Subway station channel pedestrian abnormal overtaking behavior identification method

Abstract

The invention discloses a subway station channel pedestrian abnormal overtaking behavior identification method, which comprises the following steps: acquiring pedestrian motion videos in morning and evening commuting peak periods of a subway station channel, and extracting pedestrian motion feature data when a plurality of overtaking behaviors occur; firstly, a correlation analysis method is adopted to screen out an overtaking behavior key characterization index G which generates the most significant correlation influence on the walking speed of other pedestrians, then a quartile difference method is adopted to calculate an upper limit threshold Tu of the overtaking behavior key characterization index G so as to obtain a discrimination index M, then the discrimination index M is utilized to carry out negative and positive judgment on the overtaking behaviors, and finally the negative and positive judgment is carried out on the overtaking behaviors. And finally, according to the ROC curve, obtaining an optimal judgment threshold value G * of the key characterization index G of the exceeding behavior, and completing the judgment of the abnormal exceeding behavior by using the optimal judgment threshold value G *. Based on the quartile difference and the ROC curve, the abnormal pedestrian overtaking behavior of the subway station channel is recognized, and a scientific basis is provided for abnormal overtaking behavior recognition and control, refined passenger flow organization and facility design and the like.

Description

Technical Field

[0001] The present invention relates to the technical field of urban rail transit operation management, and in particular to a method for identifying abnormal overtaking behaviors of pedestrians in subway station passages. Background Art

[0002] The pedestrian passage is a main passenger service facility in subway stations. Passengers generally need to pass through the passage when entering or leaving the station or transferring. As one of the main pedestrian behaviors in subway station passages, in the case of high population density, pedestrians will be affected by psychological factors or other multiple factors, resulting in changes in their movement characteristics and behavior patterns, and it is easy to induce improper overtaking behaviors. Improper overtaking behaviors are significantly abnormal compared with general overtaking behaviors, which will greatly interfere with the stable movement order of pedestrian flows, reduce the passage efficiency and service level of the passage, and increase safety hazards such as collisions and stampedes in high-density crowds. Therefore, a method capable of accurately identifying abnormal overtaking behaviors in the passage is needed.

[0003] Currently, research on the identification of abnormal pedestrian behaviors generally tends to first define a normal behavior pattern and use statistical means to describe the data distribution characteristics, thereby establishing the range of normal values and evaluating the deviation degree of new observed values from the normal values to determine whether the behavior is abnormal. However, to ensure the application accuracy of such statistical methods, it is necessary to assume that the data distribution is symmetric. However, the data of pedestrian behavior characteristics is affected by many factors such as pedestrian attributes, facility layout, and pedestrian flow state, and its data distribution often shows an asymmetric characteristic. Therefore, only using statistical methods to determine the threshold range of the characteristic data of pedestrian overtaking behaviors in subway station passages and using this to distinguish abnormal overtaking behaviors of pedestrians often results in a low recognition accuracy. Summary of the Invention

[0004] The present invention provides a method for identifying abnormal overtaking behaviors of pedestrians in subway station passages, which identifies abnormal overtaking behaviors of pedestrians in subway station passages based on a combination method of interquartile range and ROC curve (Receiver Operating Characteristic curve), and provides a scientific basis for the identification and control of abnormal overtaking behaviors in subway stations, refined passenger flow organization, and facility design.

[0005] The present invention can be realized through the following technical solutions:

[0006] A method for identifying abnormal overtaking behaviors of pedestrians in subway station passages, which collects pedestrian movement videos during the morning and evening commuting peak hours in subway station passages, extracts pedestrian movement feature data when several overtaking behaviors occur, first uses the correlation analysis method to screen out the key characterization index G of overtaking behaviors that have the most significant correlation impact on the walking speeds of other pedestrians, and then uses the interquartile range method to calculate the upper threshold T of the key characterization index G of overtaking behaviors u , thereby obtaining the discrimination index M, then uses the discrimination index M to determine the positivity and negativity of these overtaking behaviors to draw an ROC curve, and finally obtains the optimal discrimination threshold G of the key characterization index G of overtaking behaviors according to the ROC curve * , and uses the optimal discrimination threshold G * to complete the discrimination of abnormal overtaking behaviors

[0007] Furthermore, it includes the following steps

[0008] Step 1: Collect pedestrian movement videos during the morning and evening commuting peak hours in subway station passages, extract pedestrian movement feature data when several overtaking behaviors occur, take the pedestrian movement feature data when one overtaking behavior occurs as one sample data, and group these sample data

[0009] Step 2: For each group of sample data, first use the correlation analysis method to select the key characterization index G of overtaking behaviors that have the most significant correlation impact on the walking speeds of other pedestrians, and then use the interquartile range method to calculate the upper threshold T of the key characterization index G of overtaking behaviors u ;

[0010] Step 3: Take the ratio of the key characterization index G of each sample data in this group of sample data to the upper threshold as the discrimination index M, set the discrimination index threshold, if the discrimination index M is greater than the discrimination index threshold, then consider this sample data as a positive sample, otherwise as a negative sample

[0011] Step 4: Based on the positive and negative sample data divided in Step 3, calculate the corresponding true positive rate and false positive rate to draw the ROC curve of the discrimination index, and then obtain the optimal discrimination threshold M of the discrimination index M according to the ROC curve * , and further obtain the optimal discrimination threshold G of the key characterization index G of overtaking behaviors * ;

[0012] Step 5: According to the optimal discrimination threshold G of the key characterization index G of overtaking behaviors * , re-discriminate this group of sample data: when the key characterization index G of a certain sample data is greater than the corresponding optimal discrimination threshold G * , then discriminate it as an abnormal overtaking behavior, otherwise as a normal overtaking behavior

[0013] Further, in the fourth step, set the value range of the discrimination index threshold and take incremental values at equal step lengths. For each discrimination index threshold, re - execute step three to perform positive - negative discrimination on this set of sample data, and then combine the positive - negative discrimination results corresponding to when the discrimination index threshold takes the value of 1 to calculate the corresponding true positive rate and false positive rate. Then, use the true positive rate and false positive rate corresponding to each discrimination index threshold as coordinate points to draw an ROC curve.

[0014] Further, assume that a set of sample data has a total of n sample data. In the value range of the discrimination index threshold, take incremental values from the lower threshold to the upper threshold at a predetermined step length, and a total of m discrimination index thresholds are obtained. When the discrimination index threshold is 1, execute step three for discrimination and obtain a total of n1 positive samples and n2 negative samples, where n1 + n2 = n.

[0015] For any y - th discrimination index threshold, repeat the execution of step three for discrimination to obtain n y1 positive samples and n y2 negative samples, where n y1 +n y2 = n, y = 1, 2,..., m. Then, use the following formula to calculate the corresponding true positive rate TPR and false positive rate FPR.

[0016]

[0017] Among them, TP is the true positive sample data, that is, the number of samples belonging to n1 among n y1 samples; FN is the false negative sample data, that is, the number of samples belonging to n1 among n y2 samples; TN is the true negative sample data, that is, the number of samples belonging to n2 among n y2 samples; FP is the false positive sample data, that is, the number of samples belonging to n2 among n y1 samples.

[0018] Then, use the coordinate points (FPR, TPR) composed of the true positive rate TPR and false positive rate FPR corresponding to each discrimination index threshold to draw an ROC curve.

[0019] Further, select the discrimination index threshold corresponding to the coordinate point on the ROC curve that is close to the point (0, 1) and has the largest Youden index as the optimal discrimination threshold M * of the discrimination index M, and then use the formula G * = M * T u to calculate the optimal discrimination threshold G * of the key characterization index G of the transcending behavior. If the key characterization index G of a certain sample data is greater than the optimal discrimination threshold G *, then it is determined as an abnormal overtaking behavior, otherwise it is a normal overtaking behavior.

[0020] Among them, Youden = TPR + TNR - 1 = 1 - (FPR + FNR).

[0021] Furthermore, in step four, the area AUC under the ROC curve is obtained by using the trapezoidal method for integration. If the area AUC is greater than the AUC threshold, it is determined that the ROC curve is reasonably drawn at this time, and the optimal discrimination threshold M * is effective. Otherwise, reset the value range of the discrimination index threshold and the predetermined step size, redraw the ROC curve, and make a determination until the corresponding ROC curve is reasonably drawn.

[0022] Furthermore, use the optimal discrimination threshold M * of the discrimination index M obtained from the previous ROC curve drawing as a benchmark, and reset the value range of the discrimination index threshold and the predetermined step size for the next ROC curve drawing.

[0023] Furthermore, the pedestrian movement characteristic data includes: the pedestrian flow density ρ in the area where the overtaking behavior occurs, the average value Δv of the speed change of the affected pedestrians before and after the overtaking behavior, and three overtaking behavior characterization indexes, namely the lateral distance D H at the moment of overtaking, the overtaking time t, and the longitudinal distance D Z ,

[0024] Determine the pedestrian service level grade of the channel based on the pedestrian flow density ρ, and divide the sample data of the same pedestrian service level grade of the channel into one group;

[0025] For each group of sample data, use the Pearson correlation coefficient method to analyze the correlation between the lateral distance D H at the moment of overtaking, the overtaking time t, and the longitudinal distance D Z and the average value Δv of the speed change of the affected pedestrians. Select the overtaking behavior characterization index with the largest correlation coefficient as the key overtaking behavior characterization index G, and then use the interquartile range method to calculate the upper threshold T u .

[0026] Furthermore, the pedestrian movement characteristics are calculated by the following steps:

[0027] S1. Select the segmented pedestrian movement videos with overtaking behaviors during the morning peak period from 7:00 to 9:00 and the evening peak period from 17:00 to 19:00 at typical channels of subway stations;

[0028] S2. Import the segmented video into Adobe Premiere Pro software for frame-by-frame playback and manual counting, and count the number of pedestrians \(i\) at the time of each overtaking behavior in the survey area. Further, combine the area of the survey area to obtain the pedestrian flow density \(\rho\).

[0029] S3. Use Tracker software to obtain the position of pedestrians in each frame, so as to extract the speeds \(v_1\), \(v_2\) of the surrounding pedestrians before and after the overtaking behavior, the lateral distance \(D\) at the overtaking moment H , the overtaking time \(t\) and the longitudinal distance \(D\) at the overtaking moment Z , and the calculation formulas are as shown below;

[0030] D H =|x p -x q |

[0031] t=t2 - t1

[0032] D Z =|y p -y q |

[0033] Wherein, \(x\) p is the abscissa of the position of pedestrian \(p\) at the overtaking moment, and \(x\) q is the abscissa of the position of pedestrian \(q\) at the overtaking moment; \(t1\) is the start time of the overtaking behavior, and \(t2\) is the end time of the overtaking behavior; \(y\) p is the ordinate of the position of pedestrian \(p\) at the overtaking moment, and \(y\) q is the ordinate of the position of pedestrian \(q\) at the overtaking moment;

[0034] S4. Calculate the speed change amount \(\Delta v\) of the affected pedestrians around the overtaking behavior j , calculate the average value \(\Delta v\) of the speed change amounts of the affected pedestrians, and the calculation formula is as follows,

[0035] \(\Delta v\) j =v 2j -v 1j

[0036]

[0037] Wherein, \(v\) 1j , \(v\) 2j represent the speeds of the \(j\)th affected pedestrian before and after the overtaking behavior; \(k\) represents the number of affected pedestrians in this overtaking behavior.

[0038] Further, each group of sample data contains at least 100 sample data.

[0039] The beneficial technical effects of the present invention are as follows:

[0040] (1) The present invention takes into account the differences in the parameter distributions of pedestrian overtaking behaviors under different channel - pedestrian flow densities (under different channel service levels) in subway stations, and identifies abnormal overtaking behaviors by considering different channel service levels, thereby improving the refinement level of abnormal behavior control.

[0041] (2) The present invention analyzes the correlation between different overtaking behavior characterization parameters and the average change in the pedestrian speed before and after the occurrence of overtaking behaviors, and uses the characterization parameter that has the most significant impact on the movement speeds of other pedestrians in the channel as the key characterization index for identifying abnormal overtaking behaviors, thereby improving the practical value of the abnormal behavior identification method.

[0042] (3) Different from the previous method of only using statistical methods to obtain the upper and lower limit thresholds of the data set, the present invention proposes a method for obtaining the optimal discrimination threshold of key characterization parameters by combining the interquartile range method and the ROC curve method, thereby improving the accuracy of the abnormal behavior identification results. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 is a schematic diagram of the overall process of the present invention;

[0044] Figure 2 is a schematic diagram of the ROC curve in the embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0045] The following details the specific implementation of the present invention in conjunction with the accompanying drawings and preferred embodiments.

[0046] As Figure 1 shown, the present invention provides a method for identifying abnormal overtaking behaviors of pedestrians in subway station channels. The pedestrian movement videos during the morning and evening commuting peak hours in the subway station channels are collected, and a number of pedestrian movement feature data when overtaking behaviors occur are extracted. First, the correlation analysis method is used to screen out the key overtaking behavior characterization index G that has the most significant correlation impact on the walking speeds of other pedestrians, and then the interquartile range method is used to calculate the upper limit threshold T u of the key overtaking behavior characterization index G, so as to obtain the discrimination index M. Then, the discrimination index M is used to determine the positive and negative of these overtaking behaviors to draw an ROC curve. Finally, the optimal discrimination threshold G * of the key overtaking behavior characterization index G is obtained according to the ROC curve, and the abnormal overtaking behaviors are discriminated with the optimal discrimination threshold G * .

[0047] Specifically as follows:

[0048] Step 1: Collect the pedestrian movement videos during the morning and evening commuting peak hours in the subway station channels, extract a number of pedestrian movement feature data when overtaking behaviors occur, use the pedestrian movement feature data when one overtaking behavior occurs as a sample data, and group these sample data.

[0049] The pedestrian motion characteristic data includes the channel pedestrian flow density ρ at the time of overtaking behavior, the average value of the speed change Δv of the affected pedestrians before and after the overtaking behavior, and three overtaking behavior characterization indicators: the instantaneous lateral distance D H , the overtaking time t, and the instantaneous longitudinal distance D Z .

[0050] S11. Select a typical channel of the subway station, delimit the survey area, measure the area of the survey area, and shoot the segmented pedestrian motion videos of overtaking behavior occurring during the morning peak period from 7:00 to 9:00 and the evening peak period from 17:00 to 19:00;

[0051] S12. Import the segmented videos into Adobe Premiere Pro software for frame-by-frame playback and manual counting, count the number of pedestrians i at the time of each overtaking behavior in the survey area, and further obtain the pedestrian flow density ρ in combination with the area of the survey area;

[0052] S13. Use Tracker software to obtain the position of pedestrians in each frame, so as to extract the speeds v1, v2 of the surrounding pedestrians before and after the overtaking behavior, the instantaneous lateral distance D H , the overtaking time t, and the instantaneous longitudinal distance D Z , and the calculation formulas are as follows;

[0053] D H =|x p -x q | (1)

[0054] t = t2 - t1 (2)

[0055] D Z =|y p -y q | (3)

[0056] Among them, x p is the abscissa of the instantaneous position of pedestrian p, x q is the abscissa of the instantaneous position of pedestrian q; t1 is the start time of the overtaking behavior, t2 is the end time of the overtaking behavior; y p is the ordinate of the instantaneous position of pedestrian p, y q is the ordinate of the instantaneous position of pedestrian q;

[0057] S14. Calculate the speed change Δv of the affected pedestrians around the overtaking behavior j , calculate the average value of the speed change Δv of the affected pedestrians, and the calculation formula is as follows,

[0058] Δv j =v2j -v 1j (4)

[0059]

[0060] wherein, v 1j and v 2j represent the speed of the j-th affected pedestrian before and after the overtaking behavior occurs; k represents the number of affected pedestrians in this overtaking behavior.

[0061] S15. From the above parameters (ρ, Δv, D H , t, D Z ), a sample data of an overtaking behavior is constituted, and thus these sample data are grouped.

[0062] Step 2. Using the channel pedestrian flow density ρ as the clustering index, according to the HCM pedestrian service level grading standard shown in Table 1, all the collected sample data are grouped, and each group contains at least 100 sample data, that is, at least 100 overtaking behavior sample data are collected under the same channel service level.

[0063] Table 1 HCM2000 (Highway capacity manual 2000, Road Capacity Manual)

[0064]

[0065]

[0066] When grouping, first use the pedestrian flow density ρ of each sample data as the classification index to determine the channel service level grade, and then divide the sample data of the same channel service level grade into one group, so as to group all the collected overtaking behavior sample data. To ensure the accuracy of abnormal behavior recognition, each group should contain at least 100 abnormal behavior sample data.

[0067] Step 2. For each group of sample data, first use the correlation analysis method to select the pedestrian movement feature with the largest correlation coefficient as the key characterization index G of the overtaking behavior, and then use the interquartile range method to calculate the upper threshold T of the key characterization index G of the overtaking behavior u .

[0068] S21. For each group of sample data, use the Pearson correlation coefficient method to analyze the correlation between the three overtaking behavior characterization indexes, namely the lateral distance D at the overtaking moment H , the overtaking time t and the longitudinal distance D at the overtaking moment Z and the average value of the speed change amount Δv of the affected pedestrians, and select the overtaking behavior characterization index with the largest correlation coefficient as the key characterization index G of the overtaking behavior, that is, from the lateral distance D at the overtaking moment H, Beyond time t and beyond the instantaneous longitudinal distance D Z Select one of them as the key characterization index G of the overtaking behavior. If the one with the strongest correlation with the average value of the speed change Δv is the instantaneous lateral distance D during overtaking H , then D can be selected H As the key characterization index G of the pedestrian overtaking behavior, it is a characterization parameter of the overtaking behavior that most significantly affects the movement speed of other pedestrians in the channel

[0069] The Person correlation coefficient calculation method can be selected, and its calculation formula is shown in Equation (6).

[0070]

[0071] Among them, r is the Person correlation coefficient, and X selects the instantaneous lateral distance D during pedestrian overtaking H , beyond time t and beyond the instantaneous longitudinal distance D Z These three indicators

[0072] S22. Calculate the upper threshold T of the key characterization index G using the interquartile range method u , taking D H As an example of the key characterization index G of the overtaking behavior, it will be explained in detail

[0073] ① According to the selected n sample data, use the following formula to calculate the average value H And the standard deviation of the instantaneous lateral distance D during overtaking And standard deviation

[0074]

[0075] Where: Is the instantaneous lateral distance in the i-th sample data, n is the number of samples Is the average value of the instantaneous lateral distance Is the standard deviation of the instantaneous lateral distance

[0076] ② Arrange the n In the sample data of this group from small to large to form an ordered data set {x1, x2,..., x n}}, and divide it into four equal parts. Q1 is the number ranked at the 25% position, called the lower quartile; Q3 is the number ranked at the 75% position, called the upper quartile; IQR is the interquartile range

[0077] ③ According to the value of n, calculate Q1 and Q3 respectively

[0078] When n = 2k + 1 (k = 0, 1,...) is an odd number, the calculation formulas for the lower quartile Q1 and the upper quartile Q3 are as shown in Equation (9).

[0079]

[0080] When n = 2k (k = 1, 2,...) is an even number, the calculation formulas for the lower quartile Q1 and the upper quartile Q3 are as shown in Equation (10).

[0081]

[0082] In the formula, {} represents the decimal symbol.

[0083] ④ Calculate the interquartile range IQR according to the calculation results of the lower quartile Q1 and the upper quartile Q3, as shown in Equation (11).

[0084] IQR = Q3 - Q1 (11)

[0085] ⑤ Calculate the lateral distance D at the moment of pedestrian overtaking H Upper limit threshold T u , and the calculation formula is as shown in Equation (12).

[0086] T u = Q3 + 1.5IQR (12)

[0087] Among them, T u represents the upper limit threshold of the lateral distance D at the moment of pedestrian overtaking H .

[0088] Step 3: Use the ratio of the key characterization index of the overtaking behavior of each sample data in this group of sample data to the upper limit threshold as the discrimination index M, set the discrimination index threshold. If the discrimination index M is greater than the discrimination index threshold, then the sample data is considered a positive sample, otherwise it is a negative sample. Taking the lateral distance D at the moment of overtaking H as an example of the key characterization index G of the overtaking behavior for detailed description, at this time

[0089] ① Use the ratio of the lateral distance D at the moment of overtaking H to its upper limit threshold T u as the discrimination index for the abnormal overtaking behavior of pedestrians , and the calculation formula is as shown in Equation (13).

[0090]

[0091] Among them: is the ratio of the lateral distance value at the moment of overtaking of the i-th overtaking behavior sample to the upper limit threshold T u .

[0092] ②According to to determine the positivity and negativity of the sample data of the i-th overtaking behavior in this group of data, taking the discrimination index threshold value of 1 as an example for illustration:

[0093] When is greater than the discrimination index threshold of 1, the sample data corresponding to this overtaking behavior is regarded as a positive sample, otherwise it is a negative sample, and so on. The discrimination of n sample data is completed, and a total of n1 positive samples and n2 negative samples are obtained, where n1 + n2 = n.

[0094] Step 4: Based on the positive and negative sample data divided in Step 3, calculate the corresponding true positive rate and false positive rate to draw the ROC curve of the discrimination index, and obtain the optimal discrimination threshold M of the discrimination index M * from the ROC curve, and further obtain the optimal discrimination threshold G of the key characterization index G of the overtaking behavior * Taking the lateral distance D H at the moment of overtaking as an example of the key characterization index G of the overtaking behavior for detailed illustration, at this time G = D H 、 Specifically as follows:

[0095] S41. Manually define the value range of the discrimination index threshold and take values at equal step lengths. For each discrimination index threshold, re-execute Step 3 to determine the positivity and negativity of this group of sample data, and then combine the positivity and negativity discrimination results corresponding to the discrimination index threshold value of 1 to calculate the corresponding true positive rate and false positive rate, and then use the true positive rate and false positive rate corresponding to each discrimination index threshold as coordinate points to draw an ROC curve.

[0096] ①Assume that the initial threshold of the discrimination index threshold is 1, the lower threshold is 0.5, and the upper threshold is 1.5. Starting from 0.5, take values by increasing a predetermined step length of 0.01 each time. Therefore, there are a total of [(1.5 - 0.5) / 0.01 + 1] = 101 threshold points, and at this time m = 101.

[0097] ③For any discrimination index threshold, re-execute Step 3 for discrimination to obtain n y1 positive samples and n y2 negative samples, where n y1 + n y2 = n, y = 1, 2,..., m represents the y-th discrimination index threshold, and then use the following formula to calculate the corresponding true positive rate TPR and false positive rate FPR,

[0098]

[0099] Among them, TP is the true positive sample data, that is, n y1The number of samples among the samples that belong to n1 (the number of positive samples corresponding to the discrimination index threshold of 1); FN is the false negative sample data, that is, n y2 The number of samples among the samples that belong to n1 (the number of positive samples corresponding to the discrimination index threshold of 1), TN is the true negative sample data, that is, n y2 The number of samples among the samples that belong to n2 (the number of negative samples corresponding to the discrimination index threshold of 1); FP is the false positive sample data, that is, n y1 The number of samples among the samples that belong to n2 (the number of negative samples corresponding to the discrimination index threshold of 1);

[0100] ③ For each discrimination index threshold, the coordinate points (FPR, TPR) composed of the true positive rate TPR and the false positive rate FPR, there are 101 coordinate points in total, and the ROC curve is drawn using SPSS software.

[0101] ④ Select the discrimination index threshold corresponding to the coordinate point on the ROC curve that is close to the point (0, 1) and has the largest Youden index as the optimal discrimination threshold M of the discrimination index M * , where, Youden = TPR + TNR - 1 = 1 - (FPR + FNR). If the instantaneous lateral distance D H at this time is used as the key characterization index G of the overtaking behavior, then the discrimination index threshold corresponding to the coordinate point with the largest Youden index is the discrimination index optimal discrimination threshold

[0102] In order to verify that the drawn ROC curve is reasonable and the determined optimal discrimination threshold M * is correct, we use the trapezoidal method to integrate to obtain the area AUC under the ROC curve. If the area AUC is greater than the AUC threshold, it is determined that the drawn ROC curve is reasonable and the determined optimal discrimination threshold M * is correct, and step ⑤ is executed. Otherwise, the value range and the predetermined step size of the discrimination index threshold are reset, the ROC curve is redrawn, and the determination is made until the drawn ROC curve is reasonable, and then the subsequent steps are continued accordingly.

[0103] There are 101 coordinate points in this embodiment, so 100 trapezoids can be obtained. The calculation formula for each trapezoid is shown in formula (16):

[0104]

[0105] Where: S k is the area of the kth curvilinear trapezoid in the ROC curve coordinate system;

[0106] y k-1 and y kThey are the vertical coordinates of the upper and lower bases of the k-th curvilinear trapezoid respectively;

[0107] Δx k is the height of the k-th curvilinear trapezoid.

[0108] Take the AUC threshold as 0.75. When the area AUC is greater than 0.75, it indicates that the ROC curve drawn at this time is reasonable. At this time, the obtained can be used as an effective index for identifying pedestrians' abnormal overtaking behavior;

[0109] If the area AUC is less than 0.75, it is necessary to optimize the value range and step precision of the discriminant index threshold, redraw the ROC curve, and make a determination until the drawn ROC curve is reasonable. The optimal discriminant threshold of the discriminant index obtained from the previous ROC curve can be used as a benchmark to reset the value range of the discriminant index threshold and the predetermined step for the next ROC curve drawing. That is, according to the optimal discriminant threshold obtained from the ROC curve at this time At this time, r = 0, indicating the number of times the value of the current discriminant index threshold is taken. Reset the value range of the discriminant index threshold as the lower limit threshold is The upper limit threshold is Adjust the step size to 0.004 and transfer to step ②. If the area AUC still does not meet the determination condition, it is necessary to continue to adjust the value range and step precision of the discriminant index threshold. At this time, r = r + 1 indicates the next value of the discriminant index threshold until the determination condition is met.

[0110] ⑤ Use the formula G * = M * T u to calculate the optimal discriminant threshold G * of the key characterization index G of the overtaking behavior.

[0111] Based on this optimal discriminant threshold G * , for each sample data in this group of sample data, judge the positive and negative: if the key characterization index G of the overtaking behavior of a certain sample data is greater than the optimal discriminant threshold G * , then judge it as an abnormal overtaking behavior, otherwise it is a normal overtaking behavior.

[0112] Suppose the lateral distance D H at the moment of overtaking is used as the key characterization index G of the overtaking behavior, then from the formula the optimal discriminant threshold of the key characterization index D H of the overtaking behavior can be calculated Make an abnormal determination of the overtaking behavior according to the above rules.

[0113] To verify the feasibility of the recognition method of the present invention, we conduct the following experiments:

[0114] Select a subway station passage in Shanghai, delimit the investigation area and measure its area. Shoot segmented pedestrian movement videos of overtaking behavior occurring during the morning rush hour from 7:00 to 9:00 and the evening rush hour from 17:00 to 19:00. Count the number of pedestrians \(i\) at each occurrence of overtaking behavior in the investigation area, and calculate the pedestrian flow density \(\rho\).

[0115] Import the video into Tracker software to obtain the position of each frame of the pedestrian, and extract the speeds \(v_1\), \(v_2\) of the surrounding pedestrians before and after the overtaking behavior, the lateral distance \(D\) at the moment of overtaking H , the overtaking time \(t\) and the longitudinal distance \(D\) at the moment of overtaking Z , and further calculate the average value of the speed change \(\Delta v\) of the affected pedestrians. From \((\rho,\Delta v,D H ,t,D Z ) constitute a sample data of an overtaking behavior, and a total of 924 sample data are obtained, among which 856 are valid data.

[0116] Using the pedestrian flow density \(\rho\) in the passage as a classification index, group the 856 overtaking behavior sample data, and each group should contain at least 100 abnormal behavior sample data. The number of sample data in each group is shown in the following table:

[0117]

[0118] Select the sample data under the D-level service level, which contains a total of 147 sample sizes. Based on the 147 sample data, use the Pearson correlation coefficient method to analyze the correlation between the lateral distance \(D\) at the moment of pedestrian overtaking H , the overtaking time \(t\) and the longitudinal distance \(D\) at the moment of overtaking Z and the average speed change \(\Delta v\) of the affected pedestrians.

[0119] The calculation results are shown in the table:

[0120]

[0121]

[0122] According to the calculated Person correlation coefficient table, select the lateral distance \(D\) at the moment of pedestrian overtaking with the largest correlation coefficient H as the key characterization index of overtaking behavior.

[0123] Calculate the average value H of the lateral distance \(D\) at the moment of pedestrian overtaking in the 147 sample data and the standard deviation Arrange the 147 in this group of data from smallest to largest to form an ordered data set \(\{x_1,x_2,\cdots,x 147}, calculate Q1, Q3, and IQR. The data samples are shown in the following table:

[0124]

[0125]

[0126] IQR = Q3 - Q1 = 0.15

[0127] Further calculate the lateral distance D at the moment of pedestrian overtaking H Upper threshold T u .

[0128] T u = Q3 + 1.5IQR = 1.285

[0129] Take the ratio of the lateral distance D at the moment of overtaking of each sample data H to the upper threshold T u as the discriminant index for pedestrian abnormal overtaking behavior. According to to discriminate the positivity and negativity of these 147 overtaking behavior sample data: when is greater than the discriminant index threshold 1, regard the overtaking behavior data sample as a positive sample. A total of 15 positive samples and 132 negative samples are obtained. Set the initial threshold of the discriminant index threshold to 1, the lower threshold to 0.5, and the upper threshold to 1.5. Start from 0.5 and change the threshold by 0.01 step each time, and re-discriminate the positivity and negativity of the 147 overtaking behavior data samples. Calculate the true positive rate (TPR) and false positive rate (FPR) corresponding to each threshold, and use SPSS software to draw the ROC curve. The drawn ROC curve is as

[0130] shown. Figure 2 .

[0131] According to Equation (16), calculate AUC = 0.907, which is greater than or equal to 0.75, indicating that this ROC curve can be used to identify pedestrian abnormal overtaking behavior. Calculate the Youden index of each point on the ROC curve, and it reaches the maximum value Y = 0.819 at the coordinate point (0.181, 1.000), and obtain the optimal discriminant threshold According to calculate to obtain another optimal discriminant threshold Re-screen this group of sample data to obtain 11 positive samples and 136 negative samples.

[0132] Although the specific embodiments of the present invention have been described above, those skilled in the art should understand that these are only examples. Without departing from the principle and essence of the present invention, various changes or modifications can be made to these embodiments. Therefore, the protection scope of the present invention is defined by the appended claims.

Claims

1. A method for identifying abnormal overtaking behaviors of pedestrians in a subway station passageway, characterized in that: Collect pedestrian movement videos during the morning and evening peak commuting hours at subway station passages, extract several pieces of pedestrian movement feature data when overtaking behaviors occur. First, use the correlation analysis method to screen out the key characterization index G of overtaking behaviors that have the most significant correlation impact on the walking speeds of other pedestrians. Then, use the interquartile range method to calculate the upper threshold T of the key characterization index G of overtaking behaviors u , thereby obtaining the discrimination index M. Then, use the discrimination index M to determine the positivity and negativity of these overtaking behaviors to draw an ROC curve. Finally, obtain the optimal discrimination threshold G of the key characterization index G of overtaking behaviors based on the ROC curve * , and use the optimal discrimination threshold G * to complete the discrimination of abnormal overtaking behaviors.

2. The method for identifying abnormal overtaking behavior of pedestrians in the subway station passage according to claim 1, wherein It includes the following steps: Step 1: Collect pedestrian movement videos during the morning and evening commuting peak hours in the subway station passageway, extract the pedestrian movement feature data when several overtaking behaviors occur, take the pedestrian movement feature data when one overtaking behavior occurs as a sample data, and group these sample data. Step 2: For each group of sample data, first use the correlation analysis method to select the key characterization index G of the overtaking behavior that has the most significant correlation effect on the walking speeds of other pedestrians, and then use the interquartile range method to calculate the upper threshold T of the key characterization index G of the overtaking behavior u ; Step 3: Use the ratio of the key characterization index G of the overtaking behavior of each sample data in this group of sample data to the upper limit threshold as the discrimination index M, set the discrimination index threshold. If the discrimination index M is greater than the discrimination index threshold, then this sample data is considered a positive sample, otherwise it is a negative sample. Step 4: Based on the positive and negative sample data divided in Step 3, calculate the corresponding true positive rate and false positive rate to plot the ROC curve of the discrimination index, and then obtain the optimal discrimination threshold M of the discrimination index M according to the ROC curve * , and further obtain the optimal discrimination threshold G of the key characterization index G of the transcendent behavior * ; Step 5. According to the optimal discrimination threshold G of the key characterization index G of the overstepping behavior * , re-discriminate this group of sample data: when the key characterization index G of a certain sample data is greater than the corresponding optimal discrimination threshold G * , then it is discriminated as an abnormal overstepping behavior, otherwise it is a normal overstepping behavior.

3. The method for identifying abnormal overtaking behavior of pedestrians in the subway station passage according to claim 2, characterized in that: In the said Step 4, set the value range of the discrimination index threshold and perform incremental value taking with equal step lengths. For each discrimination index threshold, re - execute Step 3 to perform positive - negative discrimination on this group of sample data, and then combine the positive - negative discrimination results corresponding to when the discrimination index threshold takes the value of 1, calculate the corresponding true positive rate and false positive rate, and then use the true positive rate and false positive rate corresponding to each discrimination index threshold as coordinate points to draw an ROC curve.

4. The method for identifying abnormal overtaking behavior of pedestrians in the subway station passage according to claim 3, wherein: Suppose a group of sample data has a total of n sample data, perform incremental value taking at a predetermined step length from the lower limit threshold to the upper limit threshold within the value range of the discrimination index threshold, and a total of m discrimination index thresholds are obtained. When the discrimination index threshold is 1, execute Step 3 to discriminate and obtain a total of n1 positive samples and n2 negative samples, where n1 + n2 = n. For any y-th discrimination index threshold, step three is repeatedly executed for discrimination to obtain n y1 positive samples and n y2 negative samples, where n y1 + n y2 = n, y = 1, 2,..., m. Then, the corresponding true positive rate TPR and false positive rate FPR are calculated using the following formula Among them, TP is the true positive sample data, that is, the number of samples belonging to n1 among n y1 samples; FN is the false negative sample data, that is, the number of samples belonging to n1 among n y2 samples, TN is the true negative sample data, that is, the number of samples belonging to n2 among n y2 samples; FP is the false positive sample data, that is, the number of samples belonging to n2 among n y1 samples; Then, use the coordinate points (FPR, TPR) composed of the true positive rate TPR and false positive rate FPR corresponding to each discrimination index threshold to draw an ROC curve.

5. The method for identifying abnormal overtaking behavior of pedestrians in the subway station passage according to claim 4, wherein: Select the discrimination index threshold corresponding to the coordinate point on the ROC curve that is close to the point (0, 1) and has the largest Youden index as the optimal discrimination threshold M of the discrimination index M * , and then use the formula G * = M * T u to calculate the optimal discrimination threshold G of the key characterization index G of the overbehavior * . If the key characterization index G of the overbehavior of a certain sample data is greater than the optimal discrimination threshold G * , then it is determined as an abnormal overbehavior, otherwise it is a normal overbehavior Among them, Youden = TPR + TNR - 1 = 1-(FPR + FNR).

6. The method for identifying abnormal overtaking behavior of pedestrians in the subway station passage according to claim 2, characterized in that: In step four, the area under the ROC curve, AUC, is obtained by using the trapezoidal method for integration. If the area AUC is greater than the AUC threshold, it is determined that the ROC curve is reasonably plotted at this time, and the optimal discrimination threshold M * is valid; otherwise, the value range of the discrimination index threshold and the predetermined step size are reset, the ROC curve is redrawn, and the determination is made until the corresponding ROC curve is reasonably plotted.

7. The method for identifying abnormal overtaking behavior of pedestrians in the subway station passage according to claim 6, characterized in that: Using the optimal discrimination threshold M of the discrimination index M obtained from the previous ROC curve drawing * as a benchmark, reset the value range of the discrimination index threshold and the predetermined step size for the next ROC curve drawing.

8. The method for identifying abnormal overtaking behavior of pedestrians in the subway station passage according to claim 2, characterized in that: The pedestrian movement characteristic data includes: the pedestrian flow density ρ in the area where overtaking behavior occurs, the average value Δv of the speed change of the affected pedestrians before and after overtaking behavior, and three overtaking behavior characterization indicators, namely the instantaneous lateral distance D during overtaking H , the overtaking time t, and the instantaneous longitudinal distance D during overtaking Z , Determine the pedestrian service level grade of the passageway according to the pedestrian flow density ρ, and divide the sample data with the same pedestrian service level grade of the same passageway into one group. For each set of sample data, the Pearson correlation coefficient method is used to analyze the correlation between the overtaking instantaneous lateral distance D H , the overtaking time t, and the overtaking instantaneous longitudinal distance D Z and the average value of the speed change Δv of the affected pedestrians. The overtaking behavior characterization index with the largest correlation coefficient is selected as the key overtaking behavior characterization index G, and then the interquartile range method is used to calculate the upper threshold T u of the key overtaking behavior characterization index G.

9. The method for identifying abnormal overtaking behavior of pedestrians in the subway station passage according to claim 8, wherein The following steps are used to calculate and obtain the pedestrian movement features: S1: Select the segmented pedestrian movement videos of overtaking behaviors that occur during the morning peak hours from 7:00 to 9:00 and evening peak hours from 17:00 to 19:00 in the typical passageway of the subway station. S2: Import the segmented video into Adobe Premiere Pro software for frame - by - frame playback and manual counting, count the number of pedestrians i when each overtaking behavior occurs in the surveyed area, and further obtain the pedestrian flow density ρ in combination with the area of the surveyed area. S3. Use the Tracker software to obtain the position of pedestrians in each frame, so as to extract the speeds v1 and v2 of the surrounding pedestrians before and after the overtaking behavior, the lateral distance D at the overtaking moment H , the overtaking time t, and the longitudinal distance D at the overtaking moment Z . The calculation formula is as follows D H = |x p - x q | t = t2 - t1 D Z = |y p - y q | where x p is the abscissa of the instant position when pedestrian p overtakes, and x q is the abscissa of the instant position when pedestrian q overtakes; t1 is the start time of the overtaking behavior, and t2 is the end time of the overtaking behavior; y p is the ordinate of the instant position when pedestrian p overtakes, and y q is the ordinate of the instant position when pedestrian q overtakes; S4. Calculate the speed change amount Δv of the affected pedestrians around the overtaking behavior j , calculate the average value Δv of the speed change amounts of the affected pedestrians. The calculation formula is as follows Δv j = v 2j - v 1j where, v 1j and v 2j represent the speeds of the j-th affected pedestrian before and after the overtaking behavior occurs; k represents the number of affected pedestrians in this overtaking behavior.

10. The method for identifying abnormal overtaking behavior of pedestrians in the subway station passage according to claim 8, wherein: Each group of sample data contains at least 100 sample data.

Images