each category. The characteristic text data refers to the text data that can accurately represent the dangerous driving behavior of the driver, reducing the analysis amount and processing amount of the text data, accelerating the speed of identifying whether the driver has dangerous driving behavior, ensuring the timeliness and accuracy of the reminder instruction, and thus reducing the incidence of traffic accidents.

[0088] In some embodiments of the present application, a training data set is generated according to the characteristic text data, and a dangerous driving behavior prediction model is constructed according to the training data set, including:

[0089] For the same characteristic text data category matrix H TEstablish a time reference line for the historical preset time period, and set multiple acquisition time nodes based on a preset time interval;

[0090] Obtain the characteristic text data of multiple categories at each acquisition time node of the same time reference line, compare the characteristic text data with the preset normal standard text data interval, mark the pending dangerous data according to the comparison result, calculate the danger coefficient of each pending dangerous data, and divide the pending dangerous data into true dangerous data and false dangerous data according to the danger coefficient;

[0091] Generate the danger index of the current acquisition time node according to the true dangerous data, danger coefficient and weight coefficient of the corresponding category at each acquisition time node, and set the initial attention time node and the end attention time node within the current historical preset time period according to the danger index;

[0092] Intercept the true dangerous data between the initial attention time node and the end attention time node in each time reference line, and set the intercepted true dangerous data as a set of training data;

[0093] Construct a training data set from multiple sets of training data corresponding to multiple time reference lines, and perform neural network training to obtain a corresponding dangerous driving behavior prediction model.

[0094] In this embodiment, the preset normal standard text data interval is constructed according to the characteristic text data corresponding to the driver's normal driving multiple times. When the danger coefficient is larger and until the danger coefficient of the corresponding pending dangerous data at the last acquisition time node is greater than the preset danger coefficient threshold, it is a true danger coefficient. False dangerous data refers to the data that confuses the recognition of the driver's dangerous driving behavior and cannot accurately determine that the driver has dangerous driving behavior.

[0095] In this embodiment, the initial attention time node refers to the start time node when the danger index is in an upward trend, and the end attention time node refers to the start time node when the danger index changes from an upward trend to a downward trend.

[0096] In this embodiment, by analyzing the characteristic text data of each category, true dangerous data and danger coefficients are obtained, so as to obtain the danger index of the corresponding acquisition time node. According to the danger index, the training data within the corresponding time period is screened, so as to construct a dangerous driving behavior prediction model, which can accurately predict or identify whether the driver has dangerous driving behavior, thereby improving road safety and traffic efficiency.

[0097] In some embodiments of the present application, generating the danger index of the current acquisition time node according to the true dangerous data, danger coefficient and weight coefficient of the corresponding category at each acquisition time node includes:

[0098] Obtain the characteristic text data that is not within the preset normal standard text data range and whose difference in characteristic text data is greater than the preset data difference threshold according to the comparison result, and set the corresponding characteristic text data as the pending dangerous data;

[0099] Generate the risk coefficient corresponding to the pending dangerous data according to the deviation degree between the difference in characteristic text data of the pending dangerous data and the preset data difference threshold;

[0100] If the risk coefficient at the last acquisition time node of the pending dangerous data is less than the preset risk coefficient threshold, set the corresponding pending dangerous data as false dangerous data;

[0101] If the risk coefficient at the last acquisition time node of the pending dangerous data is greater than the preset risk coefficient threshold, set the corresponding pending dangerous data as true dangerous data;

[0102] Generate the risk index A of the current acquisition time node according to the true dangerous data at each acquisition time node according to the category, risk coefficient, and the weight coefficient of the corresponding category;

[0103]

[0104] Among them, m0 is the number of categories corresponding to the true dangerous data, Zv,j is the difference in characteristic text data of the vth true dangerous data in the jth category, Z0v,j is the preset data difference threshold of the vth true dangerous data in the jth category, ∣Zv,j - Z0,j∣ is the risk coefficient of the vth true dangerous data in the jth category, bj is the weight coefficient of the jth category, and r,j is the number of true dangerous data in the jth category.

[0105] In this embodiment, calculate the risk index at each acquisition time node through the double summation formula, determine the initial attention time node and the end attention time node within each historical preset period according to the risk index, so as to screen out the data that can accurately represent the driver's dangerous driving behavior, and construct a dangerous driving behavior prediction model.

[0106] In some embodiments of the present application, based on the dangerous driving behavior prediction model, obtain the risk prediction index of the first text data, including:

[0107] Collect the first text data within the preset period, perform preprocessing and standardization processing on the first text data to obtain the first standard text data of different categories, and screen out the first characteristic text data of the first standard text data of different categories;

[0108] Establish a time reference line for the preset period, and set multiple first preset acquisition nodes based on the recognition accuracy;

[0109] Input the first feature text data of multiple categories of the same first preset acquisition node into the dangerous behavior prediction model to obtain the dangerous prediction indexes of multiple first preset acquisition nodes.

[0110] In this embodiment, the preset time period refers to the first 30 minutes before the current driving time period of the driver, and the first text data includes voice data and text data within the preset time period.

[0111] In some embodiments of the present application, judging whether the driver has dangerous driving behavior according to the dangerous prediction index includes:

[0112] Arrange the dangerous prediction indexes at multiple first preset acquisition nodes in the time sequence of the first preset acquisition node to obtain the dangerous prediction index sequence corresponding to the first text data;

[0113] Obtain the dangerous prediction index interval of the first text data and the average growth rate of the dangerous prediction index according to the dangerous prediction index sequence;

[0114] If the dangerous prediction index interval is less than the preset dangerous index threshold or the average growth rate is greater than the preset growth rate threshold, it is determined that the driver has dangerous driving behavior.

[0115] In this embodiment, the dangerous prediction index interval is set according to the maximum dangerous prediction index and the minimum dangerous prediction index in the dangerous prediction index sequence. The average growth rate of the dangerous prediction index is obtained by averaging the growth rates of adjacent dangerous prediction indexes in the dangerous prediction index sequence. The preset growth rate threshold is the maximum growth rate of the dangerous index under the normal driving behavior of the driver.

[0116] In some embodiments of the present application, evaluating the application of the first reminder instruction according to the second text data includes:

[0117] Obtain the second text data, process the second text data to obtain second standard text data of different categories, and screen out the second feature text data of the second standard text data of different categories;

[0118] Input the second feature text data of different categories at the same time node into the dangerous behavior prediction model to obtain the dangerous prediction index at the corresponding time node, and generate the decline rate of the dangerous prediction index of the second text data according to the dangerous prediction indexes at adjacent time nodes;

[0119] Predict the application duration until the dangerous prediction index is less than the preset dangerous prediction index threshold according to the decline rate of the dangerous prediction index of the second text data;

[0120] Generate the application evaluation result of the first reminder instruction according to the application duration.

[0121] In this embodiment, the second text data refers to the real-time voice data and real-time text data after the first reminder instruction. The decline rate refers to the decline speed of the risk prediction index. The preset risk prediction index threshold is the maximum risk prediction index set when the driver does not have dangerous driving behavior. When the decline rate is faster and the application duration is shorter, the application evaluation result of the first reminder instruction is larger. When the decline rate is slower and the application duration is longer, the application evaluation result of the first reminder instruction is smaller.

[0122] In some embodiments of the present application, determining whether to generate a second reminder instruction according to the application evaluation result includes:

[0123] Presetting an application evaluation threshold in advance;

[0124] If the application evaluation result is less than the application evaluation threshold, generate a second reminder instruction;

[0125] If the application evaluation result is greater than the application evaluation threshold, do not generate a second reminder instruction.

[0126] In summary, the present invention provides a method for identifying a driver's dangerous driving behavior based on text data, including: obtaining voice data and text data within multiple historical preset time periods, processing the voice data and text data to obtain standard text data and constructing multiple text data reference sets; analyzing the standard text data in the multiple text data reference sets and screening out characteristic text data, generating a training data set according to the characteristic text data, constructing a dangerous driving behavior prediction model according to the training data set; collecting first text data, obtaining a risk prediction index of the first text data based on the dangerous driving behavior prediction model; judging whether the driver has dangerous driving behavior according to the risk prediction index. If so, generating a first reminder instruction, obtaining second text data, performing an application evaluation on the first reminder instruction according to the second text data, and judging whether to generate a second reminder instruction according to the application evaluation result; by processing the voice data and text data, obtaining characteristic text data of multiple categories, generating a training data set from the characteristic text data of multiple categories in the dangerous driving time period, constructing a dangerous driving behavior prediction model, analyzing the real-time text data according to the dangerous driving behavior prediction model, accurately identifying whether the current driver has dangerous driving behavior, if so, issuing a reminder instruction, and judging whether the risk index drops according to the text data after the reminder instruction, ensuring the high efficiency and accuracy of the identification of the driver's dangerous driving behavior, and improving road safety and traffic efficiency.

[0127] The above are only the preferred embodiments of the present application. It should be noted that for those of ordinary skill in the art, without departing from the technical principle of the present application, several improvements and replacements can be made, and these improvements and replacements should also be regarded as the protection scope of the present application.

Claims

1. A method for identifying dangerous driving behaviors of drivers based on text data, characterized in that: include: Acquire voice data and text data within multiple historical preset time periods, process the voice data and text data, obtain standard text data and construct multiple text data reference sets; Analyze the standard text data in multiple text data reference sets and filter out feature text data, generate a training data set based on the feature text data, and build a dangerous driving behavior prediction model based on the training data set; Collecting first text data, and obtaining a danger prediction index of the first text data based on a dangerous driving behavior prediction model; Determine whether the driver has dangerous driving behavior according to the danger prediction index, and if so, generate a first reminder instruction, obtain second text data, perform application evaluation on the first reminder instruction according to the second text data, and determine whether to generate a second reminder instruction according to the application evaluation result; The speech data and text data are processed to obtain multiple standard text data and construct multiple text data reference sets, including: Based on speech recognition technology, the speech data is converted into text data, and the text data is preprocessed, wherein the preprocessing includes unifying time units, removing outliers, and filling missing values; The preprocessed text data is subjected to standardization processing, wherein the standardization processing includes classifying the preprocessed text data to obtain text data of multiple categories, and standardizing the text data of the same category; Among them, x′ (i) is the i-th text data x in the same category (i) The standard text data, n is the number of data in the same category; Constructing a text data reference set based on different categories of standard text data within the same historical preset period; Analyze the standard text data in multiple text data reference sets and filter out the characteristic text data, including: Construct a standard text data category matrix H for the standard text data in each text data reference set; Among them, xj′ (i) is the i-th standard text data in the j-th category, where j=1, 2, ...m, i=1, 2, ...n; Analyze the standard text data of the same category in the standard text data category matrix H to obtain the fluctuation value and change trend of the standard text data of the same category, generate the characteristic coefficient of the corresponding standard text data according to the fluctuation value and the change trend, and set the standard text data with a characteristic coefficient greater than a preset characteristic coefficient threshold as the characteristic text data in the corresponding category; Construct a feature text data category matrix H based on feature text data in multiple categories in the same text data reference set T ; Among them, x1 T is the characteristic factor in the first category, s1′ (i) is the characteristic coefficient corresponding to the i-th standard text data in the first category, ∝ is the screening formula, where the screening formula is used to screen out the standard text data whose characteristic coefficient is greater than the preset characteristic coefficient threshold, xj T is the characteristic factor in the jth category, sj′ (i) is the characteristic coefficient corresponding to the i-th standard text data in the j-th category, where j = 1, 2, ..., m; Generate a training data set based on the feature text data, and build a dangerous driving behavior prediction model based on the training data set, including: For the same feature text data category matrix H T Establish a time reference line for a historical preset period, and set multiple collection time nodes based on the preset time interval; Acquire characteristic text data of multiple categories at each acquisition time node of the same time reference line, compare the characteristic text data with a preset normal standard text data interval, mark pending dangerous data according to the comparison result, calculate the risk factor of each pending dangerous data, and divide the pending dangerous data into true dangerous data and false dangerous data according to the risk factor; Generate a danger index for the current collection time node based on the true danger data, danger coefficient, and weight coefficient of the corresponding category at each collection time node, and set the initial attention time node and the terminal attention time node within the current historical preset period based on the danger index; intercepting the true danger data between the initial attention time node and the terminal attention time node in each time reference line, and setting the intercepted true danger data as a set of training data; A training data set is constructed by using multiple sets of training data corresponding to multiple time reference lines, and a neural network training is performed to obtain a corresponding dangerous driving behavior prediction model; The hazard index of the current collection time node is generated according to the true hazard data, hazard coefficient and weight coefficient of each collection time node, including: According to the comparison result, characteristic text data that is not in the preset normal standard text data interval and whose characteristic text data difference is greater than the preset data difference threshold is obtained, and the corresponding characteristic text data is set as pending dangerous data; Generate a risk coefficient corresponding to the pending risk data according to the degree of deviation between the characteristic text data difference of the pending risk data and the preset data difference threshold; If the risk factor at the last collection time node of the pending risk data is less than the preset risk factor threshold, the corresponding pending risk data is set as false risk data; If the risk factor at the last collection time node of the pending risk data is greater than the preset risk factor threshold, the corresponding pending risk data is set as true risk data; Generate the danger index A of the current collection time node according to the true danger data at each collection time node according to the category, danger coefficient and the weight coefficient of the corresponding category; Wherein, m0 is the number of categories corresponding to true dangerous data, Zv,j is the feature text data difference of the vth true dangerous data in the jth category, Z0v,j is the preset data difference threshold of the vth true dangerous data in the jth category, |Zv,j-Z0,j| is the danger coefficient of the vth true dangerous data in the jth category, bj is the weight coefficient of the jth category, and r,j is the number of true dangerous data in the jth category; Performing an application evaluation on the first reminder instruction according to the second text data includes: Acquire second text data, process the second text data to obtain second standard text data of different categories, and filter out second characteristic text data of the second standard text data of different categories; Inputting the second feature text data of different categories at the same time node into the dangerous behavior prediction model to obtain the dangerous prediction index at the corresponding time node, and generating the decreasing rate of the dangerous prediction index of the second text data according to the dangerous prediction index at the adjacent time nodes; According to the decreasing rate of the danger prediction index of the second text data, predicting the application time until the danger prediction index is less than a preset danger prediction index threshold; An application evaluation result of the first reminder instruction is generated according to the application duration.

2. The method for identifying dangerous driving behaviors of drivers based on text data as claimed in claim 1, characterized in that: Based on the dangerous driving behavior prediction model, a danger prediction index of the first text data is obtained, including: Collecting first text data within a preset time period, preprocessing and standardizing the first text data to obtain first standard text data of different categories, and screening out first characteristic text data of the first standard text data of different categories; Establishing a time reference line for a preset time period, and setting a plurality of first preset collection nodes based on recognition accuracy; The first characteristic text data of multiple categories of the same first preset collection node are input into the dangerous behavior prediction model to obtain the dangerous prediction indexes of the multiple first preset collection nodes.

3. The method for identifying dangerous driving behaviors of drivers based on text data as claimed in claim 2, characterized in that: The driver is judged to have dangerous driving behavior based on the risk prediction index, including: Arrange the risk prediction indexes at the plurality of first preset collection nodes according to the time sequence of the first preset collection nodes to obtain a risk prediction index sequence corresponding to the first text data; According to the risk prediction index sequence, the risk prediction index interval of the first text data and the mean value of the growth rate of the risk prediction index are obtained; If the danger prediction index interval is less than the preset danger index threshold or the growth rate mean is greater than the preset growth rate threshold, it is determined that the driver has dangerous driving behavior.

4. The method for identifying dangerous driving behaviors of drivers based on text data as claimed in claim 3, characterized in that: Determining whether to generate a second reminder instruction according to the application evaluation result includes: Pre-set application evaluation thresholds; If the application evaluation result is less than the application evaluation threshold, a second reminder instruction is generated; If the application evaluation result is greater than the application evaluation threshold, the second reminder instruction is not generated.

Images