Detection and Analysis Method for Airport Taxi Capacity Flow Based on Security Monitoring Videos

Abstract

A method for detecting and analyzing the airport taxi capacity flow based on security monitoring videos includes the following steps: Step 1: Divide the prediction period; Step 2: Identify passengers based on airport security monitoring videos; Step 3: Identify taxis based on airport security monitoring videos; Step 4: Predict the passenger flow arrival rate; Step 5: Predict the taxi capacity flow arrival rate; Step 6: Predict the passenger queuing time; Step 7: Predict the taxi queuing time. Thus, the present invention provides a better and more appropriate real-time perception of passenger flow demand and capacity supply, and can better accurately predict the queuing times of passengers and taxi drivers according to the supply-demand matching analysis, effectively improving the airport landside traffic service level and being highly innovative.

Description

Technical Field

[0001] The present invention relates to the technical field of airport taxi detection and analysis, and particularly to a method for detecting and analyzing the taxi capacity flow based on security monitoring videos at the airport. Background Art

[0002] With the development of social economy, civil aviation, as a fast and safe means of transportation, has currently become the first choice for long-distance travelers. Due to the particularity of civil aviation transportation's requirements for airspace, airports are mostly built in open areas far from the city center, and the land transportation connecting the city and the airport has become an important part of the entire journey of passengers. Since civil aviation passengers usually carry luggage and have relatively high requirements for travel comfort, convenient and comfortable taxis have become the main choice for civil aviation passengers to go to or leave the airport.

[0003] Due to the concentration of arriving passengers at the airport (passengers arriving at the airport by plane) following the arrival of flights, arriving passengers need to queue up to wait for taxis to leave the airport, and taxis waiting to pick up passengers at the airport also need to queue up in the storage yard to wait for passengers. For passengers and taxi drivers, obtaining an accurate prediction of the queuing time is beneficial for them to decide whether to wait according to their own needs and relieve the anxiety during the waiting process. For the airport management department, mastering the real-time passenger flow demand and taxi capacity supply situation can be used to carry out traffic management and dispatching work, improving the airport service level. Therefore, the real-time perception of passenger flow demand and capacity supply, and then accurately predicting the queuing time of passengers and taxi drivers based on supply-demand matching analysis, is of great significance for improving the land transportation service level of the airport.

[0004] The existing technologies mainly include the following several:

[0005] A method for dynamic matching, capacity monitoring and early warning, and intelligent allocation of airport connection transportation is disclosed in the Chinese invention patent with the application number 201410038974.9. This method predicts the passenger flow demand for taking taxis based on data such as date attributes, weather attributes, and the number of flights, and uses the difference between this value and the taxi supply as the basis for early warning and allocation. On the one hand, this method lacks a specific description of the method for real-time perception of passenger flow demand and taxi capacity; on the other hand, this method lacks a detailed analysis of the matching process between passenger flow and taxi capacity flow, and only uses the difference between supply and demand as the basis for allocation.

[0006] A system and method for predicting the supply-demand state of airport taxis are disclosed in the Chinese invention patent with the application number 201410817069.3. This method installs a device on the plane to collect the demand for passengers to take taxis after getting off the plane, and then sends the demand data to the management department of the destination airport for capacity allocation. This method requires additional information collection devices to be installed on the plane, with relatively high costs and lack of convenience.

[0007] A method for intelligent taxi dispatching at an airport is disclosed in a Chinese invention patent with the application number 201811568700.5. This method obtains passenger flow demand data through a code scanning device installed at the gate, and obtains taxi capacity data through a sensing device installed at the parking space, thereby achieving a one-to-one matching of passenger flow and capacity. This method also requires additional installation of information collection devices, and the perception of taxi capacity is limited to the passenger boarding and alighting area, with a small scope.

[0008] A method for predicting the queuing time of taxi passengers at a passenger transport hub based on image recognition is disclosed in a Chinese invention patent with the application number 202210052861.9. This method obtains passenger flow demand and taxi capacity data through cameras installed at the airport and image recognition technology, and analyzes the matching status between passenger flow and capacity through the M / M / C queuing model to calculate the passenger queuing time. On the one hand, this method only uses the real-time passenger flow and taxi capacity data obtained by video recognition, lacking the prediction of future traffic flow; on the other hand, the M / M / C queuing model used in this method cannot describe the phenomenon of reduced service efficiency caused by lane congestion in the taxi boarding and alighting area of the airport, and only gives an estimate of the passenger queuing time, lacking the queuing time for taxi drivers to refer to.

[0009] Therefore, in view of the above defects, the designer of the present invention, through painstaking research and design, and integrating the experience and achievements of being engaged in related industries for many years, has studied and designed a method for detecting and analyzing the taxi capacity flow at an airport based on security monitoring video to overcome the above defects. Summary of the Invention

[0010] The purpose of the present invention is to provide a method for detecting and analyzing the taxi capacity flow at an airport based on security monitoring video, which can effectively solve the defects of the prior art, provide better and more suitable real-time perception of passenger flow demand and capacity supply, better accurately predict the queuing time of passengers and taxi drivers according to the supply-demand matching analysis, and effectively improve the land-side traffic service level of the airport, with high innovation.

[0011] To achieve the above purpose, the present invention discloses a method for detecting and analyzing the taxi capacity flow at an airport based on security monitoring video, which is characterized by including the following steps:

[0012] Step 1: Divide the prediction time period. Divide 24 hours of a day into 48 time periods in half-hour units: Among them, d is the label of the characteristic day, including 7 categories: ordinary working day (d1), ordinary weekend (d2), day before holiday (d3), first day of holiday (d4), ordinary holiday (d5), last day of holiday (d6), day after holiday (d7);

[0013] Step 2: Passenger identification based on airport security surveillance videos. Using the airport security surveillance system, select the camera facing the entrance of the passenger queuing area, and identify the cumulative number of passengers entering the queuing area at each time period. Among them, is the time period serial number;

[0014] Step 3: Taxi identification based on airport security surveillance videos. Using the airport security surveillance system, select the camera facing the entrance of the taxi storage yard, and identify the cumulative number of taxis entering the storage yard at each time period. Among them is the time period serial number;

[0015] Step 4: Prediction of passenger flow arrival rate. Obtained from Step 2 The arrival rate of airport taxi passengers in the time period is (persons / min). Since this data can only be obtained after the end of the time period , for the passengers and vehicles arriving within this time period, it is necessary to predict the arrival rate of the current time period according to the arrival rate collected in the previous time period and the historical data of the passenger arrival rate on the same characteristic day;

[0016] Step 5: Prediction of taxi capacity flow arrival rate. Obtained from Step 3 The arrival rate of airport taxi vehicles in the time period is (vehicles / min). This data also needs to be obtained after the end of the time period ; Therefore, similar to Step 4, according to the arrival rate collected in the previous time period and the historical data of the vehicle arrival rate on the same characteristic day, predict the arrival rate of the current time period;

[0017] Step 6: Prediction of passenger queuing time. According to queuing theory, the process of airport passengers queuing to take a taxi is regarded as an M / M / 1 queuing system, that is, a passenger queuing system. The process of passengers entering the boarding and alighting area is always released in batches. Therefore, 1 batch of passengers is regarded as 1 customer, and the service time is the time t required for 1 batch of passengers to board the taxi P , which is obtained through on-site observation and statistics; thus, the service rate of the passenger queuing system can be obtained as Based on the passenger flow arrival rate obtained in Step 4, calculate the passenger queuing time according to Little's theorem of queuing theory

[0018] where N P is the number of passengers released in each batch;

[0019] Step 7: Prediction of taxi queuing time. According to queuing theory, the process of taxis queuing at the airport for passengers can also be regarded as a queuing system, that is, a vehicle queuing system. The process of taxis entering the boarding and alighting area is always released in batches. Therefore, one batch of vehicles is regarded as one customer, and the service time is the time t required for one batch of vehicles to pick up passengers and drive away from the boarding and alighting area. C ;

[0020] Obviously, the boarding and alighting area and the connected roadways before and after it together constitute a road system. The vehicle service time t C will be affected by the road congestion status. When congestion occurs, t C will increase to When the congestion dissipates, t C will gradually decrease from to and can be obtained through observation and statistics. The double-threshold M / M / 1 queuing model is used to model this queuing system and determine the taxi queuing time.

[0021] Among them: In step 2, machine vision recognition and centroid tracking algorithm are used to recognize the cumulative number of passengers entering the queuing area in each period. The specific process is as follows:

[0022] Step 2.1: Access the security monitoring video stream of the camera at the entrance of the passenger queuing area, and set virtual detection lines on the necessary route for passengers to enter the queuing area in the video frame.

[0023] Step 2.2: Use the YOLOv5 algorithm to identify and label passenger targets in the video frame.

[0024] Step 2.3: Use the centroid tracking algorithm for the identified and labeled passenger targets to identify the moving trajectory of their centroids.

[0025] Step 2.4: When the passenger centroid trajectory obtained in step 2.3 collides with the virtual detection line set in step 2.1, perform cumulative counting of the number of passengers.

[0026] Among them: In step 3, machine vision recognition and centroid tracking algorithm are used. The specific process is as follows:

[0027] Step 3.1: Access the security monitoring video stream of the camera facing the entrance of the taxi storage yard, and set virtual detection lines on the necessary route for taxis to enter the storage yard in the video frame.

[0028] Step 3.2: Use the YOLOv5 algorithm to identify and label taxi targets in the video frame.

[0029] Step 3.3: Use the centroid tracking algorithm to identify the moving trajectory of the centroid of the identified and marked taxi target;

[0030] Step 3.4: When the centroid trajectory of taxis obtained in step 3.3 collides with the virtual detection line set in step 3.1, the number of taxis is counted cumulatively.

[0031] Among them: In step 4, the combined model of ARMA and Kalman filter is used for prediction. The Kalman filter model can effectively utilize the information of historical data of the same period and has the advantage of fast calculation speed; while the ARMA model has better analysis and modeling capabilities for the evolution trend of real-time data. Therefore, the two models are combined to give full play to the advantages of different models and improve prediction accuracy.

[0032] Among them: The prediction process is as follows:

[0033] ① According to the passenger flow arrival rate obtained in the i-1 period, an ARMA model is constructed to predict the passenger flow arrival rate at time i (as shown in Formula 1):

[0034]

[0035] In the formula, To use the ARMA model The estimated value of , p is the autoregressive order, q is the moving regression order, ε is the error term, μ, γ, θ are the parameters to be estimated for the model, which can be estimated using the least squares method;

[0036] ② Estimate the noise covariance matrix P(i) (as shown in Formula 2), where the state transfer matrix A=I, and Q is the white noise covariance matrix:

[0037] P(i) - =AP(i-1)A T +Q(2)

[0038] ③ Update the Kalman gain K(i) (as shown in Formula 3), where the measurement matrix H = I, and R is the white noise covariance matrix:

[0039] K(i)=P(i)H T [HP(i) - H T +R] -1 (3)

[0040] ④ The average passenger arrival rate of the same characteristic day and the same period in the forecast day As the observed data (such as formula 4), the passenger flow arrival rate in time period i is obtained The final predicted value of

[0041] ⑤ Update the noise covariance matrix P(i) (as shown in Formula 5):

[0042] P(i)=[IK(i)H]P(i) - (5)

[0043] According to the above steps, a combined prediction based on Kalman filter update and ARMA model prediction as the core can be achieved. The formula (4) The passenger arrival rate in period i The final predicted value.

[0044] Among them: In step five, the combined model of ARMA and Kalman filter is used for prediction. The Kalman filter model can effectively utilize the information of historical data of the same period and has the advantage of fast calculation speed; while the ARMA model has better analysis and modeling capabilities for the evolution trend of real-time data. Therefore, the two models are combined to give full play to the advantages of different models and improve prediction accuracy.

[0045] Among them: The prediction process is as follows:

[0046] ① According to the vehicle arrival rate obtained in the i-1 period, an ARMA model is constructed to predict the vehicle arrival rate at time i (as shown in Formula 1):

[0047]

[0048] In the formula, To use the ARMA model The estimated value of , p is the autoregressive order, q is the moving regression order, ε is the error term, μ, γ, θ are the parameters to be estimated for the model, which can be estimated using the least squares method;

[0049] ② Estimate the noise covariance matrix P(i) (as shown in Formula 2,), where the state transfer matrix A=I, Q is the white noise covariance matrix:

[0050] P(i) - =AP(i-1)A T +Q(2,)

[0051] ③ Update the Kalman gain K(i) (as shown in Formula 3,), where the measurement matrix H = I, and R is the white noise covariance matrix:

[0052] K(i)=P(i)H T [HP(i) - H T +R] -1 (3,)

[0053] ④ The average vehicle arrival rate of the same characteristic day and the same period in the forecast day As the observed data (such as Equation 4), the vehicle arrival rate in the i-th period is obtained The final predicted value

[0054]

[0055] ⑤ Update the noise covariance matrix P(i) (such as Equation 5):

[0056] P(i) = [I - K(i)H]P(i) - (5)

[0057] According to the above steps, a combined prediction based on Kalman filter update and centered on ARMA model prediction can be realized. The obtained by Equation (4) is the vehicle arrival rate in the i-th period The final predicted value

[0058] Among them: The double-threshold M / M / 1 queuing model in Step 7 is as follows

[0059] There are two states of the service rate of the vehicle queuing system in the congested state and the non-congested state The state transition process is

[0060] ① When the number of batches of queuing vehicles in the vehicle queuing system is [0, U], the system service rate is

[0061] ② When the number of batches of queuing vehicles in the vehicle queuing system continues to increase to [U + 1, ∞], the system service rate drops to

[0062] ③ After the service rate of the vehicle queuing system drops to if the congestion gradually dissipates over time; when the number of batches of queuing vehicles drops to [0, D - 1], the system service rate increases to

[0063] On the basis of obtaining the predicted value of the taxi capacity flow arrival rate in Step 5 the waiting time W of vehicles in period i can be calculated by Equation 7 i C :[[]]END]]

[0064]

[0065] In the formula, U and D are the system congestion state thresholds, which can be obtained through observation and statistics; ρ l and ρ h can be obtained by Equation 8 and Equation 9

[0066]

[0067]

[0068] M is the number of vehicles released in each batch; π(0,1) is the system idle probability.

[0069] Among them: π(0,1) is determined according to Formula 10-14:

[0070]

[0071]

[0072]

[0073]

[0074]

[0075] In the formula, π is the system state probability. For example, π(i,1) represents the probability that the system state is non-congested when there are i batches of rental vehicles in the system; π(i,2) represents the probability that the system state is congested when there are i batches of rental vehicles in the system.

[0076] From the above content, it can be seen that the detection and analysis method of airport taxi capacity flow based on security monitoring video of the present invention has the following effects:

[0077] 1. Effectively solve the problems of high hardware cost and inconvenience in the current airport passenger flow and taxi capacity flow monitoring technologies. Utilize the existing security monitoring videos at the airport and identify the passenger and taxi flows through machine vision algorithms.

[0078] 2. Overcome the problem of poor timeliness in the current airport passenger flow and taxi capacity flow prediction technologies. Utilize the real-time and historical passenger and taxi flows identified by machine vision algorithms, and perform short-term prediction on the passenger and taxi flows through a combined model of ARMA model and Kalman filter.

[0079] 3. Aiming at the problem that it is difficult to describe the impact of lane congestion on service efficiency by using the basic queuing theory model in the current airport taxi capacity flow analysis technology, a queuing model based on double-threshold control is proposed to accurately estimate the queuing time under lane congestion conditions.

[0080] 4. Aiming at the problem that there is a lack of queuing time for drivers to refer to in the current airport taxi queuing system, the present invention proposes a double-system queuing model for passengers and vehicles based on double-threshold control to estimate the queuing times of both passengers and taxi drivers simultaneously.

[0081] The detailed content of the present invention can be obtained through the following description and the accompanying drawings. Brief Description of the Drawings

[0082] Figure 1 Shows a schematic diagram of the detection and analysis method of the airport taxi capacity flow based on the security monitoring video of the present invention.

[0083] Figure 2 Shows a schematic diagram of the airport taxi queuing system in the present invention.

[0084] Figure 3 Shows the state flow diagram of the dual-threshold queuing model in the present invention. Detailed implementation manners

[0085] See Figure 1 、 Figure 2 and Figure 3 ,Shows the detection and analysis method of the airport taxi capacity flow based on the security monitoring video of the present invention.

[0086] The detection and analysis method of the airport taxi capacity flow based on the security monitoring video includes the following steps:

[0087] Step 1: Since there are obvious fluctuations in the arrival rates of the passenger flow and the taxi capacity flow at different times at the airport, it is necessary to divide the whole day into time periods before predicting the arrival rates of the passenger flow and the taxi capacity flow, so as to facilitate subsequent predictions for different time periods. To ensure that the time granularity of the prediction is fine enough, 24 hours of a day are divided into 48 time periods in half-hour units: Among them, d is the label of the characteristic day, including 7 categories such as ordinary working day (d1), ordinary weekend (d2), the day before the holiday (d3), the first day of the holiday (d4), ordinary holiday (d5), the last day of the holiday (d6), and the day after the holiday (d7). For example, when i = 1 and d = d1, there is Indicates the time period from 0:00 to 0:30 on an ordinary working day. The reason for distinguishing different characteristic days is that the passenger flow and the taxi capacity flow on the same characteristic day have similar change rules. Statistical historical data of the passenger flow and the train flow on the same characteristic day are applied to the prediction models in steps 4 and 5 in order to improve the prediction accuracy.

[0088] Step 2: Based on the passenger recognition of the airport security monitoring video, using the airport security monitoring system, select the camera facing the entrance of the passenger queuing area (point 1, as Figure 2 shown), and use machine vision recognition and centroid tracking algorithm to identify the cumulative number of passengers entering the queuing area in each time period Among them, Is the time period serial number.

[0089] The specific process of the above algorithm is:

[0090] Step 2.1: Access the security surveillance video stream of the camera at the entrance of the passenger queue area (point 1), and set a virtual detection line on the route that passengers must take to enter the queue area in the video image.

[0091] Step 2.2: Use the YOLOv5 algorithm to identify and label passenger targets in the video.

[0092] Step 2.3: Use the centroid tracking algorithm to identify the moving trajectory of the centroid of the identified and marked passenger targets.

[0093] Step 2.4: When the passenger center of mass trajectory obtained in step 2.3 collides with the virtual detection line set in step 2.1, the number of passengers is counted cumulatively.

[0094] The YOLOv5 algorithm and the centroid tracking algorithm are mature public algorithms in this field, and their contents will not be repeated here.

[0095] Step 3: Taxi identification based on airport security surveillance video: Using the airport security surveillance system, select the camera facing the entrance of the taxi parking lot (point 2, such as Figure 2 ), using machine vision recognition and centroid tracking algorithms to identify the cumulative number of taxis entering the parking lot at each time period in The time period number.

[0096] The specific process of the above algorithm is as follows:

[0097] Step 3.1: Access the security monitoring video stream of the camera facing the entrance of the taxi parking lot (point 2), and set a virtual detection line on the route that taxis must take to enter the parking lot in the video image.

[0098] Step 3.2: Use the YOLOv5 algorithm to identify and label the taxi targets in the video.

[0099] Step 3.3: Use the centroid tracking algorithm to identify the moving trajectory of the centroid of the identified and marked taxi target.

[0100] Step 3.4: When the centroid trajectory of taxis obtained in step 3.3 collides with the virtual detection line set in step 3.1, the number of taxis is counted cumulatively.

[0101] Step 4: Passenger flow arrival rate prediction, obtained from step 2 The airport taxi passenger arrival rate during the period is (person / min). It is available only after the end of the time period, so for passengers and vehicles arriving during this time period, it is necessary to calculate the time according to the previous time period. The collected arrival rate and the historical data of the passenger arrival rate on the same characteristic days are used to predict the arrival rate in the current period.

[0102] To improve the accuracy, an ARMA and Kalman filter combined model is used for prediction. The Kalman filter model can effectively utilize the information of historical data in the same period and has the advantage of fast operation speed; while the ARMA model has better analysis and modeling capabilities for the evolution trend of real-time data. Therefore, combining the two models can give full play to the advantages of different models and improve the prediction accuracy. The prediction process is as follows:

[0103] ①Construct an ARMA model based on the passenger flow arrival rate obtained in the (i - 1) period to predict the passenger flow arrival rate at time i (as shown in Equation 1):

[0104]

[0105] In the formula, is the estimated value of using the ARMA model, p is the autoregressive order, q is the moving regression order, μ is the constant term, γ is the autocorrelation coefficient, and ε is the error term.

[0106] ②Estimate the noise covariance matrix P(i) (as shown in Equation 2), where the state transition matrix A = I:

[0107] P(i) - = AP(i - 1)A T + Q(2)

[0108] ③Update the Kalman gain K(i) (as shown in Equation 3), where the measurement matrix H = I:

[0109] K(i) = P(i)HT[HP(i) - H T + R]-1(3)

[0110] ④Use the mean value of the passenger flow arrival rate at the same time period on the same characteristic days in the history of the prediction day as the observation data (as shown in Equation 4) to obtain the final predicted value of the passenger flow arrival rate at time i

[0111] ⑤Update the noise covariance matrix P(i) (as shown in Equation 5):

[0112] P(i) = [I - K(i)H]P(i) - (5)

[0113] According to the above steps, a combined prediction based on Kalman filter update and centered around ARMA model prediction can be achieved. The obtained by formula (4) is the final predicted value of the passenger arrival rate in the i-th period.

[0114] Step 5: Prediction of the arrival rate of taxi transport capacity. It can be obtained from Step 3 that the arrival rate of airport taxis in the (vehicles / min), and this data also needs to be obtained after the period ends. Therefore, similar to Step 4, based on the arrival rate collected in the previous period and the historical data of the arrival rate of vehicles on the same characteristic days, the arrival rate in the current period is predicted. The specific process is similar to Step 4, and an ARMA and Kalman filter combined model is also used for prediction. The Kalman filter model can effectively utilize the information of historical data in the same period and has the advantage of fast operation speed; while the ARMA model has better analysis and modeling capabilities for the evolution trend of real-time data. Therefore, the two models are combined to give full play to the advantages of different models and improve the prediction accuracy.

[0115] Among them: The prediction process is as follows:

[0116] ①Construct an ARMA model based on the vehicle arrival rate obtained in the (i - 1)-th period to predict the vehicle arrival rate at the i-th moment (as shown in formula 1):

[0117]

[0118] In the formula, is the estimated value of using the ARMA model, p is the autoregressive order, q is the moving regression order, ε is the error term, and μ, γ, θ are the parameters to be estimated in the model, which can be estimated by the least squares method;

[0119] ②Estimate the noise covariance matrix P(i) (as shown in formula 2), where the state transition matrix A = I and Q is the white noise covariance matrix:

[0120] P(i) - = AP(i - 1)A T + Q(2, )

[0121] ③Update the Kalman gain K(i) (as shown in formula 3), where the measurement matrix H = I and R is the white noise covariance matrix:

[0122] K(i) = P(i)H T [HP(i) - H T + R] -1(3,)

[0123] ④Use the average vehicle arrival rate at the same time period on the historical same characteristic day as the prediction day as the observed data (as shown in Formula 4) to obtain the vehicle arrival rate at time period i as the final predicted value of the vehicle arrival rate at time period i

[0124]

[0125] ⑤Update the noise covariance matrix P(i) (as shown in Formula 5):

[0126] P(i)=[I - K(i)H]P(i) - (5,)

[0127] According to the above steps, a combined prediction based on Kalman filter update and centered on ARMA model prediction can be realized. The obtained by Formula (4) is the vehicle arrival rate at time period i as the final predicted value

[0128] Step 6: Prediction of passenger queuing time. According to queuing theory, the process of airport passengers queuing to take a taxi can be regarded as an M / M / 1 queuing system (hereinafter referred to as the passenger queuing system), and the service desk is the Figure 2 boarding and alighting area as shown. Due to airport management measures, the process of passengers entering the boarding and alighting area is always released in batches. Therefore, 1 batch of passengers is regarded as 1 customer, and the service time is the time t required for 1 batch of passengers to board the taxi P , which is obtained through on-site observation and statistics. Thus, the service rate of the passenger queuing system can be obtained as

[0129] According to Little's theorem of the M / M / 1 queuing model, on the basis of obtaining the passenger flow arrival rate prediction value in Step 4 , calculate the waiting time W of passengers in time period i through Formula 6 i P :

[0130]

[0131] In the formula, N P is the number of passengers released in each batch

[0132] Step 7: Prediction of taxi queuing time. According to queuing theory, the process of airport taxis queuing to wait for passengers can also be regarded as a queuing system (hereinafter referred to as the vehicle queuing system), and the service desk is the Figure 2The boarding and alighting area shown. Due to airport management measures, the process of rental vehicles entering the boarding and alighting area is always released in batches. Therefore, one batch of vehicles is regarded as one customer, and the service time is the time t required for one batch of vehicles to pick up passengers and drive out of the boarding and alighting area. C .

[0133] Obviously, the boarding and alighting area and the driving lanes connected before and after it together constitute a section of road system. The vehicle service time t C will be affected by the congestion state of the road. When congestion occurs, t C will increase to When the congestion dissipates, t C will gradually decrease from to and can be obtained through observation and statistics. To characterize this phenomenon, the present invention uses a double-threshold M / M / 1 queuing model to model this queuing system.

[0134] There are two states of the service rate of the vehicle queuing system, namely, the congested state and the non-congested state. The state transition process is as Figure 3 shown. The process is as follows:

[0135] ① When the number of batches of queuing vehicles in the vehicle queuing system is in [0, U], the system service rate is

[0136] ② When the number of batches of queuing vehicles in the vehicle queuing system continues to increase to [U + 1, ∞], the system service rate drops to

[0137] ③ After the vehicle queuing system service rate drops to , if the congestion gradually dissipates over time. When the number of batches of queuing vehicles drops to [0, D - 1], the system service rate increases to

[0138] On the basis of obtaining the predicted value of the taxi capacity flow arrival rate in step 5, the waiting time W of vehicles in period i can be calculated through formula 7 i C :

[0139]

[0140] In the formula, U and D are the system congestion state thresholds and can be obtained through observation and statistics. ρ l and ρ h can be obtained through formula 8 and formula 9.

[0141]

[0142]

[0143] M is the number of vehicles released in each batch; π(0,1) is the system idle probability, which can preferably be calculated according to Formula 10-14:

[0144]

[0145]

[0146]

[0147]

[0148]

[0149] In the formula, π is the system state probability. For example, π(i,1) represents the probability that the system state is non-congested when there are i batches of rental vehicles in the system; π(i,2) represents the probability that the system state is congested when there are i batches of rental vehicles in the system.

[0150] From the above, the advantages of the present invention are as follows:

[0151] 1. Effectively solve the problems of high hardware cost and inconvenience in the current airport passenger flow and taxi capacity flow monitoring technologies. By using the existing security monitoring videos at the airport, the passenger and taxi flows are identified through machine vision algorithms.

[0152] 2. Overcome the problem of poor timeliness in the current airport passenger flow and taxi capacity flow prediction technologies. By using the real-time and historical passenger and taxi flows identified through machine vision algorithms, the passenger and taxi flows are short-term predicted through a combined model of ARMA model and Kalman filter.

[0153] 3. Aiming at the problem that the current airport taxi capacity flow analysis technology is difficult to describe the impact of lane congestion on service efficiency by using the basic queuing theory model, a queuing model based on double-threshold control is proposed to accurately estimate the queuing time under lane congestion conditions.

[0154] 4. Aiming at the problem that the current airport taxi queuing system lacks the queuing time for drivers to refer to, the present invention proposes a double-system queuing model for passengers and vehicles based on double-threshold control to estimate the queuing times of both passengers and taxi drivers simultaneously.

[0155] It is obvious that the above description and record are merely examples and not intended to limit the disclosure, application or use of the present invention. Although the embodiments have been described in the examples and described in the drawings, the present invention is not limited to the specific examples illustrated in the drawings and described in the examples as the currently considered best mode for implementing the teachings of the present invention. The scope of the present invention will include any embodiments falling within the foregoing specification and the appended claims.

Claims

1. A method for detecting and analyzing the airport taxi capacity flow based on security monitoring videos, characterized in that The steps include: Step 1: Divide the prediction period. Divide 24 hours of one day into 48 periods at half-hour intervals: Among them, d is the label of the characteristic day, including 7 categories: ordinary working day (d1), ordinary weekend (d2), day before holiday (d3), first day of holiday (d4), ordinary holiday (d5), last day of holiday (d6), day after holiday (d7); Step 2: Passenger identification based on airport security surveillance videos. Using the airport security surveillance system, select the camera facing the entrance of the passenger queuing area to identify the cumulative number of passengers entering the queuing area at each time period. Among them, is the time period serial number; Step 3: Taxi identification based on airport security surveillance videos. Using the airport security surveillance system, select the camera facing the entrance of the taxi storage lot, and identify the cumulative number of taxis entering the storage lot at each time period. where is the time period serial number; Step 4: Prediction of passenger arrival rate. Obtained from Step 2 The arrival rate of airport taxis for the time period is (persons / min). This data can only be obtained after the end of the time period. Therefore, for the passengers and vehicles arriving within this time period, it is necessary to predict the arrival rate of the current time period based on the arrival rate collected in the previous time period and the historical data of the arrival rates of passengers on the same characteristic days; ​ Step 5: Prediction of the arrival rate of taxi transport capacity flow. Obtained from Step 3 The arrival rate of airport taxi vehicles during the time period is (vehicles / min). This data also needs to be obtained after the end of the time period. Therefore, similar to Step 4, based on the arrival rate collected during the previous time period and the historical data of the arrival rate of vehicles on the same characteristic days, predict the arrival rate of the current time period; Step 6: Prediction of passenger queuing time. According to queuing theory, the process of airport passengers queuing for taxis is regarded as an M / M / 1 queuing system, that is, a passenger queuing system. The process of passengers entering the boarding and alighting area is always released in batches. Therefore, 1 batch of passengers is regarded as 1 customer, and the service time is the time t required for 1 batch of passengers to board the taxi. P , which is obtained through on-site observation and statistics; thus, the service rate of the passenger queuing system can be obtained as Based on the passenger flow arrival rate obtained in Step 4, calculate the passenger queuing time according to Little's theorem of queuing theory where N P is the number of passengers released per batch; Step 7: Taxi queuing time prediction. According to queuing theory, the process of taxis queuing at the airport for picking up passengers can also be regarded as a queuing system, that is, a vehicle queuing system. The process of taxis entering the boarding and alighting area is always released in batches. Therefore, one batch of vehicles is regarded as one customer, and the service time is the time t required for one batch of vehicles to pick up passengers and drive away from the boarding and alighting area. C ; Obviously, the boarding and alighting area and the adjacent lanes together form a section of the road system, and the vehicle service time t C is affected by the congestion state of the road. When congestion occurs, t C will increase to When the congestion dissipates, t C will gradually decrease from to and can be obtained through observation and statistics. The double-threshold M / M / 1 queuing model is used to model the queuing system and determine the taxi queuing time.

2. The detection and analysis method of the airport taxi capacity flow based on the security monitoring video according to claim 1, wherein: In step 2, machine vision recognition and centroid tracking algorithm are used to identify the cumulative number of passengers entering the queuing area in each period. The specific process is as follows: Step 2.1: Access the security surveillance video stream of the camera at the entrance of the passenger queue area, and set a virtual detection line on the route that passengers must take to enter the queue area in the video image; Step 2.2: Use the YOLOv5 algorithm to identify and label the passenger targets in the video; Step 2.3: Use the centroid tracking algorithm to identify the moving trajectory of the centroid of the identified and marked passenger targets; Step 2.4: When the passenger center of mass trajectory obtained in step 2.3 collides with the virtual detection line set in step 2.1, the number of passengers is counted cumulatively.

3. The method for detecting and analyzing the airport taxi capacity flow based on the security monitoring video according to claim 1, wherein: In step 3, machine vision recognition and centroid tracking algorithms are used. The specific process is as follows: Step 3.1: Access the security surveillance video stream of the camera facing the entrance of the taxi parking lot, and set a virtual detection line on the route that taxis must take to enter the parking lot in the video screen; Step 3.2: Use the YOLOv5 algorithm to identify and mark the taxi targets in the video; Step 3.3: Use the centroid tracking algorithm to identify the moving trajectory of the centroid of the identified and marked taxi target; Step 3.4: When the centroid trajectory of taxis obtained in step 3.3 collides with the virtual detection line set in step 3.1, the number of taxis is counted cumulatively.

4. The method for detecting and analyzing the airport taxi capacity flow based on the security monitoring video according to claim 1, wherein: In step 4, the combined model of ARMA and Kalman filter is used for prediction. The Kalman filter model can effectively utilize the information of historical data of the same period and has the advantage of fast calculation speed; while the ARMA model has better analysis and modeling capabilities for the evolution trend of real-time data. Therefore, the two models are combined to give full play to the advantages of different models and improve prediction accuracy.

5. The detection and analysis method of airport taxi capacity flow based on security monitoring video according to claim 4, characterized in that: The prediction process is as follows: ① According to the passenger flow arrival rate obtained in the i-1 period, an ARMA model is constructed to predict the passenger flow arrival rate at time i (as shown in Formula 1): wherein, is the estimated value of using the ARMA model, p is the autoregressive order, q is the moving regression order, ε is the error term, and μ, γ, and θ are the parameters to be estimated in the model, which can be estimated by the least squares method; ② Estimate the noise covariance matrix P(i) (as shown in Formula 2), where the state transfer matrix A=I, and Q is the white noise covariance matrix: P(i) - = AP(i - 1)A T + Q(2) ③ Update the Kalman gain K(i) (as shown in Formula 3), where the measurement matrix H = I, and R is the white noise covariance matrix: K(i) = P(i)H T [HP(i) - H T +R] -1 (3) ④Use the average value of the passenger arrival rate at the same time period on the historical same characteristic days as of the prediction date as the observed data (as shown in Equation 4) to obtain the final predicted value of the passenger arrival rate in the i-th time period ⑤ Update the noise covariance matrix P(i) (as shown in Formula 5): P(i) = [I - K(i)H]P(i) - (5) According to the above steps, a combined prediction based on Kalman filter update and centered around ARMA model prediction can be achieved. What is obtained by formula (4) is the passenger arrival rate during the i-th period as the final predicted value.

6. The detection and analysis method of the airport taxi capacity flow based on the security monitoring video according to claim 4, characterized in that: In step five, the combined model of ARMA and Kalman filter is used for prediction. The Kalman filter model can effectively utilize the information of historical data of the same period and has the advantage of fast calculation speed; while the ARMA model has better analysis and modeling capabilities for the evolution trend of real-time data. Therefore, the two models are combined to give full play to the advantages of different models and improve prediction accuracy.

7. The detection and analysis method of airport taxi capacity flow based on security monitoring video according to claim 6, characterized in that: The prediction process is as follows: ① According to the vehicle arrival rate obtained in the i-1 period, an ARMA model is constructed to predict the vehicle arrival rate at time i (as shown in formula 1'): wherein, is the estimated value of using the ARMA model, p is the autoregressive order, q is the moving regression order, ε is the error term, and μ, γ, θ are the parameters to be estimated in the model, estimated by the least squares method; ② Estimate the noise covariance matrix P(i) (such as formula 2'), where the state transfer matrix A = I, Q is the white noise covariance matrix: P(i) - = AP(i - 1)A T + Q(2’) ③ Update the Kalman gain K(i) (as shown in Formula 3'), where the measurement matrix H = I and R is the white noise covariance matrix: K(i) = P(i)H T [HP(i) - H T +R] -1 (3’) ④Using the average vehicle arrival rate at the same time period on the historical same characteristic days as the prediction day as the observed data (such as formula 4'), the vehicle arrival rate at the i-th time period is obtained as the final predicted value ⑤ Update the noise covariance matrix P(i) (such as formula 5'): P(i) = [I - K(i)H]P(i) - (5’) According to the above steps, a combined prediction based on Kalman filter update and centered around ARMA model prediction can be achieved. What is obtained by formula (4’) is the vehicle arrival rate at time period i as the final predicted value.

8. The detection and analysis method of airport taxi capacity flow based on security monitoring video according to claim 1, characterized in that: The double-threshold M / M / 1 queuing model in Step 7 is as follows: The service rate of the vehicle queuing system under congested and non-congested states There are two states, and the state transition process is as follows: ① When the number of batches of queuing vehicles in the vehicle queuing system is in the range of [0, U], the system service rate is ②When the number of batches of queuing vehicles in the vehicle queuing system continues to increase to [U + 1, ∞], the system service rate drops to ③ After the service rate of the vehicle queuing system drops to if the congestion gradually dissipates over time; when the number of batches of queuing vehicles drops to [0, D - 1], the service rate of the system increases to On the basis of obtaining the predicted value of the taxi capacity flow arrival rate in step 5 the waiting time W of the vehicle in period i can be calculated by formula 7 i C : where U and D are the system congestion state thresholds, which can be obtained through observation and statistics; ρ l and ρ h can be obtained through Equations 8 and 9; M is the number of vehicles released in each batch; π(0,1) is the system idle probability.

9. The detection and analysis method of airport taxi capacity flow based on security monitoring video according to claim 8, characterized in that: π(0,1) is determined according to Equation 10-14: π(i, 1) = π(0, 1)ρ l i , i = 1, 2, ..., D - 1(10) In the formula, π is the system state probability, π(i,1) represents the probability that the system state is non-congested when there are i batches of rental vehicles in the system; π(i,2) represents the probability that the system state is congested when there are i batches of rental vehicles in the system.

Images