A method for classifying driver portraits of operating vehicles based on vehicle-mounted video recognition

Abstract

The application discloses a kind of based on vehicle video identification's operating vehicle driver portrait classification method, comprising: collecting data, generating active safety early warning information detailed data table;Generate overtime fatigue driving record table;Extract the early warning record corresponding to each driver, generate early warning frequency record table;Extract the key data in early warning video record data, calculate evaluation index;Calculate lane offset;According to evaluation index and lane offset, the reaction degree of driver under different early warning types is described, and the classification result of driver under different early warning types is obtained.The application describes the driving behavior of driver by in-depth extraction of video information, classifies the driver according to driving behavior, and can efficiently and accurately capture the driving characteristics of the driver through real-time video on the vehicle, and the classification of driving behavior and user portrait analysis can provide strong theoretical support for the establishment of the driving safety evaluation system of key operating vehicles.

Description

TECHNICAL FIELD

[0001] The application relates to an operating vehicle driver portrait classification technology, in particular to an operating vehicle driver portrait classification method based on vehicle-mounted video recognition. BACKGROUND

[0002] With the large-scale development of road construction and the transportation industry, large trucks account for a larger proportion on expressways, and the road safety of key operating vehicles has gradually attracted attention. At the same time, research on the driving behavior analysis and personality portrait classification of truck drivers has gradually increased.

[0003] Traditional operating vehicles are not equipped with intelligent vehicle-mounted devices, and cannot track and record vehicle speed, position information, deflection angle and other driving data, so as to accurately classify driving behavior and drivers. The current driver portrait fuzzy classification method mainly uses questionnaires to let drivers evaluate and score driving process, behavior habit and reaction state, and further analyzes the recovered questionnaires, so as to classify the driver personality characteristics and behavior characteristics. However, the method cannot achieve accurate classification. SUMMARY

[0004] The application aims to provide an operating vehicle driver portrait classification method based on vehicle-mounted video recognition, propose different classification discriminators for different active safety warning categories, and effectively solve the difficulty of extracting accurate driving values in the laboratory in the daily driving environment, so as to improve the operating vehicle driver portrait classification standard.

[0005] Technical scheme: the application provides an operating vehicle driver portrait classification method based on vehicle-mounted video recognition, which comprises the following steps:

[0006] Step 1, collect driver personal information data, vehicle parking record data, active safety warning data and warning video record data, and associate the driver personal information data, vehicle parking record data, active safety warning data and warning video record data to generate an active safety warning information detailed data table;

[0007] Step 2, extract the data of the active safety warning information detailed data table with the warning category of overtime fatigue, and generate an overtime fatigue driving record table; extract the corresponding warning record of each driver in the active safety warning information detailed data table, and generate a warning frequency record table; the warning frequency record table includes the trigger frequency of the forward collision warning, lane deviation warning and split fatigue warning of the current driver respectively;

[0008] Step 3, extracting key data in the pre-warning video record data, and calculating evaluation indexes by using the key data; the key data includes license plate number, latitude and longitude, time and speed information; the evaluation indexes include reaction time, speed change before pre-warning and speed change after pre-warning;

[0009] Step 4, calculating lane deviation by using the front-facing video in the pre-warning video record data;

[0010] Step 5, according to the evaluation indexes and the lane deviation, describing the reaction degree of the driver under different pre-warning types, clustering the indexes, and obtaining the classification result of the driver under different pre-warning types; the pre-warning types include forward collision pre-warning, lane deviation pre-warning and distraction fatigue pre-warning.

[0011] Further, in Step 3, extracting key data in the pre-warning video record data, and calculating evaluation indexes by using the key data includes:

[0012] Extracting pictures by frame extraction on the pre-warning video;

[0013] Detecting the generated pictures, extracting the text in the pictures, and generating a pre-warning detailed data table according to the text;

[0014] Obtaining the reaction time, speed change before pre-warning and speed change after pre-warning according to the data in the pre-warning detailed data table.

[0015] Further, Step 4 specifically includes: obtaining a lane deviation picture set by frame extraction on the lane video, performing lane line detection, recognition and fitting on the picture set by using computer vision recognition technology, and calculating the lane deviation change amount of the vehicle per second.

[0016] Further, in Step 5, the driver classification process under forward collision pre-warning is: taking the reaction time of the driver as the measurement standard of the reaction speed of the driver, taking the speed change after pre-warning and the lane deviation as the reaction intensity of the driver after receiving the pre-warning, performing standard normalization on the values of the three indexes, finding the best classification number by using elbow rule and contour coefficient method, and then using K-mean++ for clustering to obtain the classification of the driver under forward collision pre-warning;

[0017] The driver classification process under lane deviation pre-warning is: taking the deviation duration as the measurement standard of the reaction speed of the driver, taking the speed change after pre-warning and the lane deviation as the reaction intensity of the driver after receiving the pre-warning, performing standard normalization on the values of the three indexes, finding the best classification number by using elbow rule and contour coefficient method, and then using K-mean++ for clustering to obtain the classification of the driver under lane deviation pre-warning;

[0018] The driver classification process under the distraction fatigue warning is: taking the distraction fatigue duration of the driver as the measurement standard of the reaction speed of the driver, taking the total time length of the accumulated overtime fatigue driving as the acceptance degree of the driver to the warning, and taking the total warning number of the distraction fatigue warning as the index for the driver to judge the driver violation, after normalizing the values of the three indexes, the elbow rule and the contour coefficient method are used to find the best classification number, and then K-mean++ is used for clustering to obtain the classification of the driver under the distraction fatigue warning.

[0019] Further, the driver personal information data, vehicle parking record data, active safety warning data and warning video record data are associated to generate an active safety warning information detailed data table, specifically:

[0020] The active safety warning data is associated with the warning video record data according to the warning number, the active safety warning data is associated with the driver personal information data according to the driver name, and the active safety warning data is associated with the vehicle parking record data according to the license plate number, and the active safety warning information detailed data table is constructed by using the data after association.

[0021] Further, the overtime fatigue driving record table includes a license plate number, a driver name, a fatigue driving start time, a fatigue driving end time and a duration.

[0022] Advantages: compared with the prior art, the present application has the following advantages:

[0023] 1. The present application extracts millimeter-level driving speed changes and second-level lane deviation amounts by methods such as text recognition and lane line detection of vehicle-mounted video, overcoming the difficulty of obtaining accurate driving data without the help of installed pressure sensing devices in real driving environment;

[0024] 2. The present application extracts effective driver behavior characteristics from a large amount of real driving data, classifies the characteristics, obtains the driving behavior threshold values of different categories of drivers, and the calculated threshold values can be further used for driver driving habit evaluation system, achieving numerical accurate classification that cannot be achieved by questionnaire survey;

[0025] 3. The present application proposes different indexes reflecting the intensity of driver reaction and the speed of reaction for different warning types, avoiding the influence of different warning types on the driver's reaction, such as the driver's reaction after the forward collision warning, which is more reflected in the change of speed, while the driver's reaction after the lane deviation warning is more reflected in the change of lane deviation. BRIEF DESCRIPTION OF DRAWINGS

[0026] Figure 1 is the overall flowchart of the method of the present application;

[0027] Figure 2 is a data information collection structure diagram of the method of the present application;

[0028] Figure 3 is a vehicle-mounted video data display diagram of the method of the present application;

[0029] Figure 4 is a lane deviation detection example display diagram of the method of the present application;

[0030] Figure 5 is a driver classification diagram under forward collision warning of the model of the present application;

[0031] Figure 6 is a driver classification diagram under lane departure warning of the model of the present application;

[0032] Figure 7 is a driver classification diagram under distraction and fatigue warning of the model of the present application. DETAILED DESCRIPTION

[0033] In order to make the purpose, technical scheme and advantages of the present application more clear, the present application is further described in detail below in combination with the drawings and examples.

[0034] The object of the present application is to classify the driving behavior of key operating vehicle drivers, extract effective data reflecting driving behavior from vehicle-mounted forward video and driver facial video, propose classification determination indexes under each warning category (forward collision warning, lane deviation warning and distraction and fatigue warning) combined with the above data, use these indexes to perform K-means++ algorithm clustering on the driving habits of drivers under different warning types, and obtain behavior classification and discrimination threshold data of operating vehicle drivers.

[0035] The data used in this embodiment are provided by a certain dangerous goods transportation company in Nanjing, and include driving and staying data of GPS network type key operating vehicles equipped with active safety warning systems, driver information data, vehicle basic data, vehicle-mounted forward camera video and in-vehicle driver facial video data within 4 months from December 2021 to March 2022.

[0036] As Figure 1 , the operating vehicle driver portrait classification method based on vehicle-mounted video recognition described in this embodiment includes the following steps:

[0037] Step 1, collect driver personal information data, vehicle parking record data, active safety warning data and warning video record data, associate the driver personal information data, vehicle parking record data, active safety warning data and warning video record data, and generate an active safety warning information detailed data table.

[0038] As Figure 2As shown, the driver personal information data includes: name, age and gender. The vehicle parking record data includes: license plate number, driver name, parking start time and end time. The active safety warning data includes: warning number, warning category, license plate number, driver name, warning start time and warning duration. The warning video record data is the video data of 5 seconds before and after the warning start time.

[0039] The active safety warning data invalid values are processed, the null data in the active safety warning data is deleted in the whole row, the data with a warning duration greater than 10 seconds is treated as an abnormal value, and the row is deleted. The active safety warning information detailed data table is established by using the processed active safety warning data.

[0040] The active safety warning data is associated with the warning video record data according to the warning number, associated with the driver personal information data according to the driver name, and associated with the vehicle parking record data according to the license plate number. The active safety warning information detailed data table is constructed by using the data after association.

[0041] Step 2, extracting the data of the active safety warning information detailed data table with the warning category of overtime fatigue, generating an overtime fatigue driving record table; extracting the corresponding warning record of each driver in the active safety warning information detailed data table, generating a warning frequency record table; the warning frequency record table includes the trigger frequency of the forward collision warning, lane deviation warning and split fatigue warning of the current driver respectively.

[0042] The overtime fatigue driving record table includes license plate number, driver name, fatigue driving start time, fatigue driving end time and duration. The driving that meets the following conditions is defined as overtime fatigue driving: (1) more than 3.5 hours of continuous driving between 5 am and 22 pm enters fatigue time; (2) more than 1.5 hours of continuous driving between 22 pm and 5 am the next morning enters fatigue time. According to the regulations, the driver needs to stop and rest after entering the fatigue driving time, but there are still violations due to subjective willingness not to stop or objective reasons that cannot stop during the actual driving process. Since the frequency of driver warning triggers increases during the fatigue driving period, an overtime fatigue driving record table needs to be generated for subsequent analysis. The specific implementation method is to compare the parking start and end times in the vehicle parking record data table with the warning start time extracted from the overtime fatigue data, calculate from the warning trigger time until the time when the vehicle starts to stop and rest, and record it in the overtime fatigue driving record table. For example, the driver receives a prompt at 20 pm that he has been driving for 3.5 hours and enters the overtime fatigue driving period. If the driver cannot stop at this time while driving on the highway, he finally rests at the highway service area at 20:30. It is considered that the 30 minutes from 20 pm to 20:30 are fatigue driving time. The extracted overtime fatigue driving record is shown in Table 1.

[0043] Table 1: Overtime fatigue driving record

[0044] License plate number ID Driver ID Start time End time Duration 310230 927 2022 / 03 / 20 23:30:07 2022 / 03 / 20 23:35:36 329 seconds 310229 935 2022 / 03 / 20 19:52:09 2022 / 03 / 20 19:55:18 189 seconds 310227 930 2022 / 03 / 20 19:31:24 2022 / 03 / 20 19:31:36 12 seconds … … … … …

[0045] For each driver, the corresponding warning records are extracted in the active safety warning information detailed data table according to the name, and the corresponding trigger times are calculated separately according to the forward collision warning, lane deviation warning and distraction fatigue warning. The above data is recorded in the warning number record table. The extracted warning number record is shown in Table 2.

[0046] Table 2: Warning number record

[0047]

[0048] Step 3, extract key data in warning video record data, and calculate evaluation index using key data; the key data includes license plate number, latitude and longitude, time (millisecond level) and speed information; the evaluation index includes reaction time, speed change before warning and speed change after warning.

[0049] Extracting key data in warning video record data and calculating evaluation index using key data includes:

[0050] Frame extraction is performed on the warning video to extract pictures;

[0051] Detect the generated picture, extract the text in the picture, and generate a warning detailed data table according to the text;

[0052] According to the data in the warning detailed data table, the reaction time, the speed change before the warning, and the speed change after the warning are obtained.

[0053] The current video data length is fixed at 10 seconds, and each 5 seconds before and after the warning trigger. In order to reduce the complexity of text recognition, the video is extracted to obtain pictures. After testing, there are 21 pictures per second in the vehicle recording video, and a frequency of every 4 pictures can obtain a picture set with an interval of 200 milliseconds. The generated picture set is detected to obtain an image specification of 1280*720 pixels. According to the position of the text information in the picture, the coordinates of the to-be-recognized area are found. According to Figure 3 As shown in the figure, the to-be-recognized area content includes license plate number, date and time, speed, longitude and latitude. The recognition method used in this embodiment is to call Tesseract-OCR package under Python environment, to extract information in the picture set in the corresponding to-be-recognized area coordinate range through the tool package, and to generate a warning detailed data table, which includes: warning number, license plate number, date and time, speed and longitude and latitude. The generated warning detailed data table after content recognition is preprocessed: (1) forward filling processing is performed on all speed null data; (2) for any two data with speed difference greater than 5KM / H, the default is to recognize error and make the speed value of the latter data the same as the former one; (3) delete the whole row operation for the data with empty warning number. Based on the extracted warning detailed data table, the reaction time, the speed change before the warning, and the speed change after the warning are extracted as part of the judgment standard for later driver classification. The reaction time is the difference between the warning trigger time and the time when the speed change occurs after the warning. The calculation of the speed change before the warning needs to consider different situations, regardless of whether the speed before the warning changes multiple times, only the change trend closest to the warning trigger time is taken. For example, the speed is continuously accelerated from 5 seconds before the warning to 3 seconds before the warning, and continuously decelerated from 3 seconds before the warning to the warning trigger time, then it is considered that the driver decelerates before the warning, and the speed difference between 3 seconds before the warning and the warning trigger time is taken as the speed change before the warning. The speed change after the warning is similar to the speed change before the warning.

[0054] Step 4, calculate the lane offset using the vehicle front video in the warning video recording data.

[0055] The lane video is extracted to obtain a lane deviation picture set, and computer vision recognition technology is used to detect, recognize and fit the lane lines of the picture set to calculate the lane offset change per second of the vehicle.

[0056] As Figure 4As shown, video frame extraction and image preprocessing: the lane video is extracted again, and because the lane deviation distance changes little in millisecond level time, the frequency of extracting every 21 frames is adopted to obtain a picture set with an interval of 1 second. By introducing HSL color space and Sobel edge extraction algorithm, irrelevant colors are suppressed and white lane line edges are detected and extracted, the influence of irrelevant areas on recognition is reduced by specifying the lane line detection area, and then the image in the detection area is perspective transformed to convert the image into a binary black and white overhead view, and in the new two-dimensional plane, two columns with dense white pixels (lane lines) in the overhead view are detected, which are the approximate positions of the left and right lane lines.

[0057] Lane line positioning and fitting: by extracting the approximate horizontal coordinates of the left and right lane lines, the sliding window method is used to search the grid near the horizontal coordinate points to determine the horizontal and vertical coordinate points of the left and right lane lines, and the polyfit function in the numpy package under the Python environment is used to fit the horizontal and vertical coordinate points of the left and right lane lines with a second-order polynomial, and the function expressions of the left and right lane lines are as follows:

[0058] Left lane:

[0059] Right lane:

[0060] Where a l , b l , c l are the second-order polynomial coefficients of the left lane after fitting, obtained by the polyfit function; a r , b r , c r are the second-order polynomial coefficients of the right lane after fitting, obtained by the polyfit function.

[0061] Calculate the lane deviation distance: the lane deviation distance is determined by the lane center line position and the pixel difference of the image center line, and the expression is:

[0062] d = d v - d c , if

[0063] Where the image center line distance d c is expressed as:

[0064]

[0065] In the formula, w is the image width, which is 1280 pixels in this embodiment; u is the length of a unit pixel, in meters.

[0066] The lane center line position distance d v is expressed as:

[0067]

[0068] where y l,imax is the left lane line position on horizontal coordinate, y r,imax is the right lane line position on horizontal coordinate.

[0069] Step 5, according to the evaluation index and lane deviation, the reaction degree of the driver under different warning types is described, the index is clustered, and the classification result of the driver under different warning types is obtained; the warning types include forward collision warning, lane deviation warning and distraction fatigue warning.

[0070] The driver classification process under forward collision warning is as follows: taking the reaction time of the driver as the measurement standard of the reaction speed of the driver, taking the speed change after warning and lane deviation to represent the reaction intensity of the driver after receiving the warning, after standard normalization of the three index values, the elbow rule and contour coefficient method are used to find the best classification number, and then K-mean++ is used for clustering to obtain the classification of the driver under forward collision warning. The index threshold of each type of driver after clustering is shown in Table 3.

[0071] Table 3 Index threshold of forward collision warning driver

[0072] Average reaction time Average speed change amount Average lateral deviation amount Aggressive 1.83s 2 m / s 3.44m Predictive 0.69s 4.32 m / s 1.57m Calm 1.7s -3.5 m / s 1.5m

[0073] In this example, the forward collision warning driver classification is as shown in Table 3, which is divided into aggressive type, foresight type and calm type. The aggressive type driver has a longer average reaction time after warning prompt, a large average lateral deviation, and no obvious speed change, indicating that the aggressive type driver tends to follow the front vehicle and find an overtaking opportunity to avoid collision by lane changing; the foresight type driver has the shortest average reaction time, smaller lateral deviation, and slight speed change, indicating that this type of driver perceives the risk in advance and prepares for response before the warning prompt, so the alarm is removed in a short time; the calm type driver has a longer average reaction time, the smallest lateral deviation, and a significant speed change, which indicates that this type of driver will observe the driving environment and make a reasonable response after receiving the warning signal, and the average lateral deviation is extremely small during this process, without lane changing behavior. Figure 5

[0074] ​The driver classification process under lane departure warning is as follows: the duration of lane departure is used as a measure of the driver's reaction speed, and the speed change and lane deviation after the warning represent the intensity of the driver's reaction after receiving the warning. After standardizing the three indicator values, the elbow rule and silhouette coefficient method are used to find the optimal number of classifications, and then K-means++ is used for clustering to obtain the driver classification under lane departure warning. The indicator thresholds for each type of driver after clustering are shown in Table 4 below.

[0075] Table 4. Thresholds for Lane Departure Warning Drivers

[0076] Average duration Average speed change amount Average lateral deviation amount Aggressive 1.9s 0.47 m / s 2.76m Predictive 0.96s 2.92 m / s 1.54m Mild 1.49s -0.59 m / s 1.35m Calm 2.42s -1.08 m / s 1.49m

[0077] In this example, the lane departure warning driver classification is as follows: Figure 6 As shown, drivers are categorized into four types: aggressive, anticipatory, mild, and calm. Aggressive drivers exhibit a large lateral drift, a moderate warning duration, and no significant speed change, similar to aggressive drivers in forward collisions, typically changing lanes instead of slowing down to avoid the warning. Anticipatory drivers have the shortest warning duration, a small lateral drift, and a large average speed change, indicating that they are aware of the drift before the warning and quickly return to the center of the road. Mild drivers have a moderate to small drift duration and a moderate to small average lane departure, with no significant speed change, indicating that their driving behavior is typical and mild, consistent with normal driver responses to warnings. Calm drivers have the longest drift duration, a moderate lateral drift, and a significant deceleration change, similar to the forward collision warning classification, and will further decide how to respond to the warning based on their own judgment.

[0078] The driver classification process under distraction and fatigue warning is as follows: the duration of driver distraction and fatigue is used as a measure of driver reaction speed; the total accumulated time spent driving while fatigued is used as a measure of driver acceptance of the warning; and the total number of distraction and fatigue warnings is used as an indicator of driver's judgment of violation. After normalizing the values ​​of the three indicators, the elbow rule and silhouette coefficient method are used to find the optimal number of classifications. Then, K-means++ is used for clustering to obtain the driver classification under distraction and fatigue warning. The indicator thresholds for each type of driver after clustering are shown in Table 5 below.

[0079] Table 5. Thresholds for Distraction and Fatigue Warnings for Drivers

[0080] Average duration Average fatigue time per kilometer Average number of fatigue times per kilometer Easily fatigued 3.9s 20 min 156 times Mild 3.6s 15 min 72 times Cautious 3s 17 min 64 times

[0081] In this example, the driver classification for distraction and fatigue warning is as follows: Figure 7As shown, the fatigue type is divided into fatigue type, moderate type and cautious type. The fatigue type driver has the most fatigue times per kilometer, the fatigue driving time per kilometer is slightly higher than that of other types, and the warning duration is long, indicating that the driver is likely to be tired during driving and is used to continuous fatigue driving. The moderate type driver has a long warning duration, fewer fatigue times (times per kilometer), and a small fatigue driving time (minutes per kilometer). The driver of this type conforms to the reaction of the conventional fatigue driving driver and tends to obey the warning prompt and stop for rest. The cautious type driver has the shortest warning duration, indicating that the driver can quickly respond and enter the alarm state after the system prompts fatigue. The smallest fatigue times (times per kilometer) and the small fatigue driving time (minutes per kilometer) also indicate that the driver has cautious driving habits.

[0082] The present application can intuitively depict the driving behavior of the driver by deeply extracting the video information, and classify the driver according to the driving behavior to find the problems existing in the driving process of the driver. Through the real-time video on the vehicle, the driving characteristics of the driver can be efficiently and accurately captured. Compared with the questionnaire survey method, this kind of analysis method can statistically analyze a large amount of driving data, provide more real and reliable digital proof, and classify the driving behavior and analyze the user portrait, which can provide strong theoretical support for the establishment of the driving safety evaluation system of the key operation vehicle.

Claims

1. A method for classifying the portrait of an operating vehicle driver based on vehicle video recognition, characterized in that, The method comprises the following steps: Step 1, collecting driver personal information data, vehicle parking record data, active safety warning data and warning video record data, associating the driver personal information data, the vehicle parking record data, the active safety warning data and the warning video record data to generate an active safety warning information detailed data table; Step 2, extracting data with an overtime fatigue warning category in the active safety warning information detailed data table to generate an overtime fatigue driving record table; extracting the warning records corresponding to each driver in the active safety warning information detailed data table to generate a warning frequency record table; the warning frequency record table includes the trigger frequencies of forward collision warnings, lane deviation warnings and distraction fatigue warnings of the current driver; Step 3, extracting key data in the warning video record data and calculating evaluation indexes by using the key data; the key data includes license plate numbers, latitude and longitude, time and speed information; the evaluation indexes include reaction times, speed changes before and after warnings and lane deviation amounts; Step 4, calculating lane deviation amounts by using the front-facing vehicle video in the warning video record data; Step 5, describing the reaction degrees of the driver under different warning types according to the evaluation indexes and the lane deviation amounts, clustering the indexes and obtaining the classification results of the driver under different warning types; the warning types include forward collision warnings, lane deviation warnings and distraction fatigue warnings.

2. The method of claim 1, wherein In step 3, the key data in the warning video record data is extracted, and the evaluation indexes are calculated by using the key data, which comprises: Frame extraction is performed on the warning video to extract pictures; The generated pictures are detected to extract the text in the pictures, and the warning detailed data table is generated according to the text; The reaction times, the speed changes before and after warnings and the lane deviation amounts are obtained according to the data in the warning detailed data table.

3. The method of claim 1, wherein Step 4 specifically comprises: frame extraction is performed on the lane video to obtain a lane deviation picture set, lane line detection, recognition and fitting are performed on the picture set by using computer vision recognition technology, and the lane deviation change amount of the vehicle per second is calculated.

4. The method of claim 1, wherein In step 5, the driver classification process under forward collision warnings is as follows: the reaction time of the driver is taken as a measurement standard of the reaction speed of the driver, the speed change after the warning and the lane deviation amount are taken to represent the reaction intensity of the driver after receiving the warning, the three index values are normalized, the elbow rule and the contour coefficient method are used to find the best classification number, K-mean++ is used for clustering, and the classification of the driver under forward collision warnings is obtained; The driver classification process under lane deviation warnings is as follows: the deviation duration is taken as a measurement standard of the reaction speed of the driver, the speed change after the warning and the lane deviation amount are taken to represent the reaction intensity of the driver after receiving the warning, the three index values are normalized, the elbow rule and the contour coefficient method are used to find the best classification number, K-mean++ is used for clustering, and the classification of the driver under lane deviation warnings is obtained; The driver classification process under the distraction fatigue warning is: taking the driver distraction fatigue duration as the measurement standard of the driver reaction speed, taking the total time length of the accumulated overtime fatigue driving as the acceptance degree of the driver to the warning, taking the accumulated total warning number of the distraction fatigue warning as the index of the driver judging the driver violation, normalizing the three index value standards, using the elbow rule and the contour coefficient method to find the best classification number, and then using K-mean++ to cluster to obtain the classification of the driver under the distraction fatigue warning.

5. The method of claim 1, wherein The driver personal information data, the vehicle parking record data, the active safety warning data and the warning video record data are associated to generate the active safety warning information detailed data table, specifically: The active safety warning data is associated with the warning video record data according to the warning number, the active safety warning data is associated with the driver personal information data according to the driver name, the active safety warning data is associated with the vehicle parking record data according to the license plate number, and the active safety warning information detailed data table is constructed by using the data after the association.

6. The method of claim 1, wherein The overtime fatigue driving record table includes the license plate number, the driver name, the fatigue driving start time, the fatigue driving end time and the duration.

Images