An adaptive signal control method for pedestrian traffic

Abstract

The application discloses a kind of adaptive signal control methods of pedestrian traffic, comprising the following steps: S1: obtaining each time period pedestrian traffic flow in each import direction and real-time pedestrian traffic flow on zebra crossing;S2: generate data matrix according to historical pedestrian traffic flow;S3: using time series prediction method, generate data matrix according to the data matrix;S4: the time series clustering of obtained prediction data matrix is carried out, and the optimal clustering point is obtained, to obtain all classification number and classification sequence;S5: calculate the prediction data matrix after classification, obtain predicted pedestrian traffic flow;S6: calculate pedestrian crossing time, combine predicted pedestrian traffic flow and pedestrian crossing time, set the signal light control time of each time period each direction;S7: the end time of each time period green light, compare real-time pedestrian traffic flow with predicted pedestrian traffic flow, adjust signal light control time according to comparison result;After real-time detection green light ends, according to whether there is stranded pedestrian on zebra crossing, compare prediction data with real-time data, selectively extend green light time, sufficiently meet the demand of pedestrian traffic.

Description

Technical Field

[0001] This invention relates to the field of urban intelligent traffic signal control technology, and specifically to an adaptive signal control method for pedestrian traffic. Background Technology

[0002] In recent years, urban traffic congestion has become increasingly severe, and intelligent signal control methods are an important means to solve this problem. However, current adaptive control methods for traffic signals are mainly based on the analysis of vehicle demand, lacking effective adaptive control methods for slow-moving traffic systems, especially pedestrian signals. Furthermore, existing pedestrian control methods primarily analyze pedestrian crossing characteristics and determine the minimum green light time based on pedestrian walking speed to meet basic pedestrian crossing needs. For intersections near commercial areas, schools, and tourist attractions, where pedestrian traffic is high, simply relying on pedestrian walking speed to calculate green light time is insufficient to meet the large volume of crossing demand. The lack of effective adaptive traffic control methods based on pedestrian data makes it difficult for traffic signals to respond in real-time to changes in pedestrian traffic flow.

[0003] For example, Chinese Patent CN114724392A, published on June 30, 2015, discloses a road traffic signal control method and system based on video detection of pedestrians crossing the street. The method includes the following steps: Step S1: Acquire video images of pedestrians in the waiting area outside the pedestrian crossing; Step S2: When the pedestrian traffic light is red, calculate the number of pedestrians in the waiting area based on the pedestrian video images, and start timing when the first pedestrian is detected; Step S3: Determine whether the number of pedestrians in the waiting area exceeds a preset number, or whether the waiting time of the first pedestrian exceeds a preset time. If either of these conditions is met, proceed to Step S4; if neither is met, return to Step S2; and Step S4: Switch the pedestrian traffic light to a green light. This invention can automatically switch the pedestrian traffic light from a red light to a green light based on the number of pedestrians or the waiting time by acquiring pedestrian video images, thereby reducing pedestrian waiting time and improving traffic management efficiency. However, in practical applications, it cannot cope with situations where real-time data is lost due to equipment interruption or detection anomalies, and it cannot solve the problem of overlapping waiting areas in different directions at intersections. Summary of the Invention

[0004] The technical problem this invention aims to solve is that existing technologies cannot perfectly achieve adaptive traffic management. It proposes an adaptive signal control method for pedestrian traffic, which can predict pedestrian traffic flow based on historical data and dynamically adjust the signal scheme in conjunction with real-time data.

[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: an adaptive signal control method for pedestrian traffic, comprising the following steps:

[0006] S1: Obtain pedestrian traffic flow and real-time pedestrian traffic flow;

[0007] S2: Generate a data matrix based on historical pedestrian traffic flow;

[0008] S3: Use time series forecasting methods to forecast the data matrix;

[0009] S4: Perform temporal clustering to find the optimal cluster points;

[0010] S5: Calculate the classified prediction data matrix to obtain the predicted pedestrian traffic flow;

[0011] S6: Calculate pedestrian crossing time and set traffic light control times for each time period and direction;

[0012] S7: Compare real-time pedestrian traffic flow with predicted pedestrian traffic flow and adjust the traffic light control time accordingly.

[0013] An adaptive signal control method for pedestrian traffic first acquires pedestrian traffic flow data from each approach direction using video detection data. Then, sampling is performed at 5-minute intervals, dividing the day into 288 time periods, resulting in a 4*288 data matrix. Next, historical pedestrian traffic data from several consecutive days is extracted, and the LSTM method is used to predict pedestrian traffic flow at the intersection for each time period on the new day. Then, the predicted pedestrian traffic flow is subjected to temporal clustering using a dynamic Fisher algorithm, and finally, sequence classification is performed. The pedestrian traffic flow after sequence classification is calculated, and then the pedestrian crossing time is calculated. Combining the predicted pedestrian traffic flow and pedestrian crossing time, the signal light control times for each time period and direction on the day are set. The real-time pedestrian traffic flow obtained from the video is compared with the predicted pedestrian traffic flow, and the set signal light control times are adjusted to better meet pedestrian traffic needs.

[0014] Preferably, step S2 samples historical pedestrian traffic flow based on time interval t, and the data matrix is ​​shown below:

[0015]

[0016] Among them: Q n Let q be the pedestrian traffic volume on day n. i，j To predict pedestrian traffic flow at the i-th intersection during time period j on a given day, The historical pedestrian traffic flow data is sampled at time intervals t. Since there are four directions of traffic at the intersection, a final result is obtained. The data matrix Q.

[0017] Preferably, step S4 includes the following: setting an initial segmentation range K = [k1, k2, k3, ..., k max For each direction i, the data is [q]. i，1 q i，2 q i，3 , ..., q i，288 ], i∈(s,n,e,w); calculate the class diameter for all possible classifications: in, For [q] i，1 q i，2 q i，3 , ..., q i，288 The mean of ] is used to obtain the objective function: After obtaining predicted pedestrian traffic data from historical pedestrian traffic data, time-series clustering is required to divide the time period range of traffic characteristics. The dynamic Fisher algorithm can be used to accomplish this operation.

[0018] Preferably, the prediction data matrix in step S5 is as follows:

[0019]

[0020] Where: t i，n For different time intervals in each direction, i∈(s,n,e,w); Q i，n1 This represents the classification sequence corresponding to the current time interval in the current direction. After completing the sequence classification, each sequence is treated as a time period. The length of the time period is recorded as the number of subsequence elements multiplied by 5 minutes. The traffic flow value of this time period is recorded as the sum of the subsequence traffic flows, thereby calculating the pedestrian traffic flow after sequence classification.

[0021] Preferably, step S6 calculates the pedestrian crossing time using the following formula:

[0022]

[0023] Where: q is the pedestrian flow rate; t i，j The green light time corresponds to the current i-th inlet sequence flow; T is the average signal period; l is the pedestrian crossing width; Δs is the longitudinal safety clearance; V p t represents walking speed; min The time required for a single pedestrian to cross the street; t lThe time loss parameter is used to calculate pedestrian crossing time. This parameter needs to take into account the single pedestrian flow, crosswalk width, longitudinal safety distance, and walking speed. Since some pedestrians who are waiting at the crosswalk will speed up when the light turns red, and pedestrians will judge whether they can cross the intersection based on the countdown or green light indication time and choose to go through or wait, the time loss parameter needs to be added. In addition, special cases such as single pedestrian crossings also need to be considered.

[0024] Preferably, step S7 includes the following: Let the real-time pedestrian flow be q, the predicted pedestrian flow be q′, and the pedestrian density in a certain direction at the end of the green light be k. p，i The threshold number of people who do not pass through a green light is k. max If q ≥ q′ * 80%, and k p，i >k max This triggers a pedestrian overload alarm, marks the alarm phase, and extends the green light time for that alarm phase by Δg. i , When the actual traffic flow is less than or equal to 80% of the predicted pedestrian flow, and the pedestrian density at the end of the green light is greater than the threshold for the number of people who have not crossed the green light, a pedestrian overload alarm will be generated, the alarm phase will be marked, and the green light time will be extended accordingly.

[0025] Preferably, step S7 includes an adjustable time threshold t. max If there are multiple alarm phases, the extended green light time Δg for all alarm phases i The sum of the adjusted times for all alarm phases must be less than the adjustable time threshold. The traffic light control system has an adjustable time threshold. When an alarm phase is triggered, the green light time at the current moment can be extended, and subsequent traffic light control times will be shifted backwards. However, the sum of the adjusted times for all alarm phases on that day cannot exceed the adjustable time threshold.

[0026] Preferably, the time prediction method employs the LSTM method to obtain the predicted pedestrian flow Q. n+1 The time series prediction method uses LSTM, an improved recurrent neural network, as input data for the signal control method. LSTM is an improved recurrent neural network that can solve the problem of RNNs being unable to handle long-sequence dependencies. Based on this method, the 5-minute pedestrian flow data for each direction at the intersection on the new day is predicted and used as input data for the signal control method.

[0027] The substantial effect of this invention is that it designs an adaptive signal control method for pedestrian traffic, which predicts the latest day's pedestrian flow data based on historical data and generates a set of green light times for each direction to meet historical pedestrian needs; after the green light ends in real time, it compares the predicted data with the real-time data based on whether there are lingering pedestrians at the zebra crossing, and selectively extends the green light time to fully respond to pedestrian traffic needs. Attached Figure Description

[0028] Figure 1 This is a flowchart of the steps in Example 1;

[0029] Figure 2 This is a comparison chart of LSTM-based pedestrian flow prediction in Example 1;

[0030] Figure 3 This is a pedestrian flow prediction map for Example 1. Detailed Implementation

[0031] The specific embodiments of the present invention will be further described below with reference to the accompanying drawings.

[0032] Example 1:

[0033] An adaptive signal control method for pedestrian traffic, such as Figure 1 As shown, the process includes the following steps: S1: Obtain pedestrian traffic flow at various times and on zebra crossings in each direction; S2: Generate a data matrix based on historical pedestrian traffic flow; S3: Use time series forecasting to predict the data matrix based on the generated data matrix; S4: Perform time-series clustering on the obtained predicted data matrix, find the optimal cluster points, and obtain the number of all categories and the category sequence; S5: Calculate the classified predicted data matrix to obtain the predicted pedestrian traffic flow; S6: Calculate pedestrian crossing time, and combine the predicted pedestrian traffic flow and pedestrian crossing time to set the traffic light control time for each direction and time period; S7: At the end of the green light for each time period, compare the real-time pedestrian traffic flow with the predicted pedestrian traffic flow, and adjust the traffic light control time based on the comparison result.

[0034] First, pedestrian traffic flow and the real-time number of pedestrians on zebra crossings are obtained based on video detection data. Then, historical pedestrian traffic flow data is acquired, sampled at 5-minute intervals, and the daily data is reconstructed into a 4*288 data matrix Q, where each row represents data for one direction. A daily data matrix is ​​generated based on the historical data volume.

[0035] ......

[0036]

[0037] Next, a time series forecasting method is adopted, with LSTM (Long Short Term Memory Network) being the preferred approach for pedestrian flow prediction. LSTM is an improved recurrent neural network that can address the problem of RNNs' inability to handle long-sequence dependencies, and it has wide applications in time series forecasting. Using historical data Q1-Q... n Predict the 5-minute pedestrian flow data for each direction at the intersection for the entire day, denoted as Q. n+1 , which serves as the input data for the signal control method.

[0038] A pedestrian flow prediction model was trained using pedestrian crossing data from the south entrance zebra crossing at a certain intersection from February 9th to March 6th, 2023. Following the S1-S2 method, the dataset was divided into a prediction set (70%) and a training set (30%). Figure 2 The image shows the results of pedestrian flow prediction based on LSTM. Figure 3 As shown, this illustrates a comparison of data from one particular day.

[0039] Based on the time series prediction model obtained from the forecast, a prediction was made for the dataset on March 7th, resulting in dataset Q. n+1 Temporal clustering is performed to divide the time period range of traffic characteristics. Preferably, the dynamic Fisher algorithm is used, setting the initial segmentation range K = [k1, k2, k3, ..., k max For each direction i, the data [q] i，1 q i，2 q i，3 , ..., q i，288 ], i∈(s,n,e,w). Calculate the class diameter for all possible classifications: in, For [q] i，1 q i，2 q i，3 , ..., q i，288 The mean of ].

[0040] The objective function is

[0041] After the initial clustering is completed, the categorical subsequence Q` is obtained. n+1 =[Q s，1 Q s，2 Q s，n Find the optimal cluster point while changing the classification K value until the number of all categories is found.

[0042] Ordered Sample Classification Theorem: The optimal K-class split must be formed by adding one more class to the optimal K-1 class split based on a certain truncated field. Based on this principle, we therefore obtain: To divide the data into k classes, we first need to find the dividing point n. k-1 The minimum value of equation (3) is obtained, that is: B(n, k) = [B(n k-1 ,k-1)]+D(k)

[0043] Calculate pedestrian flow after sequence classification. Each sequence represents a time period. The time period length is recorded as the number of elements in the subsequence multiplied by 5 minutes. The flow value for that time period is recorded as the sum of the flow values ​​of the subsequences, denoted as .

[0044]

[0045] Referring to the Road Capacity Manual (HCM2010), the space required for pedestrians to queue or stand is 0.75m², the average shoulder width of a pedestrian is 0.75m, and there is a safe swing distance of 0.25m when crossing the street. Therefore, the lateral safety distance for pedestrians crossing the street is 1m. Considering forward and backward movement, the longitudinal safety distance Δs is 1.5m, and the walking speed V... p = 1.2 m / s. The time lost during startup acceleration is negligible. Figure 1 An exemplary schematic diagram of pedestrian crossing feature analysis according to the present invention is shown.

[0046] Assuming the pedestrian crossing demand is q, and the time required for a dense formation to cross the crosswalk is t:

[0047]

[0048] Among them, t min Take the table below, where n is the number of lanes for motor vehicles corresponding to the pedestrian crossing entrance:

[0049]

[0050] When pedestrian traffic is oversaturated, even when the traffic light turns red, some pedestrians remain on the crosswalk. The formula above calculates the time required for the pedestrian flow to fully cross. In reality, after the light turns red, some pedestrians remaining on the crosswalk will accelerate their crossing. When actually crossing the street, pedestrians will judge whether they can cross the intersection based on the countdown or the flashing green light, choosing to proceed or wait. Therefore, the crossing time parameter t is added. l Then we have:

[0051]

[0052]

[0053] Within each traffic light cycle, the time required for a pedestrian to cross the street is

[0054]

[0055] Calculate the average time required for pedestrians to cross the street in each time period and each cycle. Where T is the average signal cycle, and ti,j is the green light time corresponding to the j-th sequence of traffic flow at inlet i.

[0056] After calculating all entrances, the required green light time for pedestrians crossing in each direction at various times of the day is determined. This information is then sent to the signal control system for traffic control. Following the above steps, the predicted traffic flow data is divided into time periods and the corresponding pedestrian green light times are calculated, resulting in the data in the table below:

[0057] Time period start point End of time period Green light time 2023-03-07 00:00:00 2023-03-07 8:00:00 18 2023-03-07 8:00:00 2023-03-07 08:15:00 18 2023-03-07 08:15:00 2023-03-07 08:25:00 21 2023-03-07 08:25:00 2023-03-07 08:35:00 24 2023-03-07 08:35:00 2023-03-07 08:40:00 20 2023-03-07 08:40:00 2023-03-07 08:45:00 30 2023-03-07 08:45:00 2023-03-07 08:50:00 26 2023-03-07 08:50:00 2023-03-07 08:55:00 18 2023-03-07 08:55:00 2023-03-07 09:00:00 29 2023-03-07 09:00:00 2023-03-07 09:05:00 24 2023-03-07 09:05:00 2023-03-07 09:10:00 20 2023-03-07 09:10:00 2023-03-07 09:35:00 22 2023-03-07 09:35:00 2023-03-07 17:35:00 18 2023-03-07 17:35:00 2023-03-07 17:50:00 24 2023-03-07 17:50:00 2023-03-07 17:55:00 18 2023-03-07 17:55:00 2023-03-07 18:55:00 21 2023-03-07 18:55:00 2023-03-08 23:55:00 18

[0058] Subscribe to real-time pedestrian density data and record the most recent pedestrian density data at the end of each green light cycle. Obtain the 5-minute pedestrian flow value corresponding to the current time period, denoted as q, and predict the 5-minute pedestrian flow for that time period today as q'.

[0059] If q ≥ q′ * 80% and the pedestrian density k in a certain direction i at the end of the green light... p，i >k max If k is the pedestrian overload alarm, then an alarm will be triggered. max This is the threshold number of people who do not pass through when the green light is on.

[0060] An alarm is currently in effect. Mark the alarm phase and record the signal pattern for this cycle. Set the adjustable time threshold to t. max The alarm phase i time increases by Δg i .

[0061] If the cached alarm information only contains single-phase alarms, then If there are multiple phase alarms, the total adjustment of all alarm phases shall not exceed t. max If it exceeds, then t max According to the final density k of each phase p，i Distribute equally.

[0062] The above embodiments merely illustrate several implementation methods of the present invention, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention.

Claims

1. An adaptive signal control method for pedestrian traffic, characterized in that, Includes the following steps: S1: Obtain pedestrian traffic flow and real-time pedestrian traffic flow; S2: Generate a data matrix based on historical pedestrian traffic flow; S3: Use time series forecasting methods to forecast the data matrix; S4: Perform temporal clustering to find the optimal cluster points; S5: Calculate the classified prediction data matrix to obtain the predicted pedestrian traffic flow; S6: Calculate pedestrian crossing time and set traffic light control times for each time period and direction; S7: Compare real-time pedestrian traffic flow with predicted pedestrian traffic flow and adjust the traffic light control time accordingly; Let the real-time pedestrian flow be q, the predicted pedestrian flow be q', and the pedestrian density in a certain direction at the end of the green light be k. p,i The threshold number of people who do not pass through a green light is k. p,max ,like , and k p,i >k p,max This triggers a pedestrian overload alarm, marks the alarm phase, and extends the green light time for that alarm phase. , for .

2. The adaptive signal control method for pedestrian traffic according to claim 1, characterized in that, Step S2 involves sampling historical pedestrian traffic flow based on time interval t, and the data matrix is ​​shown below: ， Among them: Q n Let q be the pedestrian traffic volume on day n. i,j To predict pedestrian traffic flow at the i-th intersection during time period j on a given day, .

3. The adaptive signal control method for pedestrian traffic according to claim 1, characterized in that, Step S4 includes the following: setting the initial segmentation range. The data for each direction i is , ; Calculate the class diameter for all possible classifications: ,in, for The mean; thus, the objective function is obtained: .

4. An adaptive signal control method for pedestrian traffic according to claim 1, 2, or 3, characterized in that, The prediction data matrix for step S5 is shown below: ， in: For different time intervals in each direction, ; This is the classification sequence corresponding to the current time interval in the current direction.

5. An adaptive signal control method for pedestrian traffic according to claim 1, 2, or 3, characterized in that, Step S6 calculates the pedestrian crossing time using the following formula: ， Where: q represents pedestrian traffic flow; Let T be the green light time corresponding to the current i-th sequence flow rate; T be the average signal period; and l be the pedestrian crossing width. For longitudinal safety clearance; Walking speed; The time required for a single pedestrian to cross the street; This is the time loss parameter.

6. The adaptive signal control method for pedestrian traffic according to claim 1, characterized in that, An adjustable time threshold is set in step S7. If there are multiple alarm phases, the extended green light time for all alarm phases. The sum is less than the adjustable time threshold.

7. An adaptive signal control method for pedestrian traffic according to claim 1, 2, or 3, characterized in that, The time series prediction method uses the LSTM method to obtain the predicted pedestrian traffic flow. As input data for signal control methods.

Images