Road traffic multi-direction dynamic management and control method based on traffic flow monitoring

Abstract

The present application relates to the technical field of traffic dynamic management and control, in particular to a road traffic multidirectional dynamic management and control method based on vehicle flow monitoring, which can dynamically adjust road traffic management and control measures according to the road traffic operation state, thereby improving the traffic efficiency and reducing the traffic energy consumption. In view of the shortcomings and deficiencies in the prior art, the present application proposes a method for multidirectional dynamic management and control of large vehicle flow based on vehicle flow monitoring, which can monitor and predict traffic flow data in real time, then convert the traffic flow data into a matrix and a coordinate data model in multiple dimensions, and analyze and process the data by an artificial intelligence algorithm, thereby efficiently and accurately outputting a red-green light control strategy, so as to effectively improve the traffic control capability. Compared with the prior art, the storage capacity and cost of traffic data are greatly reduced, and the data processing efficiency is improved by dimensionality reduction reconstruction of real-time traffic data such as video, picture and radar signal.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of traffic dynamic management and control, in particular to a road traffic multi-directional dynamic management and control method based on traffic flow monitoring, which can dynamically adjust road traffic management and control measures according to the road traffic operation state, thereby improving the traffic efficiency and reducing the traffic energy consumption. BACKGROUND

[0002] With the increase of motor vehicles, traffic congestion and low vehicle traffic efficiency problems are increasing, and due to the backwardness of traffic data processing technology, the vehicle waiting time for red light is long, the large traffic flow is difficult to smoothly pass, the energy consumption is high, the pollution is serious, and the traffic management work is under great pressure.

[0003] The existing methods to relieve the pressure of traffic management mainly include road traffic expansion, control and reduction of total traffic volume (number limit passing, etc.), encouraging public transportation and bicycle travel, tidal lane, peak-shifting passing, etc., but the effect is not obvious. Currently, there are also technologies for improving management experience and strategies by analyzing historical traffic data, and the more typical one is the "green wave passing" technology, but through technical analysis and practice, it is proved that "green wave passing" can only be applied to specific passing direction and passing speed, and the practicability is poor.

[0004] Currently, there are also a small number of artificial intelligence algorithm technologies that improve traffic management and control capabilities by real-time collection and analysis of traffic flow data and optimization of traffic light timing, but practice has proved that these methods are not comprehensive in data feature mining, low in data utilization efficiency, and general in technical methods, and secondly, they are too dependent on the "black box" operation of artificial intelligence, cannot quantitatively explain the content of data analysis and operation, and cannot avoid the "AI uncertainty" problem. If such technologies are widely used in the field of public safety such as traffic management, there will inevitably be safety hazards. SUMMARY

[0005] The present application proposes a road traffic multi-directional dynamic management and control method based on traffic flow monitoring, which can monitor the real-time state of road traffic and give accurate traffic control instructions accordingly, thereby effectively improving the traffic efficiency and reducing congestion.

[0006] The present application achieves the following measures:

[0007] A road traffic multi-directional dynamic management and control method based on traffic flow monitoring, characterized in that it comprises the following steps:

[0008] Step 1: Constructing the traffic intersection area model: the intersection stop line and the extension line of the stop line are used as boundaries, the central intersection area of the traffic intersection is set as a control area, and the extended areas of roads in multiple directions are set as monitoring areas; in each monitoring area, the area flowing out of the monitoring area to the control area is set as an out area, and the area flowing into the monitoring area from the control area is set as an in area; Step 2: planning the passing path of the traffic intersection area model established in Step 1: taking each out area of the last intersection monitoring area as a starting point, connecting the out area of the current intersection in left, straight and right directions to construct a connection relationship diagram of the out area of the current intersection and the out area of the last intersection; when there are special vehicles in the current intersection, other connection relationship diagrams can be reconstructed for joint use;

[0009] Step 3: data acquisition and index extraction: real-time collection of video, picture and radar signal information of vehicle conditions in the out area of the current traffic intersection and one or N (N≥1) traffic control areas and multiple direction out areas, storage and use of the original data, K≥1, the value of K depends on the total length of the out area and the length of the area slice; then, the vehicle features, vehicle position and vehicle motion state are detected, tracked and the feature information index is extracted, the extracted feature information index is stored and used in an m i *n i atrix, 0≤i≤K; subsequently, further data analysis index extraction is performed on the stored data features, including the number and position of vehicles in each direction in K areas, vehicle density in K areas, vehicle inflow cumulative amount and vehicle outflow cumulative amount, the vehicles in each direction in K areas include in-transit vehicles and waiting vehicles;

[0010] Step 4: judging the large vehicle flow of the current intersection, including judging the large vehicle flow flowing out of the last intersection in multiple directions and judging the large vehicle flow already existing in the current intersection;

[0011] Step 5: predicting the passing time of the large vehicle flow already existing in the current intersection:

[0012] Step 5-1: predicting the passing time of the waiting vehicles in the current intersection: obtaining the waiting vehicle release time curve through data set training, first, obtaining the data set of vehicle quantity and passing time, setting the vehicle quantity as x and the vehicle passing time as y, the number of obtained sample groups is Q, and the i-th sample group is recorded as (x i ,y i ); secondly, setting the vehicle passing time as g(x), and solving g(x), there are two implementation methods as follows:

[0013] 1) Calculate the average or weighted average passing time of the sample under each vehicle quantity x as the predicted value of the vehicle passing time, or determine the predicted value by using the method of gradient descent to reduce the prediction error, and finally obtain the calculation error R, then correct the error, set the error threshold Rth, when R < Rth, the calculation is ended, when R < Rth, calculate x i The error of a single point is set as r i = g(x i ; α0, α1, …, α i )- y i , set the single-point error threshold as rth, when r i < rth, this point is regarded as valid data, when r n > rth, this point is regarded as an outlier and is excluded from the calculation process;

[0014] 2) Formulate the vehicle passing time to obtain the vehicle passing time prediction function g(x), which can be a segmented function or a non-segmented function, and the polynomial fitting formula of g(x) in each segment is:

[0015] g(x; α0, α2, …, α n ) = α0x n-1 + α1x n-2 + α2x n + … + α i

[0016] Bring the x i of the K groups of samples into the formula, and set the fitting error as Determine the parameter set α0, α1, …, α n by calculating the error between the prediction function and the true y i ;

[0017] Then, correct the fitting error, set the error threshold as Rth, when R < Rth, the fitting is ended, when R < Rth, calculate x i The error of a single point is set as r i = g(x n ; α0, α1, …, α i )- y i , set the single-point error threshold as rth, when r i < rth, this point is regarded as valid data and continues to the next round of fitting, when r await > rth, this point is regarded as an outlier and is excluded from the next round of fitting process;

[0018] Step 5-2: Calculate the position corresponding to the large vehicle release time: the to-be-line vehicle obtained by step 4 is recorded as N onwaylThe current vehicle flow release time is predicted, and the prediction result is g(N await ), then the large vehicle flow distance threshold (i.e. the position of the large vehicle flow release time) is calculated according to the release time under the premise that the in-transit vehicle flow speed v onway is known. According to the passing time, a time identity is set:

[0019]

[0020] Wherein, δ is the allowed time error term, t await

[0021] is the time required for the vehicle to be released to pass through the monitoring area,

[0022] t onway is the time when the in-transit vehicle flow reaches the boundary of the monitoring area and the control area;

[0023] The distance threshold L onway of the in-transit vehicle flow is obtained:

[0024]

[0025] Step 5-3: Calculate the large vehicle flow release time and release time:

[0026] Under the condition of achieving priority release in this area, according to the situation of the in-transit large vehicle flow, other in-transit vehicles and vehicles to be released, the optimal passing time and release time are determined, and the time and time allow a certain error, which can be adjusted according to the actual time. When there is no other in-transit vehicle or vehicle to be released in front of the large vehicle flow, the release time is the time when the large vehicle flow reaches the stop line; when there is an in-transit vehicle or a vehicle to be released in front of the large vehicle flow, the release time is the time when the large vehicle flow reaches the distance threshold L onway , and the release time T f is calculated as follows:

[0027] T f = (g(N await +N onwayl )+δ) + (h(N onway |v onway )+γ), wherein h(N onway |v onway ) is the passing time prediction result of N onway vehicles at speed v onway

[0028] The passing time prediction result of the vehicle,

[0029] δ and γ are time reservation terms;

[0030] Step 6: Large vehicle flow prediction out of the intersection in multiple directions:

[0031] Step 6-1: Determine whether the vehicle flow out cumulative amount constitutes a large vehicle flow, calculate the vehicle flow out cumulative amount data of the OUT area through the multi-direction vehicle feature matrix of the last intersection, calculate the vehicle density of the control area through the vehicle position matrix of the control area, set the density threshold value as N6 and the vehicle flow out cumulative amount threshold value as N7, when the density exceeds the threshold value and the vehicle flow out cumulative amount of the OUT area also exceeds the threshold value, calculate the time when the large vehicle flow is expected to reach the stop line of the intersection according to the length of the OUT area of the intersection and the vehicle flow speed, output the signal and time that a large vehicle flow may exist, otherwise, continue to the next step;

[0032] Step 6-2: Determine whether the vehicle to be released constitutes a large vehicle flow, obtain the driving direction and the number of vehicles to be released in each direction in the monitoring area through the vehicle feature matrix of the multi-direction OUT area of the last intersection and the connection relationship diagram of step 2, set the threshold value N8, determine whether the number of vehicles to be released in each direction and the cumulative number of vehicles to be released in different directions constitute a large vehicle flow, when the number of vehicles to be released is greater than the set threshold value, calculate the time when the large vehicle flow is expected to reach the stop line of the intersection according to the length of the OUT area of the intersection and the vehicle flow speed when the green light is released, output the signal and time that a large vehicle flow may exist, otherwise, continue to the next step;

[0033] Step 6-3: Determine whether the total number of in-transit vehicles and vehicles to be released constitutes a large vehicle flow, obtain the number of vehicles to be released in the monitoring area and the number of in-transit vehicles in the specified area through the vehicle feature matrix of the multi-direction OUT area of the last intersection, set the threshold value N9, determine whether the total number constitutes a large vehicle flow, when the total number is greater than the set threshold value, calculate the time when the large vehicle flow is expected to reach the stop line of the intersection according to the length of the OUT area of the intersection and the vehicle flow speed when the green light is released, output the signal and time that a large vehicle flow may exist, otherwise, continue to the next step;

[0034] Step 6-4: Determine whether there is an in-transit large vehicle flow, output the signal and position that a large vehicle flow may exist in each direction and the time when the large vehicle flow is expected to reach the stop line of the intersection according to the length of the OUT area of the intersection and the vehicle flow speed by using the method in step 4-3 through the vehicle feature matrix of the multi-direction OUT area of the last intersection, otherwise, give up tracking the large vehicle flow and consider that there is no large vehicle flow flowing out of the last intersection in multiple directions;

[0035] Step 7: Output the accurate control of the large vehicle flow:

[0036] Step 7-1: According to the prediction of the large vehicle flow converging at the current intersection and the detection result of the existing vehicle number at the current intersection, precise control is implemented, if it is predicted through step 4 that a large vehicle flow converges at the current intersection or it is detected that a large vehicle flow exists at the current intersection, the release time of the large vehicle flow, that is, the time of opening the green light and the release time are determined through the method in step 5, step 7-2: According to the prediction result of the large vehicle flow flowing out in multiple directions at the upper intersection, precise control is implemented, when no large vehicle flow is predicted at the current intersection, if it is predicted through step 6 that the large vehicle flow flowing out in multiple directions at the upper intersection can form a large vehicle flow, the time of the large vehicle flow reaching the stop line at the current intersection is predicted according to the release time of the signal light at the upper intersection and the length and speed of the OUT area of the current section, and the control of the current intersection is implemented, and the release time of the large vehicle flow after reaching the stop line needs to be determined according to the final vehicle number of the large vehicle flow detected at the current intersection; when the large vehicle flow flowing out in multiple directions at the upper intersection and the large vehicle flow are detected at the current intersection at the same time, the passing of the large vehicle flow at the current intersection is given priority to, so as to determine the release time of the large vehicle flow, that is, the time of opening the green light and the release time.

[0037] In step 3 of the present application, the vehicle characteristics, vehicle position and vehicle motion state are detected, tracked and feature information indexes are extracted, which can be realized by using image recognition, signal recognition artificial intelligence algorithm, and the extracted feature information indexes are stored and used according to m i *n i Matrix data feature storage and use, 0 < i < K, or by directly interacting with the vehicle to obtain interaction information, to obtain data feature indexes for storage and use.

[0038] Step 4 of the present application specifically includes the following steps:

[0039] Step 4-1: Determine whether there is a large vehicle flow converging at the current intersection, which is determined by way one or way two:

[0040] Way one, through the vehicle characteristic matrix of the entry area of the monitoring area of the current intersection, the vehicle density or vehicle number data in a historical time T l of the current intersection is obtained, the density threshold or vehicle number is set as N1, and whether there is a large vehicle flow is predicted through the vehicle density or vehicle number of the current intersection. When the density or vehicle number rises by more than the threshold value in the time interval T l , a large vehicle flow may exist; way two, a two-dimensional convolution time sequence deep learning network model is established to determine whether there is a large vehicle flow converging at the current intersection, the vehicle position matrix or video image in the time interval T l is obtained as the input end of the neural network model, the obtained multi-dimensional information matrix is taken as a sample, whether it is a large vehicle flow is taken as a label of the sample, a training data set is constructed, a convolutional neural network structure is built, and the vehicle density matrix at different time points is sent into the network to predict whether there is a large vehicle flow;

[0041] Step 4-2: If it is determined that the intersection will merge into a large vehicle flow, the number of vehicles passing through the last intersection, the release time and the average vehicle outflow are used to predict the number of large vehicle flows, and if the number of vehicles is greater than a set threshold, a signal indicating that there may be a large vehicle flow, a position and the number of vehicles in the large vehicle flow are output;

[0042] Step 4-3: Determine whether there is a large vehicle flow at the intersection, calculate the vehicle density and the number of vehicles in the K regions by using the vehicle feature matrix of the OUT region of the intersection, obtain the vehicle inflow cumulative amount data in the historical time T l , set the vehicle inflow cumulative amount threshold N2, the vehicle density threshold N3, the vehicle number threshold N4, and the in-transit large vehicle flow region length threshold N5, when the vehicle inflow cumulative amount is greater than the set threshold, a signal indicating that there may be a large vehicle flow and the position are output; the vehicle density or the average number of vehicles in the region is iteratively calculated by continuously expanding the K small regions, the maximum detection region with the vehicle density greater than the threshold or the average number of vehicles greater than the threshold is output, if the detection region length is greater than the in-transit large vehicle flow region length threshold, a signal indicating that there may be a large vehicle flow and the position are output, and the number of vehicles in the in-transit large vehicle flow and the number of other in-transit vehicles are obtained, the number of vehicles in the in-transit large vehicle flow is denoted as N onway , and the number of other in-transit vehicles is denoted as N onwayl ; otherwise, the tracking of the large vehicle flow is abandoned, and it is considered that there is no in-transit large vehicle flow at the intersection;

[0043] Step 4-4: Obtain the driving direction and the number of waiting vehicles in each direction in the monitoring region by using the vehicle feature matrix of the OUT region of the intersection, and the number of vehicle flows waiting to be released in the lane is denoted as N await .

[0044] The application also includes step 8, position acquisition of special vehicles and multi-intersection passing control, which specifically includes the following steps:

[0045] Step 8-1: Determine whether there is a special vehicle at the intersection, divide the intersection monitoring region into K regions, and obtain the special vehicle information by directly reading the vehicle marking information or through the vehicle feature matrix, if a vehicle is marked (or identified by features) as a special vehicle, obtain the position information, speed information, driving route and departure time information of the special vehicle, and set the total time from the current time to the departure time as T z ,

[0046] Step 8-2: Green light release condition calculation of the front N intersections, set the front N intersections to participate in the green light release control (determined by the driving route), obtain the position of the special vehicle and the vehicle feature matrix of the monitoring region of the front N intersections, if the special vehicle has no departure time information (i.e., it is already on the road), calculate the time for the special vehicle to reach the stop line of the i-th intersection in front by using the speed and position information of the special vehicle, and set it as STi , 0≤i≤N, the 0th intersection is the current intersection, if the special vehicle has departure time information, the vehicle speed is calculated using the set fixed vehicle speed; then, the total number of vehicles in front of the special vehicle in the monitoring area of the ith intersection in front is calculated, and the total passing time of the vehicles in front of the ith intersection monitoring area is calculated using the method of predicting the passing time in step 5, and is set as AT i ; finally, the time reservation item is set as γ, if the special vehicle has no departure time information, when ST i ≤AT i +γ, the green light release of the ith intersection is immediately performed, if the special vehicle has departure time information, the green light release needs to be delayed by T,

[0047] Step 8-3: Multi-intersection passing control is performed, when the special vehicle exists in the current intersection, the control of multiple intersections is performed by simultaneously calculating whether the release conditions of the front N intersections are met.

[0048] The present application aims at the shortcomings and deficiencies in the prior art, and proposes a method for multi-directional dynamic control of large traffic flow based on vehicle flow monitoring, which can monitor and predict traffic flow data in real time, then convert the traffic flow data into a matrix, coordinate data model, and analyze and process the data by artificial intelligence algorithm, and then efficiently and accurately output the red and green light control strategy, thereby effectively improving the traffic control capability. Compared with the prior art, (1) by dimensionality reduction reconstruction of real-time traffic data such as video, picture, radar signal, first, the storage capacity and storage cost of traffic data are greatly reduced, the data processing efficiency is improved, second, a series of quantifiable technical indexes are obtained by dimensionality reduction reconstruction, traffic data processing becomes more accurate and standardized, third, a series of quantifiable technical indexes are combined with digital twinning, virtual engine and other technologies, and the real historical traffic data can be virtually restored; (2) by using coordinate curve and matrix data model for multi-dimensional feature analysis of technical index data, data rules can be more accurately obtained, data outliers can be found, and traffic data processing visualization can be realized, thereby solving the problems of "black box" operation of artificial intelligence and "AI uncertainty", and widening the road for wide application of artificial intelligence technology in public safety fields such as traffic control; (3) the coordinate curve, matrix data model and algorithm innovatively proposed by the present application can be deeply integrated with artificial intelligence technology and products, and can provide reliable technical support for intelligent traffic control, vehicle-road cooperation and automatic driving after continuous accumulation and optimization. BRIEF DESCRIPTION OF DRAWINGS

[0049] The accompanying drawings are the flowchart of the present application. Figure 1

[0050] The accompanying drawings are the flowchart of the present application. Figure 2 ​is the flow chart for large vehicle flow judgment of the traffic state of the intersection in the present application. Figure 3 is the flow chart for large vehicle flow passing time judgment of the intersection in the present application.

[0051] is the flow chart for special vehicle management and control in the present application. Figure 4

[0052] is the flow chart for special vehicle management and control in the present application. Figure 5 is the flow chart for special vehicle management and control in the present application.

[0053] is the flow chart for special vehicle management and control in the present application. Figure 6 is the flow chart for special vehicle management and control in the present application.

[0054] is the flow chart for special vehicle management and control in the present application. Figure 7 is the flow chart for special vehicle management and control in the present application. DETAILED DESCRIPTION

[0055] The present application will be further described below in combination with the drawings and embodiments.

[0056] Embodiment 1

[0057] In this example, by setting zones for traffic conditions, constructing a two-intersection passing connection graph, acquiring traffic information such as road conditions, predicting large vehicle flow at the previous intersection in real time, and adjusting the passing situation of the present intersection, the specific process is as follows:

[0058] Step 1: Region construction: construct a traffic intersection region model, take the stop line and the extension line of the stop line as the boundary, set the central intersection area of the traffic intersection as the control area, and set the extended area of the road in multiple directions as the monitoring area; in each monitoring area, set the area flowing out of the monitoring area to the control area as the out area, and set the area flowing into the monitoring area from the control area as the in area;

[0059] Step 2: Plan the passing path of the region model: take each out area of the previous intersection monitoring area as the starting point, make connection with the out area of the present intersection in left, straight and right directions, and construct the connection relationship graph of the out area of the present intersection and the out area of the previous intersection as shown in the attached Figure 7 ;

[0060] Step 3: Data acquisition and index extraction: set the out area in multiple directions of the previous intersection to be 1250 meters long, with three lanes, and divide one area by 50 meters, real-time collect video, picture, radar signal information of the vehicle condition in 25 areas of each direction of the previous traffic road, acquire and store the original data for use; then, use image recognition, signal recognition artificial intelligence algorithm to detect, track and extract feature information index of vehicle characteristics, vehicle position and vehicle motion state, and then extract the feature information index according to m i *n​i The matrix is used for data feature storage and use, 0 < i < 25; finally, the stored data features are further subjected to data analysis index extraction calculation, specifically including the number and position of vehicles in each direction of travel in the 25 regions (including vehicles in transit and vehicles waiting to travel), vehicle density in the 25 regions, vehicle inflow cumulative amount, and vehicle outflow cumulative amount, as shown in the accompanying drawings.

[0061] Vehicle number feature extraction matrix at a certain moment:

[0062]

[0063] Vehicle density feature extraction matrix at a certain moment:

[0064]

[0065] Vehicle outflow cumulative amount: (The cumulative amount is calculated in 4 small time periods, and then summed to calculate the outflow amount from the current moment to the previous time period)

[0066] Out T1 T2 T3 T4 ALL OutE 0 0 0 0 0 OutS 0 0 0 0 0 OutN 3 2 2 1 8

[0067] Vehicle inflow cumulative amount: (The cumulative amount is calculated in 4 small time periods, and then summed to calculate the inflow amount from the current moment to the previous time period)

[0068] Out T1 T2 T3 T4 ALL OutE 16 6 6 2 40 OutS 1 2 5 2 10 OutN 1 3 0 2 6

[0069] Step 4: Large vehicle flow prediction at the intersection in multiple directions:

[0070] Step 4-1: Determine whether the vehicle outflow cumulative amount constitutes a large vehicle flow. The vehicle outflow cumulative amount data of each OUT region calculated in Step 3 is set to a vehicle outflow cumulative amount threshold of 30, and the vehicle outflow cumulative amounts of OUTE, OUTS, and OUTN regions are all above the threshold, so this step cannot determine whether a large vehicle flow exists, and the next step is continued.

[0071] Step 4-2: Determine whether the vehicle to be released constitutes a large flow. Through the vehicle feature matrix of the multi-direction OUT area of the last intersection and the connection relationship diagram of step 2, the number of vehicles to be released in each direction in the monitoring area is obtained, and a threshold of 60 is set to determine whether the vehicles to be released in each direction and the accumulated vehicles to be released in different directions constitute a large flow. The number of vehicles to be released in the OUTE area is 20, the number of vehicles to be released in the OUTS area is 10, and there is no vehicle to be released in the OUTN area. The total number of vehicles to be released is 30, which is less than the set threshold, so there is no large flow at present, and the next step of judgment is continued; Since the three OUT areas do not constitute a large flow, the threshold of the total number of vehicles is set to 55, and the density threshold is 0.5. The calculation can obtain that the total number of vehicles in the OUTS area is 60, and the density is 0.038. The OUTS area satisfies that the total number of vehicles is greater than the threshold and the density is less than the density threshold. Therefore, the red light control method is used to compress the traffic flow of the OUTS area to obtain a large flow. The OUTE area and the OUTN area are calculated in the same way;

[0072] Step 4-3: Determine whether there is an in-transit large flow. Through step 3, the vehicle density and vehicle quantity data in 25 regions (each region is 50 meters long) and the vehicle inflow cumulative amount data in a certain historical time are calculated. The vehicle inflow cumulative amount threshold is set to 50, the vehicle density threshold is 0.6, the average vehicle number threshold is 20, and the in-transit large flow region length threshold is 140 meters. The calculation shows that the vehicle outflow cumulative amount of the OUTE, OUTS, and OUTN areas all exceeds the threshold. Therefore, the 25 small regions are continuously expanded to the left and right, and the vehicle density and average vehicle number in the region are iteratively calculated. Taking the vehicle density calculation of the OUTE area as an example, the first round of calculation is performed for each small region as an independent region. The calculation results are shown in step 3. There are 5 regions with a density greater than 0.6, and the densities are 0.66, 0.61, 0.88, 0.72, and 0.6. Among them, 0.66 is the first region near the stop line, which is a waiting area, so it is excluded. In the second round of iteration, the remaining four regions are each expanded to the left and right by one region. The vehicle density values of the calculation regions are 0.54, 0.74, 0.73, and 0.48. Among them, 2 regions have a density greater than the threshold. In the third round of iteration, the vehicle density values are 0.59 and 0.58, both of which are less than the threshold. Therefore, the iteration is ended. The specific calculation table is shown in the attached figure. The second round of iteration with a density of 0.74 that meets the conditions is output. The head position of the vehicle flow in the detection region is 900 meters away from the stop line of the upper intersection, and the length is 150 meters. Because the length of the detection region is greater than the in-transit large flow region length threshold, the time when the large flow is expected to arrive at the stop line of the present intersection is output according to the length of the OUT area of the present intersection, the position of the large flow of the upper intersection away from the stop line of the upper intersection, and the vehicle flow speed;

[0073] Iteration calculation results table:

[0074]

[0075] Step 5: output accurate control of large vehicle flow:

[0076] According to the prediction results of the multi-directional outflow of large vehicles at the upper intersection, accurate control is implemented. It is predicted in step 4 that there is an on-the-way large vehicle flow in the OUTE area of the upper intersection, and the position of the on-the-way large vehicle flow is 900 meters away from the control area of the previous intersection. Therefore, combined with the release time of the signal light at the upper intersection and the length of the OUT area of the current section and the speed of the vehicle flow, the time when the large vehicle flow reaches the stop line of the current intersection is predicted, and the control of the current intersection is implemented. The release time after the large vehicle flow reaches the stop line needs to be determined according to the final number of vehicles detected at the current intersection;

[0077] Embodiment 2:

[0078] In this example, by setting zones for traffic conditions and constructing a path connection diagram, real-time large vehicle flow information (including quantity and position) in the intersection is obtained by acquiring traffic information such as road conditions, and the traffic situation at the intersection is adjusted in combination with the to-be-released vehicle flow information at the intersection. The specific process is as follows:

[0079] Step 1: Zone construction: construct a traffic intersection zone model, take the stop line and the extension line of the stop line as the boundary, set the central intersection area of the traffic intersection as the control area, and set the extended area of the road in multiple directions as the monitoring area; in each monitoring area, set the area flowing out of the monitoring area to the control area as the out area, and set the area flowing into the control area from the monitoring area as the in area;

[0080] Step 2: Plan the traffic path of the zone model: take each out area of the monitoring area of the previous intersection as the starting point, connect the out area of the current intersection in left, straight and right directions, and construct a connection relationship diagram of the out area of the current intersection and the out area of the previous intersection; the attached diagram is the same as in embodiment 1.

[0081] Step 3: Data acquisition and index extraction: the length of the outE area of the current intersection is 1250 meters, there are three straight lanes, and each area is divided into 50 meters. Real-time collection of video, picture, radar signal information of vehicle conditions is carried out in 25 areas of outE area, and original data is stored and used. Then, using image recognition, signal recognition and artificial intelligence algorithm, the characteristics, position and motion state of the vehicle are detected, tracked and the characteristic information index is extracted, and then m i *n iData feature storage and use by matrix, 0 < i < 25; finally, the stored data features are further analyzed to extract data analysis indicators, using a heat map, according to the length of the vehicle, assigning a numerical value to each vehicle, such as 5 for a regular car and 15 for a bus, assigning values to vehicles in specified locations within the 25 regions, and setting the range of the area covered by the numerical value to be 3 times the size of the vehicle, then calculating the total number of vehicle heat values in the 25 regions, as shown in the accompanying drawings; in addition, the vehicle inflow cumulative amount and the vehicle outflow cumulative amount in the outE area are counted, as shown in the accompanying drawings.

[0082] Vehicle quantity feature extraction matrix at a certain time:

[0083]

[0084]

[0085] Vehicle heat value matrix at a certain time:

[0086]

[0087] Vehicle outflow cumulative amount: (calculate the cumulative amount in 4 small time periods, then add to calculate the outflow amount from this time to the previous time period)

[0088] Out T1 T2 T3 T4 ALL OutE 0 0 0 0 0

[0089] Vehicle inflow cumulative amount: (calculate the cumulative amount in 4 small time periods, then add to calculate the inflow amount from this time to the previous time period)

[0090] Out T1 T2 T3 T4 ALL OutE 16 6 6 2 40

[0091] Step 4: Large vehicle flow detection at this intersection:

[0092] Step 4-1: Refer to the implementation method of step 4-3 in example 1, calculate the vehicle heat value data in the 25 regions through the vehicle quantity feature matrix of the OUT area of this intersection, and calculate the historical time T lThe vehicle inflow cumulative amount data, the vehicle inflow cumulative amount threshold 60, the vehicle average thermal value threshold 220, and the large vehicle flow area length threshold 150 are set. The vehicle inflow cumulative amount is calculated to be 40, which is less than the set threshold, and no signal indicating that a large vehicle flow may exist and the location thereof can be output. Then, the next step is calculated, and the 25 small areas are continuously expanded to the left and right sides, the vehicle average thermal value in the calculation area is iteratively calculated, the maximum detection area with the vehicle average thermal value greater than the threshold is output, if the detection area length is greater than the large vehicle flow area length threshold, the signal indicating that a large vehicle flow may exist and the location thereof is output, and the calculation result is shown in the drawing. The vehicle average thermal value of the third iteration meeting the condition is calculated to be 254, the head position of the vehicle flow in the area is 450 meters away from the stop line of the upper intersection, the length is 250 meters, and the number of vehicles is 82. Because the detection area length is greater than the large vehicle flow area length threshold, it can be determined that a large vehicle flow N onway = 82, and other vehicles in the area are denoted as N onwayl = 8.

[0093] The iterative calculation result table is as follows:

[0094]

[0095] Step 4-2: Using the vehicle number feature matrix of the OUT area of the intersection, the number of vehicles waiting in the monitoring area is obtained, and the number of vehicle flows waiting to be released is denoted as N await = 20.

[0096] Step 5: Predicting the large vehicle flow passing time of the intersection:

[0097] Step 5-1: Obtaining the vehicle release time curve by training the data set. First, the vehicle number and passing time data set is obtained, the vehicle number is denoted as x, the vehicle passing time is denoted as y, the number of obtained sample groups is denoted as K, the i-th sample group is denoted as (x i , y i ); secondly, the vehicle passing time is denoted as g(x), and g(x) is obtained. The vehicle passing time is calculated by formula, and the vehicle passing time prediction function g(x) is obtained. g(x) can be a segmented function or a non-segmented function, and the polynomial fitting formula of g(x) in each segment is as follows:

[0098] g(x; a0, a2, …, a n ) = a0x n + a1x n-1 + a2x n-2 + … + a n

[0099] The x i of the K sample groups is brought into the formula, and the fitting error is denoted as By calculating the error between the prediction function and the true y i , the parameter set a0, a1, …, a n is determined by minimizing the error.

[0100] Then, the fitting error is corrected, and the error threshold Rth is set. When R < Rth, the fitting ends. When R < Rth, the error of the single point is calculated and set as r i . i = g(x i ; a0, a1, …, a n ) - y i , and the single-point error threshold rth is set. When r i < rth, the point is considered as valid data and continues to the next round of fitting. When r > rth, the point is considered as an outlier and is excluded from the next round of fitting.

[0101] Step 5-2: Position of large vehicle flow release time: The release time of the current vehicle flow to be released is predicted. Through fitting calculation, the travel time prediction of 20 vehicles to be released is 15 seconds, i.e. g(N await ) = g(20) = 15. Since a small number of in-transit vehicles are detected near the to-be-released area in front of the large vehicle flow in the previous step, the number of in-transit vehicles is incorporated into the number of vehicles to be released, and the release time is calculated. The travel time prediction of 28 vehicles to be released is 18 seconds, i.e. g(N await + N onway1 ) = g(28) = 18. Then, under the premise of knowing the in-transit vehicle speed v onway = 30 km / h, the release time is used to calculate the distance threshold of the large vehicle flow (i.e. the position of the large vehicle flow release time). According to the travel time, the time identity is established:

[0102]

[0103] where δ is the allowed time error term, which is set to 0 in this embodiment,

[0104] t await is the time required for the to-be-released vehicle flow to pass through the monitoring area,

[0105] t onway is the time for the in-transit vehicle flow to reach the boundary between the monitoring area and the control area.

[0106] The distance threshold L onway of the in-transit vehicle flow is obtained:

[0107]

[0108] After calculation, L onway=(g(28)+0)*40=18*30=540 meters.

[0109] Step 5-3: Calculate the release time and release moment of large vehicle flow:

[0110] Under the condition of achieving priority release in this area, the calculation shows that there are 82 large vehicles on the way, 8 other vehicles on the way and 20 vehicles to be released. Since there are other vehicles on the way and vehicles to be released in front of the large vehicle flow, the release time is when the large vehicle flow reaches the distance threshold L onway = 540 meters, that is, when the large vehicle flow is 540 meters away from the stop line of the intersection, the large vehicle flow is released. Since the position detected by the large vehicle flow is less than 540 meters away from the stop line of the intersection, the release time is immediate.

[0111] The release time T is calculated as follows:

[0112] T=(t(N await +N onwayl )+δ)+(h(N onway |v onway )+γ), where h(N onway |v onway ) is the velocity v onwwy Next N onway

[0113] The vehicle travel time prediction results,

[0114] δ and γ are time reserved items, and are all set to 0 in this embodiment.

[0115] The release time of the large traffic flow is calculated as T = (g(28) + 0) + (h(82|30) + 0) = 18 + 50 = 68 seconds.

[0116] Step 6: Output precise control of large traffic flow:

[0117] Based on the detection results of the number of vehicles at the intersection, precise control is implemented. Step 4 detected the presence of a large traffic flow at the intersection, with 82 vehicles. There are a total of 28 vehicles waiting to be released and vehicles on the way. The method in Step 5 determines the release time (i.e., the moment the green light turns on) and the release time for the large traffic flow. The release time is now, and the release time is 68 seconds.

[0118] Example 3:

[0119] This example sets up traffic zones and constructs a traffic path connectivity map. By acquiring traffic information such as road conditions, it predicts the large traffic flow (including quantity and location) at the intersection in real time, and adjusts the traffic conditions at the intersection based on the information about the traffic flow to be released. The specific process is as follows:

[0120] Step 1: Area construction: construct the traffic intersection area model, bound by the stop line and the extension line of the stop line, set the center intersection area of the traffic intersection as the control area, set the extended area of the road in multiple directions as the monitoring area; in each monitoring area, set the area flowing out of the monitoring area to the control area as the out area, and set the area flowing into the monitoring area from the control area as the in area;

[0121] Step 2: Plan the traffic path of the area model: take each out area of the last intersection monitoring area as the starting point, connect the out area of the current intersection with the out area of the last intersection in left, straight and right directions to construct the connection relationship diagram of the out area of the current intersection with the out area of the last intersection; the attached diagram is the same as in embodiment 1.

[0122] Step 3: Data acquisition and index extraction: assuming that the length of the out area of the current intersection is 1250 meters, there are three lanes, and each area is divided into 50 meters, the out area is divided into 25 areas, and the video, picture and radar signal information of the real-time vehicle condition are collected to obtain the original data for storage and use; then, the image recognition and signal recognition artificial intelligence algorithm are used to detect, track and extract the characteristic information index of the vehicle features, vehicle position and vehicle motion state, and then the data characteristics are stored and used in the matrix, 0 < i < 25; finally, the stored data characteristics are further analyzed to extract the data analysis index, including the number and position of vehicles in the 25 areas, and the vehicle density in the 25 areas. i *n i

[0123] Step 4: Prediction of large vehicle flow into the current intersection:

[0124] Step 4-1: Establish a two-dimensional convolution time series deep learning network model, as shown in the attached Figure 6 , to predict whether there is a large vehicle flow.

[0125] ​First, the vehicle position matrix or video image in the time interval T is obtained, and the collected image or vehicle position matrix is preprocessed. The lane within a certain distance (the nearest area to the control area) of the last intersection is captured in real time by the camera, and the image is spliced; the time starts from the first vehicle entering, and multiple lane images at the same position are continuously obtained within T time, and the obtained multi-dimensional information matrix is used as the input end of the neural network model; then, the prediction model is built and the parameters are trained and optimized. The multi-dimensional information matrix obtained is used as a sample (model input), and whether it is an in-transit traffic flow is used as a label (model output) to build a training data set and build a CNN convolutional neural network structure. The data obtained at multiple times within the time interval T is respectively input into a two-dimensional convolution layer, the filter number is 32, the step is 2, the convolution kernel size is 7*7, followed by a maximum pooling layer, the step is 2, and the convolution kernel size is 3*3. Then, the same convolution layer is entered twice, the filter numbers are 64 and 128 respectively, and the remaining parameters are the same as those of the first layer. Finally, an average pooling layer is connected, a full connection layer is passed through, a linear feature layer result is output, the linear features at different times are spliced in time sequence, input into a one-dimensional convolution network, there are 3 layers, the filter numbers are 32, 64 and 128 respectively, the step is 2, and the convolution kernel size is 7. Followed by a maximum pooling layer, the step is 2, and the convolution kernel size is 3. Finally, a Dense dense layer network is built to predict whether the in-transit traffic flow exists. The network model block diagram is shown in the drawing. The cross-entropy loss function is used to calculate the loss of the network model, and the prediction result is The true label is y, and the cross-entropy loss function formula is as follows: Until the model loss is less than the predetermined threshold L, the model training is completed, the calculation parameters in the model are saved, and are used for real-time prediction; finally, the large traffic flow is predicted. The multi-dimensional information matrix is input into the model, and the prediction result is whether the in-transit traffic flow exists, and the position information of the in-transit traffic flow is obtained through the layout position of the camera.

[0126] Step 4-2: After prediction, there is a large traffic flow converging into the intersection, therefore, the number threshold is set to 50, the number of vehicles to be released within a period of time at the last intersection is 80, and the cumulative outflow of vehicles is 40. The number of vehicles to be released at the last intersection before T time is greater than the set threshold, therefore, a signal that there may be a large traffic flow is output, the predicted number of large traffic flow is 80, the position of large traffic flow is 1200 meters away from the stop line of the intersection, and is in the position of the in-transit traffic flow at the out area of the intersection; step 5: predict the large traffic flow passing time of the intersection:

[0127] Step 5-1: obtain the to-be-released traffic release time curve through data set training. First, obtain the vehicle number and passing time data set, set the vehicle number as x and the vehicle passing time as y, the number of obtained sample groups is K, and the i-th group of samples is recorded as (xi , y i ), second, set the vehicle passing time as g(x), and calculate g(x). Calculate the average passing time or weighted average passing time of the sample under each vehicle quantity x as the predicted value of the vehicle passing time, or determine the predicted value by using the gradient descent method to reduce the prediction error, and finally obtain the calculation error R. Then, correct the error, set the error threshold as Rth, when R i The error of a single point is set as r i = g(x i )- y i , set the single point error threshold as rth, when r i

[0128] Step 5-2: Position of large vehicle flow release time calculation: In this embodiment, it is set that there is no vehicle in the way in front of the large vehicle flow detected by the current intersection, and the vehicle to be released is 20. The release time of the current vehicle flow to be released is predicted, and the prediction result is g(N await ) = 15. Then, under the premise that the in-traffic vehicle speed v onway = 30, the distance threshold of the large vehicle flow (i.e. the position of the large vehicle flow release time) is calculated according to the release time. The distance threshold L onway of the in-traffic vehicle flow is obtained:

[0129] L onway = (g(N await + 0) + 0) * v onway = (15 * 30) = 450 meters.

[0130] Step 5-3: Large vehicle flow release time and release time calculation:

[0131] Under the condition that the priority release condition is reached in this area, it is calculated that there are 80 in-traffic large vehicles, 0 other in-traffic vehicles and 20 vehicles to be released. Since there are vehicles to be released in front of the large vehicle flow, the release time is the time when the large vehicle flow reaches the distance threshold L onway = 450 meters, that is, the release of the large vehicle flow is performed 450 meters away from the stop line of the intersection. The release time T is calculated as follows:

[0132] T = (g(N await ) + δ) + (h(N onway | v onway ) + γ), where h(N onway | v onway ) is the release time of N onwau vehicles under the speed v onway ​

[0133] the vehicle's travel time prediction result,

[0134] δ, γ are time reservation items, which are preset as 0 in the embodiment.

[0135] The release time of the large vehicle flow is calculated as t = (g(20) + 0) + (h(80|30) + 0) = 15 + 49 = 64 seconds.

[0136] Step 6: output the accurate control of the large vehicle flow:

[0137] According to the prediction result of the large vehicle flow at the intersection, accurate control is implemented. It is predicted by step 4 that the large vehicle flow at the intersection is 80 vehicles, and there are 20 vehicles to be released in front. Through the method in step 5, the release time of the large vehicle flow (i.e. the time of green light) and the release time are determined. The release time is 64 seconds.

[0138] Embodiment 4:

[0139] In this example, by setting zones for traffic conditions of multiple intersections, constructing a traffic path connection graph, and obtaining traffic information and feature matrices of each intersection, the special vehicles marked at the intersection are controlled by the green light of the whole road section. The specific process is as follows:

[0140] Step 1: region construction: construct a traffic intersection region model, take the stop line and the extension line of the stop line as the boundary, set the central intersection area of the traffic intersection as the control area, and set the extended area of the road in multiple directions as the monitoring area; in each monitoring area, set the area flowing out of the monitoring area to the control area as the out area, and set the area flowing into the control area from the monitoring area as the in area;

[0141] Step 2: plan the traffic path of the multi-intersection region model: take the out area of the monitoring area of the intersection as the starting point, connect the out area of the next intersection in the straight line and the direction of travel, and construct the connection relationship graph of the out area of the intersection and the out area of the next intersection. In this way, the connection relationship graph of multiple intersections is constructed, as shown in FIG. 8. Figure 7

[0142] ​Step 3: Data acquisition and index extraction: Assume that the out zone lengths of the four intersections below this intersection are 750 meters, 500 meters, 1000 meters, and 2000 meters respectively, and divide each area into 50 meters. The out zones of the four intersections are divided into 15, 10, 20, and 40 areas respectively to collect real-time video, pictures, and radar signal information of vehicle conditions, and obtain the original data for storage and use; then, for each intersection area, use image recognition and signal recognition artificial intelligence algorithms to detect, track, and extract characteristic information indicators of vehicle characteristics, vehicle positions, and vehicle movement conditions, and then press m i *n i The matrix is ​​used to store and use data features, 0<i≤number of regions; finally, the number of vehicles in the monitoring areas of the four intersections is calculated respectively.

[0143] Step 4: Acquisition of special vehicle locations and multi-intersection traffic control:

[0144] Step 4-1: Determine whether a special vehicle exists at this intersection. Assume that the intersection system has received a signal from a marked special vehicle. The special vehicle is located 600 meters from the stop line at the intersection, departs at 7:30, and has a fixed speed of 40 km / h. The special vehicle's route is to go straight through two intersections, turn left at the next intersection, and then continue straight to its destination. The current time is 7:20.

[0145] Step 4-2: Calculate the green light release conditions for this intersection and the three intersections ahead. Set up three intersections ahead to participate in green light release control: the OUTE zone of the next intersection, the OUTE zone of the next-down intersection, and the OUTN zone of the next-down-down intersection. Obtain the vehicle feature matrix for the monitoring areas of this intersection and the three intersections ahead. It is known that the fixed speed of the special vehicle is 40 km / h and it is 600 meters away from the stop line of this intersection. The time to reach the stop line of this intersection and the three intersections ahead is calculated, and the calculation results are ST0=15 seconds, ST1=27.5 seconds, ST2=52.5 seconds, and ST3=102.5 seconds, respectively. Then, the total number of vehicles ahead of the special vehicle in the monitoring area of ​​this intersection and the three intersections ahead is calculated, and the calculation results are 20, 100, 120, and 80, respectively. Using the method for predicting the travel time in step 5 of Example 2, the total travel time of the vehicles ahead of the monitoring area of ​​this intersection and the three intersections ahead is calculated, and the calculation results are AT0=15 seconds, AT1=31 seconds, AT2=43 seconds, and AT3=37 seconds, respectively. Finally, the time reserved item is set to γ=10 seconds. According to the conditions, the current intersection, the next intersection, and the next-down intersection meet ST i ≤AT i +γ, the green light is released. If the conditions at the next intersection are not met, the vehicle may not be released. Referring to the departure time of special vehicles, the green light release time of all intersections is postponed by 10 minutes.

[0146] Step 4-3: Multi-way intersection control is performed. To ensure the smoothness of the special vehicle road passage, the vehicle passage capacity of the current time through multiple intersections is calculated, and when the special vehicle departs, the green light of the current intersection, the next intersection and the intersection after the next intersection needs to be released. Among them, the vehicle passage capacity is updated every certain period of time.

[0147] The present application is directed to the shortcomings and deficiencies in the prior art, and proposes a method for multi-directional dynamic control of heavy traffic based on vehicle flow monitoring, which can monitor and predict traffic flow data in real time, then convert traffic flow data into multi-dimensional matrix and coordinate data models, and analyze and process them by artificial intelligence algorithms, and then efficiently and accurately output red and green light control strategies, thereby effectively improving traffic control capabilities. Compared with the prior art, (1) by dimensionality reduction reconstruction of real-time traffic data such as video, picture, radar signal, first, the storage capacity and cost of traffic data are greatly reduced, the data processing efficiency is improved, second, a series of quantifiable technical indicators are obtained through dimensionality reduction reconstruction, traffic data processing becomes more accurate and standardized, third, a series of quantifiable technical indicators are combined with digital twinning, virtual engine and other technologies, which can virtually restore real historical traffic data; (2) by using coordinate curve and matrix data model for multi-dimensional feature analysis of technical indicator data, data rules can be more accurately obtained, data outliers can be found, and traffic data processing visualization can be realized, thereby solving the problems of "black box" operation of artificial intelligence, "AI uncertainty" and other problems, and widening the road for the wide application of artificial intelligence technology in public safety fields such as traffic control; (3) the coordinate curve, matrix data model and algorithm innovatively proposed by the present application can be deeply integrated with artificial intelligence technology and products, and through continuous accumulation and optimization, it can provide reliable technical support for intelligent traffic control, vehicle-road cooperation and autonomous driving.

Claims

1. A multi-directional dynamic control method for road traffic based on vehicle flow monitoring, characterized in that: The following steps are involved: Step 1: Construct a traffic intersection area model: Using the intersection stop line and its extension as the boundary, the central intersection area of ​​the traffic intersection is set as the control area, and the road extension areas in multiple directions are set as the monitoring area. Within each monitoring area, the area where data flows out from the monitoring area to the control area is set as the out area, and the area where data flows in from the control area to the monitoring area is set as the in area. Step 2: Plan the traffic path for the traffic intersection area model established in Step 1: Starting from each out zone of the previous intersection monitoring area, connect the out zone of this intersection in the three directions of left, straight, and right, and build a connection relationship diagram between the out zone of this intersection and the out zone of the previous intersection; when there are special vehicles at this intersection, other connection relationship diagrams can be reconstructed and used together; Step 3: Data acquisition and index extraction: Collect the video, picture and radar signal information of the vehicle condition in real time in K areas of the out area of ​​the current traffic intersection and the previous or next N areas, the traffic road control area and the out areas in multiple directions, obtain the original data for storage and use, N≥1, K≥1, the value of K depends on the total length of the out area and the length of the area slice; then, detect, track and extract the characteristic information indicators of the vehicle characteristics, vehicle position and vehicle movement status, and calculate the extracted characteristic information indicators according to m i *n i The matrix is ​​used to store and use data features, 0<i≤K; then, the stored data features are further used to extract data analysis indicators, including the number and location of vehicles in each direction of travel within the K areas, the vehicle density within the K areas, the cumulative inflow of vehicles, and the cumulative outflow of vehicles. The vehicles in each direction of travel within the K areas include both in-transit vehicles and vehicles waiting to travel. Step 4: Determine the traffic flow at this intersection, including the presence of heavy traffic flowing out of the intersection from multiple directions and the presence of heavy traffic at this intersection. Step 5: Predict the travel time of heavy traffic at this intersection: Step 5-1: Predict the time for vehicles to pass through the intersection: Obtain the release time curve of the waiting vehicles through data set training. First, obtain the number of vehicles and the passing time data set. Let the number of vehicles be x, the vehicle passing time be y, and the number of sample groups obtained be Q. Then the i-th group of samples is recorded as (x i ,y i ); Secondly, let the vehicle passing time be g(x). There are two ways to calculate g(x): 1) Calculate the average passing time or weighted average passing time of the samples for each vehicle quantity x as the predicted value of the vehicle passing time, or use the gradient descent method to determine the predicted value to achieve the purpose of reducing the prediction error. Finally, obtain the calculation error as R. Then, perform error correction. Set the error threshold as Rth. When R < Rth, the calculation ends. When R < Rth, calculate x i The error of a single point, denoted as r i = g(x i ) - y i , set the single-point error threshold as rth. When r i < rth, this point is used as valid data. When r i > rth, this point is excluded as an outlier and does not participate in the calculation process; 2) The vehicle passing time is formulated and the vehicle passing time prediction function g(x) is obtained. g(x) can be a piecewise function or a non-piecewise function. The polynomial fitting formula of g(x) in each segment is: g(x;α0,α2,…,α n )=α0x n +α1x n-1 +α2x n-2 +…+a n Substitute the x of K groups of samples i into the formula, and set the fitting error as By calculating the error between the prediction function and the true y i , minimize the error to determine the parameter set α0, α1, …, α n ; Then, perform fitting error correction. Set the error threshold as Rth. When R < Rth, the fitting ends. When R < Rth, calculate the error of the single point of x i , denoted as r i = g(x i ; α0, α1, …, α n ) - y i . Set the single-point error threshold as rth. When r i < rth, this point is used as valid data to continue the next round of fitting. When r i > rth, this point is excluded as an outlier and does not participate in the next round of fitting process; Step 5-2: Calculate the position corresponding to the time of release of the large vehicle flow: The waiting vehicle obtained in step 4 is recorded as N await , the number of vehicles on the way is N onwayl , predict the release time of the current traffic flow to be released, and the prediction result is g(N await ), then when the on-road traffic speed v is known onway Under the premise of , the distance threshold of large traffic flow is obtained according to the release time. The distance threshold of large traffic flow is the position of the large traffic flow release moment. According to the passing time, the time identity is established: Among them, δ is the allowable time error term, t await The time required for the traffic to be released to pass through the monitoring area, t onway The time it takes for the on-the-way traffic to reach the boundary between the monitoring area and the control area; Get the distance threshold L of the on-the-way traffic flow onway : Step 5-3: Calculate the release time and release moment of large vehicle flow: Under the condition of achieving priority release in this area, the optimal passage time and release time are determined according to the situation of the large traffic flow, other vehicles on the way and vehicles to be released. This time and time allow a certain error and can be adjusted according to the actual time. When there are no other vehicles on the way or vehicles to be released in front of the large traffic flow, the release time is the time when the large traffic flow reaches the stop line; when there are vehicles on the way or vehicles to be released in front of the large traffic flow, the release time is when the large traffic flow reaches the distance threshold L onway The release time is T f The calculation is as follows: T f =(g(N await +N onwayl )+δ)+(h(N onway |v onway )+γ), where h(N onway |v onway ) is the velocity v onway Next N onway The vehicle travel time prediction results, δ and γ are time reservation items; Step 6: Prediction of heavy traffic flow from multiple directions at the upper intersection: Step 6-1: Determine whether the cumulative outflow of vehicles constitutes a large traffic flow. Calculate the cumulative outflow of vehicles in the OUT zone using the multi-directional vehicle feature matrix of the previous intersection. Calculate the vehicle density in the control zone using the vehicle position matrix of the control zone. Set the density threshold to N6 and the cumulative outflow threshold to N7. When the density exceeds the threshold and the cumulative outflow of vehicles in the OUT zone also exceeds the threshold, calculate the estimated time for this large traffic flow to reach the stop line of the intersection based on the length of the OUT zone and the traffic speed of the intersection. Output the signal and time indicating the possible presence of a large traffic flow. Otherwise, proceed to the next step. Step 6-2: Determine whether the vehicles waiting to be released constitute a large traffic flow. Using the vehicle feature matrix of the multi-directional OUT zone of the previous intersection and the connectivity diagram from step 2, obtain the driving direction within the monitoring area and the number of vehicles waiting to be released in each direction. Set a threshold N8 and determine whether the number of vehicles waiting to be released in each direction and the cumulative number of vehicles waiting to be released in different directions constitutes a large traffic flow. If the number of vehicles waiting to be released exceeds the set threshold, calculate the estimated time for this large traffic flow to reach the stop line of the intersection when the green light is released based on the length of the OUT zone of the intersection and the traffic speed. Output the signal and time indicating the possibility of a large traffic flow. Otherwise, proceed to the next step. Step 6-3: Determine whether the total number of vehicles on the road and vehicles to be released constitutes a large traffic flow. Using the vehicle feature matrix of the multi-directional OUT zone of the previous intersection, obtain the number of vehicles to be released within the monitoring area and the number of vehicles on the road within the specified area. Set a threshold N9 to determine whether the total number constitutes a large traffic flow. If it is greater than the set threshold, calculate the estimated time for this large traffic flow to reach the stop line at the intersection when the green light is released based on the length of the OUT zone of the intersection and the traffic speed. Output the signal and time indicating the possible presence of a large traffic flow. Otherwise, proceed to the next step. Step 6-4: Determine whether there is a large flow of traffic on the way. Using the vehicle feature matrix of the multi-directional OUT zone of the previous intersection and the method in step 4-3, output the signal and location of the possible large flow of traffic in each direction. Based on the length of the OUT zone of this intersection and the speed of the traffic flow, output the estimated time when this large flow of traffic will reach the stop line of this intersection. Otherwise, abandon tracking of the large flow of traffic and assume that there is no large flow of traffic flowing out of multiple directions at the previous intersection. Step 7: Output precise control of large traffic flow: Step 7-1: Implement precise control based on the prediction of large traffic flow entering the intersection and the detection results of the number of existing vehicles at the intersection. If a large traffic flow is predicted to enter the intersection or a large traffic flow is detected at the intersection through step 4, the release time of the large traffic flow, i.e. the time to turn on the green light and the release time, will be determined through the method in step 5. Step 7-2: Implement precise control based on the prediction results of large outflows from multiple directions at the upper intersection. If no large outflow is predicted at this intersection, but if step 6 predicts that vehicles flowing out of multiple directions at the upper intersection may form a large outflow, then combine the release time of the upper intersection signal light and the length and speed of the OUT zone of this section to predict the time when the large outflow will reach the stop line of this intersection, and implement control at this intersection. The release time of the large outflow after reaching the stop line needs to be determined based on the final number of large outflow vehicles detected at this intersection. When large outflows from multiple directions at the upper intersection and this intersection also detects large outflows, priority is given to the passage of the large outflow at this intersection to determine the release time of the large outflow, that is, the time when the green light is turned on and the release time.

2. A method for multi-directional dynamic control of road traffic based on vehicle flow monitoring according to claim 1, characterized in that: In step 3, the vehicle features, vehicle position, and vehicle motion status are detected, tracked, and feature information indicators are extracted using image recognition and signal recognition artificial intelligence algorithms, and the extracted feature information indicators are calculated by m i *n i The matrix is ​​used to store and use data features, 0<i≤K; or data feature indicators are obtained by directly obtaining interactive information with the vehicle for storage and use.

3. The method for multi-directional dynamic control of road traffic based on vehicle flow monitoring according to claim 1 is characterized in that: Step 4 specifically includes the following steps: Step 4-1: Determine whether there is a large flow of vehicles merging into the intersection using either Method 1 or Method 2: Method 1: Obtain a historical time T of this intersection through the vehicle feature matrix of the entrance area of ​​the monitoring area of ​​this intersection l The vehicle density or number data within the intersection is set to N1, and the vehicle density or number of vehicles at the intersection is used to predict whether there is a large traffic flow. l When the density or the number of vehicles increases beyond the threshold, there may be a large traffic flow. Method 2: Establish a two-dimensional convolutional time series deep learning network model to determine whether there is a large traffic flow that will merge into the intersection and obtain the time interval T l The vehicle position matrix or video image in the image is used as the input of the neural network model. The obtained multi-dimensional information matrix is ​​used as a sample, and whether it is a large traffic flow is used as the label of the sample. A training data set is constructed, and a convolutional neural network structure is built. The vehicle density matrix at different times is fed into the network to predict whether there is a large traffic flow. Step 4-2: If it is determined that the intersection will merge into a large traffic flow, the number of vehicles at the previous intersection, the release time, and the average vehicle outflow are used to predict the number of large traffic flows. If the number of vehicles is greater than the set threshold, the signal, location, and number of vehicles in the large traffic flow are output. Step 4-3: Determine whether there is a large traffic flow at this intersection. Calculate the vehicle density and vehicle quantity data in K areas through the vehicle feature matrix of the OUT area of ​​this intersection, and obtain the historical time T l The cumulative amount of vehicle inflow data in the system is set, and the cumulative amount threshold N2 of vehicle inflow, the vehicle density threshold N3, the vehicle number threshold N4, and the length threshold N5 of the area with large traffic flow in transit are set. When the cumulative amount of vehicle inflow is greater than the set threshold, the signal and location of the possible existence of large traffic flow are output; by continuously expanding K small areas, the vehicle density or the average number of vehicles in the area is iteratively calculated, and the maximum detection area with vehicle density greater than the threshold or the average number of vehicles greater than the threshold is output. If the length of the detection area is greater than the length threshold of the area with large traffic flow in transit, the signal and location of the possible existence of large traffic flow are output, and the number of vehicles in the large traffic flow in transit and the number of other vehicles in transit are obtained. The number of vehicles in the large traffic flow in transit is recorded as N onway , the number of other vehicles on the way is recorded as N onwayl Otherwise, the tracking of the large traffic flow is abandoned and it is considered that there is no large traffic flow on the way at this intersection; Step 4-4: Use the vehicle feature matrix of the OUT zone of this intersection to obtain the driving direction and the number of vehicles waiting to be released in each direction within the monitoring area. The number of vehicles waiting to be released in the lane is recorded as N await .

4. The method for multi-directional dynamic control of road traffic based on vehicle flow monitoring according to claim 1, characterized in that: The process also includes step 8, which is to obtain the location of special vehicles and control traffic at multiple intersections, specifically including the following steps: Step 8-1: Determine whether there is a special vehicle at this intersection. After dividing the intersection monitoring area into K areas, obtain the special vehicle information by directly reading the vehicle marking information or through the vehicle feature matrix. If a vehicle is marked or identified as a special vehicle by features, obtain the location information, speed information, driving route and departure time information of the special vehicle. Let the total time from the current time to the departure time be T z , Step 8-2: Calculate the green light release conditions for the N intersections ahead. Set the N intersections ahead to participate in the green light release control, which is determined by the driving route. After obtaining the position of the special vehicle and the vehicle feature matrix of the monitoring area of ​​the N intersections ahead, if the special vehicle has no departure time information and is already on the road, then calculate the time for the special vehicle to reach the stop line of the i-th intersection ahead based on the speed and position information of the special vehicle, and set it as ST i , 0≤i≤N, the 0th intersection is the current intersection. If the special vehicle has departure time information, the vehicle speed is calculated using the set fixed speed. Then, the total number of vehicles ahead of the special vehicle in the monitoring area of ​​the i-th intersection is calculated. Using the method for predicting travel time in step 5, the total travel time of the vehicles ahead in the monitoring area of ​​the i-th intersection is calculated and set as AT i Finally, set the time reservation item to γ. If the special vehicle does not have departure time information, when ST i ≤AT i +γ, the i-th intersection immediately releases the green light. If the special vehicle has departure time information, the green light release needs to be postponed by T; Step 8-3: Perform traffic control at multiple intersections. When there is a special vehicle at this intersection, control is performed at multiple intersections by simultaneously calculating whether the release conditions of the N intersections ahead are met.

Images