A driver abnormal gear shifting recognition method based on vehicle GPS data

Abstract

This invention discloses a method for identifying abnormal driver speed changes based on vehicle GPS data. First, the pre-acquired vehicle GPS data is preprocessed. Second, discrete speed change thresholds are calculated by dividing the GPS data into bins using vehicle speed and acceleration as key indicators, and determining the discrete acceleration threshold for each bin using the characteristic line method and elbow diagram method. Then, dynamic speed change thresholds are fitted by fitting the discrete thresholds using a modified least squares method to obtain the dynamic threshold for abnormal driver speed changes, and a mathematical equation for this threshold is derived. Finally, abnormal speed change behavior is identified by substituting the dynamic threshold equation into historical or real-time data. This invention does not rely on complex environmental parameters such as road grade, weather conditions, or traffic status, and possesses real-time performance, accuracy, and adaptability. It is applicable to driver abnormal speed change identification in various environments and can be practically applied in pre-emptive driving prevention and proactive management platforms.

Description

Technical Field

[0001] This invention belongs to the field of urban traffic safety technology, and specifically relates to a method for identifying abnormal driver speed changes based on vehicle GPS data. Background Technology

[0002] According to data from the World Health Organization's "Global Status Report on Road Safety," approximately 1.35 million people die in road traffic accidents globally each year. Among these, accidents caused by sudden driver malfunctions or driver interference are on the rise. Therefore, monitoring the vehicle's driving status during operation and issuing warnings or initiating emergency braking measures when abnormal conditions occur is of paramount importance to ensure the safety of the driver and passengers.

[0003] Current driver safety detection technologies are mostly based on monitoring one or more of the driver's control activities (such as acceleration, braking, and steering) to detect the vehicle's status. However, with the rapid development of data science and data mining technologies, the evaluation of driver safety behavior is no longer limited to real-time monitoring methods, but is increasingly focused on objectively and quantitatively evaluating the driver's safety attributes through massive amounts of heterogeneous data on driver behavior.

[0004] Driver factors are the most critical factor in traffic safety. Scientifically and objectively identifying abnormal driver shifting behavior, thereby fairly assessing drivers' safe driving skills and improving proactive risk control capabilities, is of great significance for enhancing the overall safety of public transportation. Therefore, this invention designs a method for identifying abnormal driver shifting behavior based on vehicle GPS data. Summary of the Invention

[0005] Purpose of the invention: In order to improve the level of driving safety evaluation, this invention proposes a method for identifying abnormal speed changes by drivers based on vehicle GPS data. Based on vehicle GPS data, a dynamic threshold equation for safe speed changes by drivers is fitted, thereby facilitating the identification of abnormal speed change behavior by drivers. The identification results can objectively assess the driver's safe driving level.

[0006] Technical Solution: This invention provides a method for identifying abnormal driver gear shifts based on vehicle GPS data, specifically including the following steps:

[0007] (1) Preprocess the previously acquired vehicle GPS data;

[0008] (2) Discrete speed threshold solution: Taking vehicle speed and acceleration as key indicators, the speed of GPS data is divided into bins, and the discrete acceleration threshold of each bin is determined by the characteristic line method and elbow diagram method.

[0009] (3) Dynamic speed change threshold fitting: The dynamic threshold of abnormal speed change of the driver is obtained by fitting the discrete threshold through the improved least squares method, and the mathematical equation of the threshold is obtained.

[0010] (4) Identification of abnormal speed change behavior: The dynamic threshold equation is input into historical or real-time data to identify abnormal speed change behavior.

[0011] Furthermore, the vehicle GPS data in step (1) includes a total of 6 fields, namely license plate number, GPS time, longitude, latitude, direction and speed; after data cleaning, data interpolation, vehicle turning angle and acceleration calculation, the vehicle GPS data includes two fields: vehicle speed and acceleration.

[0012] Furthermore, step (2) includes the following steps:

[0013] (21) Speed ​​binning: The speed range of the sample data is binned with a binning width of 1 km / h. The number of bins obtained at this time is more appropriate.

[0014] (22) Determine the critical acceleration value for each sub-box:

[0015] Take the a to b quantiles of acceleration in each speed compartment as feature points. Connect the quantile points of each level in sequence to obtain multiple feature lines, which can be used as candidate positions for the safe speed change threshold line.

[0016] The spacing between adjacent quantile characteristic lines reflects the absolute change in acceleration at different characteristic lines. Since the spacing between adjacent characteristic lines changes with velocity, it cannot directly reflect the degree of acceleration jump. Therefore, each characteristic value is standardized.

[0017]

[0018] Where Q(i) represents the vector composed of the acceleration values ​​of the i-th quantile feature point; Q(u) represents the vector composed of the acceleration values ​​of the highest quantile; Q(l) represents the vector composed of the acceleration values ​​of the lowest quantile; and Q'(i) represents the vector composed of the acceleration values ​​of the i-th quantile feature point after standardization.

[0019] The spacing between adjacent standardized feature lines reflects the relative change in acceleration of samples at adjacent quantiles. Considering that the safe acceleration threshold may change in different velocity ranges, let the velocity range R be the safe acceleration range.

[0020] For relatively stable regions at the threshold position, calculate the distance between adjacent standardized feature lines and denote it as the standardized neighbor difference SND. The specific processing formula is as follows:

[0021]

[0022] Where, d R (i) represents the SND index at the i-quantile characteristic line within the velocity range R; This represents the mean acceleration of the i-quantile characteristic line within a specific velocity range R;

[0023] (23) Using the elbow diagram method, arrange the i-quantile feature lines SND in sequence and draw a line graph to show the transition points of the acceleration value. The quantile feature line at the inflection point in the rising part of the line graph is the location of the safe speed change threshold.

[0024] Furthermore, step (3) includes the following steps:

[0025] (31) Determining the reliability of the safe speed change threshold:

[0026] Find a bin with a large sample size. Randomly select n samples from the population of this bin as "subsamples" to calculate the safe speed change threshold. Repeat this process m times and calculate the mean of the safe speed change threshold, denoted as a0. Each time, increase the number of samples by l and repeat the calculation of the mean of the safe speed change threshold until the subsample size is large enough. The mean of the safe speed change threshold calculated in the kth iteration is denoted as a0. k The final convergence value is denoted as a k Corresponding credibility estimate The specific calculation formula is as follows:

[0027]

[0028] Construct a confidence function that reflects the actual changes in confidence level, and according to the definition of confidence level, the function must satisfy... Given two conditions, the final confidence function is:

[0029]

[0030] In the formula, x is the sample size; p(x) is the confidence level of the safe shift threshold when the sample size is x; C1 and C2 are undetermined constants that need to be determined by fitting actual data.

[0031] (32) Fitting algorithm considering the reliability of sampling points:

[0032] Let (x, y) be a pair of observations, x = [x1] , x2,…,x n ] T And it satisfies the following: y = f(x, w), where w = [w1, w2, ..., w n ] T These are parameters to be determined.

[0033] Based on the original loss function, the reliability of the sampling points is considered, and the reliability of the sample size in each bin is calculated to obtain the reliability sequence p = [p1, p2, ..., p m ] T For a given set of m sets (m>n) of observation data (x) i ,y i Minimize the loss function:

[0034]

[0035] The final threshold curve equation obtained by fitting is:

[0036] Furthermore, step (4) is implemented as follows:

[0037] Substitute the velocity x and acceleration y from historical or real-time data points into the threshold equation. like If the data indicates abnormal speed change behavior, then the abnormal speed change behavior cannot be explained.

[0038] Beneficial effects: Compared with the prior art, the beneficial effects of the present invention are as follows: The present invention plans and fits the safe limits of driver speed change based on the actual driving data of the driver, without relying on complex environmental parameters such as road grade, weather conditions, and traffic conditions. It has real-time performance, accuracy, and adaptability, and is suitable for identifying abnormal speed change of drivers in various environments. The present invention uses data mining based on GPS data, which can effectively identify abnormal speed change behavior of drivers, realize the advance judgment of driving risks, and play a positive role in improving vehicle acceleration safety. Attached Figure Description

[0039] Figure 1 This is a flowchart of the present invention;

[0040] Figure 2 This is a velocity and acceleration graph of the sample data of this invention;

[0041] Figure 3 This is a standardized feature line diagram of the acceleration of feature points after processing according to the present invention;

[0042] Figure 4 This is a basic position diagram of the safe speed change threshold obtained by the elbow diagram method in this invention;

[0043] Figure 5 This is the confidence function curve constructed in this invention;

[0044] Figure 6 This is a confidence curve of the data for each bin in this invention;

[0045] Figure 7This is a curve fitting diagram of the deceleration threshold of the present invention; wherein, (a) is the fitting result of the first-order polynomial, with an MSE of 1.34; (b) is the fitting result of the second-order polynomial, with an MSE of 0.28; (c) is the fitting result of the third-order first-order polynomial, with an MSE of 0.27; and (d) is the fitting result of the fourth-order polynomial, with an MSE of 0.007.

[0046] Figure 8 These are correlation diagrams between abnormal acceleration behavior and actual alarm data; among them, (a) is a correlation diagram between abnormal acceleration behavior and actual alarm data under lane departure conditions; and (b) is a correlation diagram between abnormal acceleration behavior and actual alarm data under forward collision conditions.

[0047] Figure 9 This is a correlation diagram between abnormal deceleration behavior and actual alarm data; where (a) is the correlation between lane departure abnormal deceleration behavior and actual alarm; and (b) is the correlation between forward collision abnormal deceleration behavior and actual alarm. Detailed Implementation

[0048] The present invention will now be described in further detail with reference to the accompanying drawings.

[0049] like Figure 1 As shown, this invention proposes a method for identifying abnormal driver gear shifts based on vehicle GPS data, comprising the following steps:

[0050] Step 1: Preprocess the pre-acquired vehicle GPS data.

[0051] Vehicle GPS data includes six fields: license plate number, GPS time, longitude, latitude, direction, and speed. After data cleaning, data interpolation, and calculation of vehicle turning angle and acceleration, the vehicle GPS data must include two fields: vehicle speed and acceleration (hereinafter referred to as GPS data). The field details are shown in Table 1.

[0052] Table 1. Detailed list of vehicle GPS data fields

[0053]

[0054] Step 2: Discrete speed threshold solution: Taking vehicle speed and acceleration as key indicators, the speed of GPS data is divided into bins, and the discrete acceleration threshold of each bin is determined by the method of characteristics and elbow diagram method.

[0055] First, discretize the velocity, such as Figure 2The sample data is binned according to its velocity range. It's important to note that the number of bins should not be too large, otherwise the amount of data in each bin will be insufficient to obtain reliable acceleration critical values; conversely, too few bins will significantly reduce the continuity and accuracy of the model. This invention uses a bin width of 1 km / h for binning, which yields a suitable number of bins.

[0056] Next, the critical acceleration values ​​for each gearbox are determined. The a to b quantiles of acceleration within each speed gearbox are taken as characteristic points, where a can be 80 km / h and b can be 99 km / h. Connecting these quantiles sequentially yields multiple characteristic lines, which are used as candidate locations for the safe shift threshold lines.

[0057] The spacing between adjacent quantile characteristic lines reflects the absolute change in acceleration at different characteristic lines. Since the spacing between adjacent characteristic lines changes with the velocity, it cannot intuitively reflect the degree of acceleration value transition. Therefore, each characteristic value should be standardized using formula (1) first.

[0058]

[0059] Where Q(i) represents the vector composed of the acceleration values ​​of the i-th quantile feature point; Q(u) represents the vector composed of the acceleration values ​​of the highest quantile; Q(l) represents the vector composed of the acceleration values ​​of the lowest quantile; and Q'(i) represents the vector composed of the acceleration values ​​of the i-th quantile feature point after standardization. The standardized feature line is obtained by calculating the standardized value of the feature point acceleration according to formula (1).

[0060] like Figure 3 As shown, when the speed is below 35 km / h, the characteristic line between the maximum and minimum quantile characteristic lines is chaotic, indicating that this is the concentrated distribution range of acceleration and cannot be used as a basis for examining abnormal acceleration behavior. When the speed exceeds 65 km / h, the characteristic line fluctuates significantly, mainly due to the reduced data volume. When the speed exceeds 35 km / h but is less than 65 km / h, the intervals between the characteristic lines are distinct, indicating obvious differences between them. Therefore, the analysis should focus on the characteristic line for speeds exceeding 35 km / h but less than 65 km / h.

[0061] The distance between adjacent standardized feature lines can reflect the relative change in acceleration of adjacent quantile samples. Considering that the position of the safe speed change threshold may change under different speed ranges, let the speed range R be the relatively stable area of ​​the safe speed change threshold position. The distance between adjacent standardized feature lines is calculated using formula (2) and denoted as "standardized neighbor difference (SND)".

[0062]

[0063] Where, dR (i) represents the SND at the i-quantile characteristic line within the velocity interval R; This represents the mean acceleration of the i-quantile characteristic line within a specific velocity range R.

[0064] Finally, using the elbow plot method, the i-quantile characteristic lines SND are arranged sequentially to create a line graph representing the transition points of acceleration values. The quantile characteristic lines at the inflection points in the rising section of this line graph represent the locations of the safe acceleration threshold. Figure 4 As shown.

[0065] Step 3: Dynamic speed change threshold fitting: The dynamic threshold of abnormal speed change of the driver is obtained by fitting the discrete threshold using the improved least squares method protected by this invention, and the mathematical equation of the threshold is obtained.

[0066] First, to determine the reliability of the safe shift threshold, we first identify a bin with a large sample size (based on the Bootstrap method, the number of samples in the bin should exceed 8000). We then randomly select n samples from the population of this bin as "subsamples" to calculate the safe shift threshold. This step is repeated m times, and the mean of the safe shift threshold is calculated, denoted as a0. Each time, we increase the number of samples by l and repeat the above calculation steps until the subsample size is sufficiently large. The mean of the safe shift threshold calculated in the k-th iteration is denoted as a0. k The final convergence value is denoted as Finally, a is calculated using formula (3). k Corresponding credibility estimate

[0067]

[0068] Due to the influence of random sampling error, the confidence estimate has a bias. Therefore, it is necessary to construct a confidence function. This function needs to meet the actual situation of confidence change. According to the definition of confidence, this function needs to meet the two conditions of formula 4. Finally, the constructed confidence function is shown in formula (5).

[0069]

[0070]

[0071] In the formula, x is the sample size; p(x) is the confidence level of the safe shift threshold when the sample size is x; C1 and C2 are undetermined constants that need to be determined by fitting actual data. The confidence curves obtained according to formulas (2)-(5) are as follows: Figure 5 As shown, the confidence level of each bin's data is then calculated using a confidence function, as follows: Figure 6 As shown.

[0072] Next, let (x, y) be a pair of observations, x = [x1, x2, ..., xy]. n ] T And it satisfies the following functional expression:

[0073] y = f(x, w), where w = [w1, w2, ..., w n ] T These are parameters to be determined.

[0074] Traditional least squares treats all data points equally, while the improved method considers the reliability of the sampling points based on the original loss function. The reliability sequence p = [p1, p2, ..., p] is calculated by substituting the sample size of each bin into formula (5). m ] T To find the optimal estimate of parameter w, given m sets (m>n) of observation data (x... i ,y i Minimize the loss function:

[0075]

[0076] Then, the deceleration threshold curves were fitted using polynomials of orders 1 to 4, respectively, as follows: Figure 7 As shown, MSE is used as the evaluation index for goodness of fit. When M=4, the polynomial fit is the best, with an MSE of 0.07. The polynomial expression of the final threshold curve obtained by fitting is:

[0077] a tv = -5.45e -07 v 4 +7.71e -05 v 3 -3.18e -03 v 2 -2.07e -02 v+4.00, MSE is 0.07.

[0078] Step 4: Identify abnormal speed change behavior. Substitute the dynamic threshold equation described in Step 3 into historical or real-time data to identify abnormal speed change behavior.

[0079] The Spearman model was used to examine the correlation between the number of abnormal acceleration and deceleration behaviors of 27 bus drivers each day and the number of alarms from their onboard terminals on that day. The analysis results are as follows: Figure 8 and Figure 9 As shown, Figure 8 (a) is a correlation diagram between abnormal acceleration behavior and actual alarm data under lane departure conditions; Figure 8 (b) Correlation diagram between abnormal acceleration behavior and actual alarm data in the case of forward collision; Figure 9(a) Correlation between lane departure abnormal deceleration behavior and actual alarm. Figure 9 (b) Correlation between abnormal deceleration behavior and actual alarms in the face of forward collision. Both abnormal acceleration and abnormal deceleration behaviors were significantly correlated with lane departure and forward collision (P<0.01). This indicates that the bus driver abnormal behavior identification model has a good ability to assess objective behavioral risks, which indirectly confirms the effectiveness and applicability of the present invention.

Claims

1. A driver abnormal gear shift recognition method based on vehicle GPS data, characterized in that, It includes the following steps: (1) Preprocess the pre-acquired vehicle GPS data; (2) Discrete speed threshold solution: Taking vehicle speed and acceleration as key indicators, the speed of GPS data is divided into bins, and the discrete acceleration threshold of each bin is determined by the characteristic line method and elbow diagram method. (3) Dynamic speed change threshold fitting: The dynamic threshold of abnormal speed change of the driver is obtained by fitting the discrete threshold through the improved least squares method, and the mathematical equation of the threshold is obtained. (4) Identification of abnormal speed change behavior: The dynamic threshold equation is input into historical or real-time data to identify abnormal speed change behavior. Step (2) includes the following steps: (21) Speed ​​binning: The speed range of the sample data is binned with a binning width of 1 km / h. The number of bins obtained at this time is more appropriate. (22) Determine the critical acceleration value for each sub-box: The acceleration in each speed bin is taken as a feature point a to b The quantile is taken as a feature point, and the feature lines are obtained by connecting the quantile points in sequence, which are used as the candidate positions of the safety gear shifting threshold lines. The spacing between adjacent quantile characteristic lines reflects the absolute change in acceleration at different characteristic lines. Since the spacing between adjacent characteristic lines changes with velocity, it cannot directly reflect the degree of acceleration jump. Therefore, each characteristic value is standardized. in, Q ( i ) indicates the first i A vector composed of the acceleration values ​​of quantile feature points; Q ( u () represents the vector composed of the highest quantile acceleration values; Q ( l () represents the vector consisting of the lowest quantile accelerations; Q’ ( i ) indicates the standardized i A vector composed of acceleration values ​​of quantile feature points; the spacing between adjacent standardized feature lines reflects the relative change in acceleration of adjacent quantile samples. Considering that the safe acceleration threshold position may change in different speed ranges, let the speed range... R For the relatively stable region of the safe shift threshold position, the distance between adjacent standardized feature lines is calculated and denoted as the standardized adjacency difference SND. The specific processing formula is as follows: in, Indicates the speed range R Inside i SND indicator at the quantile characteristic line; This indicates that within a specific speed range R... i The mean acceleration of the quantile characteristic line; (23) Using the elbow diagram method, i The quantile characteristic lines SND are arranged in sequence, and a line graph is drawn to show the transition points of the acceleration values. The quantile characteristic line at the inflection point in the rising line graph is the location of the safe speed change threshold. The step (3) includes the following steps: (31) Calculation of the reliability of the safe speed change threshold: Find a bin with a large sample size, and randomly select from the population of the sample in the bin. n Each sample is used as a "subsample" to calculate the safe shift threshold, and this process is repeated. m The average safe shift threshold is calculated once, and denoted as... a 0 Each time increase l The average safe shift threshold is calculated repeatedly for each sample until the subsample size is large enough, where the nth sample is the largest. k The average value of the calculated safe shift threshold is denoted as . a k The final convergence value is denoted as , a k Corresponding credibility estimate The specific calculation formula is as follows: Construct a confidence function that reflects the actual changes in confidence level, and according to the definition of confidence level, the function must satisfy... Given two conditions, the final confidence function is: In the formula, x The number of samples; p ( x (The sample size is) x The reliability of the safe speed-changing threshold; C 1 and C 2 is an undetermined constant term, which needs to be determined by fitting actual data; (32) Fitting algorithm considering the reliability of sampling points: Let ( x , y ) is a pair of observations, x = [ x 1, x 2, …, x n ] T And satisfy the following ,in w = [ w 1, w 2,…, w n ] T These are parameters to be determined; based on the original loss function, the reliability of the sampling points is considered, and the reliability is calculated for the sample size of each bin to obtain a reliability sequence. p = [ p 1, p 2,…, p m ] T For a given m Group of observation data ( x i , y i ), m > n， Minimize the loss function: The final threshold curve equation obtained by fitting is: .

2. The method for identifying abnormal driver speed changes based on vehicle GPS data according to claim 1, characterized in that, The vehicle GPS data in step (1) includes six fields: license plate number, GPS time, longitude, latitude, direction and speed. After data cleaning, data interpolation, vehicle turning angle and acceleration calculation, the vehicle GPS data includes two fields: vehicle speed and acceleration.

3. The method for identifying abnormal driver speed changes based on vehicle GPS data according to claim 1, characterized in that, Step (4) achieves the following: the speed of historical data points or real-time data. x acceleration y Substitute into the threshold curve equation ,like If the data indicates abnormal speed change behavior, then the abnormal speed change behavior cannot be explained.

Images