Plane intersection traffic investigation method and system based on unmanned aerial vehicle video

Abstract

The invention relates to a plane intersection traffic investigation method and system based on an unmanned aerial vehicle video, and the method comprises the steps: firstly carrying out the photographing of an investigated plane intersection through an unmanned aerial vehicle, obtaining a video, extracting the vehicle track information in the video through a computer vision technology, extracting the corresponding road structure information from the vehicle track information, and carrying out the calculation of the road structure information. And finally, building a simulation road network by using the extracted vehicle track information and road structure information, restoring an actual road scene, and realizing traffic investigation of the plane intersection. Compared with a traditional traffic investigation method, the method is simpler, manpower and material resources are saved, more visual, intelligent and large-scale plane intersection traffic information can be obtained, and technical support can be provided for follow-up establishment of a traffic demand prediction model and formulation of a reasonable traffic management plan.

Description

Technical Field

[0001] The present application relates to the field of intelligent transportation, and specifically, to a method and system for surveying plane intersection traffic based on drone video. Background Art

[0002] At-grade intersections are scattered in every corner of the city. As the convergence point of various traffic flows on urban roads, they bear the important function of connecting the entire road network. Their operating efficiency is crucial to the operation level of the entire city's road network. Therefore, traffic surveys on at-grade intersections are of great significance in helping the government and traffic management departments to formulate reasonable traffic plans and traffic flow control measures, improve traffic efficiency, and reduce congestion and accident rates.

[0003] Traffic survey is an important part of the feasibility study of highway construction projects; its purpose is to understand the characteristics and composition of highway traffic volume in the project area, to master data such as highway traffic flow, direction and vehicle composition, to provide basic data for future traffic volume forecasts for planned highways, and also to provide a reliable basis for economic evaluation and highway design.

[0004] There are generally manual counting, questionnaire, video observation, electronic monitoring and other methods for flat intersection traffic surveys, which can be mainly divided into manual methods and methods using various sensor technologies for detection. Manual methods have large errors, high labor costs, and severely limited efficiency. The methods of using sensors for detection are to place sensors on gantry or bury them in the road surface. Basically, a set of sensor equipment is fixed to survey a flat intersection. It is a fixed traffic survey method with insufficient flexibility, single equipment function, and low equipment resource utilization.

[0005] In recent years, domestic and foreign studies using drone videos as a source of traffic data have proven that drone videos can be well applied in the field of transportation. Research in the fields of road traffic monitoring, traffic accident detection, determining road service levels, vehicle flow and pedestrian flow detection are constantly developing. Under permitted conditions, drones can collect high-definition videos of intersection traffic scenes and obtain comprehensive, real-time and efficient intersection traffic information from the captured videos.

[0006] The research on vehicle trajectory information extraction based on drone videos is becoming increasingly mature, and the efficiency of detecting and tracking vehicle targets in drone videos is constantly improving. However, there is still room for improvement in the research on extracting vehicle trajectory information while eliminating the influence of drone video jitter. At present, most of the road structure information extraction methods based on vehicle trajectory information are oriented towards GPS vehicle trajectory data. Compared with GPS vehicle trajectory data, the vehicle trajectory data extracted from drone videos has a higher collection frequency, more dense vehicle trajectory points, can be subdivided to the lane level, and has higher accuracy. However, there are few studies on extracting road information using high-frequency vehicle trajectory data obtained from drone videos, and this aspect still faces challenges. Summary of the invention

[0007] In order to overcome at least one shortcoming in the prior art, the present application provides a planar intersection traffic survey method and system based on drone video.

[0008] In a first aspect, a planar intersection traffic survey method based on drone video is provided, comprising:

[0009] Step S1, obtaining a plane intersection video sequence shot by a drone;

[0010] Step S2, processing the plane intersection video sequence using computer vision technology to extract vehicle trajectory information;

[0011] Step S3, using the spatial features and attribute features contained in the vehicle trajectory information to extract the road structure information of the plane intersection; the road structure information includes the road network structure at the lane level, the length of the red light, the location information of the waiting area, and the traffic conflict point information;

[0012] Step S4, building an intersection simulation road network model according to the vehicle trajectory information and the road structure information, and implementing a traffic survey of the plane intersection based on the intersection simulation road network model.

[0013] In one embodiment, in step S2, the planar intersection video sequence is processed using computer vision technology to extract vehicle trajectory information, including:

[0014] Step S21, selecting a reference image in a plane intersection video sequence;

[0015] Step S22, performing image registration on the current frame image in the plane intersection video sequence and the reference image to obtain a homography transformation matrix;

[0016] Step S23, extracting the vehicle trajectory in the current frame image using a vehicle target detection and tracking algorithm; and processing the vehicle trajectory using a homography transformation matrix to obtain a processed vehicle trajectory;

[0017] Step S24, repeating steps S22 and S23 until the plane intersection video sequence is traversed to obtain the final vehicle trajectory information.

[0018] In one embodiment, in step S3, the road structure information of the plane intersection is extracted by using the spatial features and attribute features contained in the vehicle trajectory information, including:

[0019] Step S31, preprocessing the vehicle trajectory information to delete abnormal vehicle trajectories and obtain preprocessed vehicle trajectory information;

[0020] Step S32, using a divisive hierarchical clustering method for each vehicle trajectory in the preprocessed vehicle trajectory information to determine the driving direction of the vehicle trajectory; the vehicle trajectories in the same driving direction constitute a vehicle trajectory data set in the driving direction, and using an agglomerative hierarchical clustering method for each vehicle trajectory data set in the driving direction to determine a representative vehicle trajectory set in the driving direction, that is, a lane-level road network structure;

[0021] Step S33, determining the vehicle trajectory in each driving direction at the stop line according to the stop point information set at the intersection stop line; calculating the vehicle's driving speed according to the vehicle trajectory; analyzing the vehicle's driving speed to determine the red light duration and waiting area position information in each driving direction;

[0022] Step S34: determining the traffic conflict point information in each driving direction according to the representative vehicle trajectory set.

[0023] In one embodiment, in step S31, the abnormal vehicle trajectory includes a redundant vehicle trajectory and a vehicle trajectory with an abnormal length;

[0024] Redundant vehicle trajectories are trajectories where the number of times the vehicle is repeatedly collected by the drone at the stop position is greater than the set number of times. The set number of times is calculated using the following formula:

[0025] C Δt =Δt×fps

[0026] Among them, C Δt is the set number of times, Δt is the sampling interval, and fps is the video frame rate obtained by the drone;

[0027] The vehicle trajectory with abnormal length is the vehicle trajectory whose trajectory length is outside the set range. The set range is calculated using the following formula:

[0028] f = fps × t

[0029] Among them, f is the set range, fps is the video frame rate obtained by the drone, and t is the duration of vehicle passage at the flat intersection.

[0030] In one embodiment, in step S32, a split hierarchical clustering method is used for each vehicle trajectory in the preprocessed vehicle trajectory information to determine the driving direction of the vehicle trajectory, including:

[0031] Calculate the difference △x between the horizontal coordinate of the terminal pixel and the horizontal coordinate of the starting pixel of the vehicle trajectory, and the difference △y between the vertical coordinate of the terminal pixel and the vertical coordinate of the starting pixel;

[0032] Determine whether |Δx|<χ or |Δy|<χ holds true, where χ is the threshold. If so, the vehicle trajectory is in the straight direction, and then the direction of the vehicle trajectory is determined based on the positive or negative value of △x or △y. If not, calculate the convexity θ of the vehicle trajectory, and then determine the direction of the vehicle trajectory based on the positive or negative value of △x, △y and θ.

[0033] In one embodiment, in step S32, for each vehicle trajectory data set in each driving direction, an agglomerative hierarchical clustering method is used to determine a representative vehicle trajectory set in the driving direction, including:

[0034] Step S321, input an empty set Re for storing output results;

[0035] Step S322: Input a vehicle trajectory data set TR in the driving direction = {T1, T2, ... T I …T N} and the empty set Rs, T I is the I-th vehicle trajectory, 1≤I≤N, N is the number of vehicle trajectories in the vehicle trajectory dataset; and the symmetric segmented path distance d between any two vehicle trajectories is calculated, and all symmetric segmented path distances d constitute the symmetric segmented path distance set D = (d -2 ,d -3 ,…d -N ,d 23 ,d 24 ,…d (N-1)N );

[0036] Step S323: Input the vehicle trajectory T in the vehicle trajectory dataset TR I The corresponding symmetric segmented path distance set D I , that is, D I =(d 1I ,d 2I ,…d (I-1)I ,d (I+1)I …d NI );

[0037] Step S324: Determine D I Does d exist in VI ≤5, where V∈[1,N] and V≠I, V represents the set of vehicle trajectories that meet the conditions, V=(V1,V2,…Vm ), m is the number of track labels in V. If so, replace V1, V2, … V m And the vehicle trajectory corresponding to I Merge into new clusters Add the new cluster to the empty set Rs and delete the cluster from TR If not, delete T in TR directly. I ;

[0038] Step S325: Determine whether the vehicle trajectory data set TR is an empty set. If not, select a new vehicle trajectory from TR as T. I , return to step S323; if yes, proceed to step S326;

[0039] Step S326: Determine whether there is only one cluster in Re or it is an empty set. If not, take Re as the new Re and take Re as the new vehicle trajectory data set TR, and return to step S323; if so, output the current Re as the final clustering result, that is, the current Re is the representative vehicle trajectory set in the driving direction.

[0040] In one embodiment, in step S33, determining the duration of the red light in each driving direction includes:

[0041] Step S331: Input the parking point information set K={K1, K2…K i ...K8}, K i Represents the parking point information at the stop line in the i-th driving direction, and at the same time inputs an empty set R for storing the results;

[0042] Step S332: Input the parking point information K at a stop line in a driving direction in K i =(x i ,y i ,id i ), (x i ,y i ) indicates the parking location information, id i The vehicle number indicating the i-th driving direction;

[0043] Step S333: Find the ID i The corresponding vehicle trajectory T, and calculate the vehicle's speed v;

[0044] Step S334: Determine whether the parking point corresponding to v=0 is unique. If so, it means that there is no waiting area in the driving direction. Calculate the corresponding red light duration R according to the parking time of the vehicle at the stop line. i If not, it means that there is a waiting area in the driving direction, and the sum of the two stop times needs to be calculated to get the corresponding red light duration R i ;

[0045] Step S335: Set the red light duration R i Add to R, determine K i Have you traversed all K? If not, delete K from K. i , and select a new K from K i As input, return to step S332, and if so, output the current R, that is, the set of red light durations in each driving direction.

[0046] In one embodiment, in step S33, determining the position information of the waiting area in each driving direction includes:

[0047] Step a: Input the parking point information set K = {K1, K2…K i ...K8}, K i Represents the parking point information at the stop line in the i-th driving direction, and inputs an empty set W for storing the results;

[0048] Step b: Input the parking point information K at the parking line in one driving direction in K o (x i ,y i ,id i ), (x i ,y i ) indicates the parking location information, id i The vehicle number indicating the i-th driving direction;

[0049] Step c: Find the id i The corresponding vehicle trajectory T, and calculate the vehicle's travel speed v;

[0050] Step d: Determine whether the parking point position corresponding to v = 0 is unique. If so, it means that there is no waiting area in this direction. If not, it means that there is a waiting area in the driving direction. Then add the two parking point position information (x1, y1) and (x2, y2) to the set W of the waiting area position information in the i-th driving direction. i , and W i Add to W;

[0051] Step e: Determine K i Have you traversed all K? If not, delete K from K. i , and select a new K from K i As input, return to step b, if yes, then output the current W, that is, the set of waiting area position information in each driving direction.

[0052] In one embodiment, in step S34, determining the traffic conflict point information in each driving direction according to the representative vehicle trajectory set includes:

[0053] Step S341: Input representative vehicle trajectory set RE = {T′1, T′2…T′ i …T′ g}, T′ i is the i-th representative vehicle trajectory, 1≤i≤g, g is the number of representative vehicle trajectories, and the input is the empty set I used to store the output results;

[0054] Step S342: Input representative vehicle trajectory Y′ i , calculate and obtain T′ i Traffic conflict point information set I i ;

[0055] Step S343: Determine T′ i Is the driving direction right? If so, directly turn I i Add to I, if not, first extract the trajectory T′ i The representative vehicle trajectories that cannot intersect in time are recorded as time-independent representative vehicle trajectories, and I is deleted. i The traffic conflict point information related to the time-disjoint representative vehicle trajectory in the updated I i Add to I;

[0056] Step S344: Determine whether all representative vehicle trajectories in RE have been traversed. If not, select the untraversed representative vehicle trajectory as the new T′ i Input, return to step S342, if yes, output the current I, that is, the set of traffic conflict point information including each driving direction.

[0057] In one embodiment, in step S4, a simulated road network model of an intersection is constructed according to the vehicle trajectory information and the road structure information, and a traffic survey of the plane intersection is performed based on the simulated road network model of the intersection, including:

[0058] Step S41, using the road structure information to customize the road network in the plane intersection simulation model;

[0059] Step S42, using the vehicle trajectory information to customize the vehicle flow in the plane intersection simulation model;

[0060] Step S43, building a plane intersection simulation model, restoring the actual road scene, realizing the traffic survey of the plane intersection, and obtaining the traffic survey results of the plane intersection.

[0061] In a second aspect, a plane intersection traffic survey system based on drone video is provided, comprising:

[0062] A video sequence acquisition module is used to acquire the plane intersection video sequence taken by the drone;

[0063] The vehicle trajectory information extraction module is used to process the plane intersection video sequence using computer vision technology to extract vehicle trajectory information;

[0064] The road structure information extraction module is used to extract the road structure information of the plane intersection by using the spatial features and attribute features contained in the vehicle trajectory information; the road structure information includes the road network structure at the lane level, the duration of the red light, the location information of the waiting area, and the information of the traffic conflict point;

[0065] The traffic survey module is used to build an intersection simulation road network model according to vehicle trajectory information and road structure information, and to implement traffic survey of the plane intersection based on the intersection simulation road network model.

[0066] Compared with the prior art, this application has the following beneficial effects:

[0067] 1. Compared with traditional manual traffic surveys and sensor detection traffic surveys, using drone videos for traffic surveys is more convenient, reduces labor costs and equipment maintenance costs, has greater flexibility and mobility, and high resource utilization.

[0068] 2. Using image registration and vehicle target detection and tracking algorithms in computer vision technology to extract vehicle trajectory information from drone videos can eliminate the impact of drone video jitter on vehicle trajectory information extraction and effectively correct the vehicle trajectory deviation caused by drone jitter.

[0069] 3. The high-frequency and dense vehicle trajectory information obtained from the drone video can be divided into lane levels, and the road structure information of the plane intersection is effectively extracted by using the spatial characteristics and attribute characteristics of the vehicle trajectory information.

[0070] 4. This application is automated and intelligent, has strong flexibility, consumes less manpower and material resources, and can effectively realize traffic surveys at flat intersections. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] The present application may be better understood by referring to the following description given in conjunction with the accompanying drawings, which together with the following detailed description are included in this specification and form a part of this specification. In the drawings:

[0072] Figure 1 A flowchart of a planar intersection traffic survey method based on drone video is shown;

[0073] Figure 2 A flowchart of extracting vehicle trajectory information is shown;

[0074] Figure 3A flowchart of obtaining the total number of vehicle trajectories affected by drone jitter is shown;

[0075] Figure 4 The diagram shows the southeast, northwest and north-south direction signs of the plane intersection and the coordinate axes of the pixel coordinate system;

[0076] Figure 5 A flow chart of determining the direction of travel of a vehicle trajectory is shown;

[0077] Figure 6 A flowchart for determining a representative set of vehicle trajectories in a driving direction is shown;

[0078] Figure 7 A schematic diagram of representative vehicle driving trajectory information is shown;

[0079] Figure 8 A flowchart for determining the duration of a red light in each direction of travel is shown;

[0080] Fig. 9 A flowchart of determining the position information of the waiting area in each driving direction is shown;

[0081] Fig.10 A flowchart of determining traffic conflict point information in each driving direction is shown;

[0082] Fig.11 A schematic diagram of traffic conflict point information at a flat intersection is shown;

[0083] Fig.12 The flowchart of defining the road network file is shown;

[0084] Fig.13 The constructed plane intersection simulation road network model is shown;

[0085] Fig.14 The structural block diagram of the planar intersection traffic survey device based on drone video is shown. DETAILED DESCRIPTION

[0086] The exemplary embodiments of the present application will be described below in conjunction with the accompanying drawings. For the sake of clarity and conciseness, not all features of the actual embodiments are described in the specification. However, it should be understood that many implementation-specific decisions can be made in the process of developing any such actual embodiments in order to achieve the specific goals of the developer, and these decisions may vary from embodiment to embodiment.

[0087] It is also necessary to explain here that, in order to avoid obscuring the present application due to unnecessary details, only the device structure closely related to the scheme according to the present application is shown in the drawings, while other details that are not closely related to the present application are omitted.

[0088] It should be understood that the present application is not limited to the described implementation forms due to the following description with reference to the accompanying drawings. In this article, where feasible, the embodiments can be combined with each other, features between different embodiments can be replaced or borrowed, and one or more features can be omitted in one embodiment.

[0089] The present application provides a method for surveying traffic at a plane intersection based on drone video. Figure 1 A flowchart of a planar intersection traffic survey method based on drone video is shown in FIG. Figure 1 , methods include:

[0090] Step S1, obtaining a plane intersection video sequence shot by a drone.

[0091] First, determine the traffic survey task and the specific survey location; the survey start and end time is: 7:30-8:30; the hovering position is: 75m above the center of the plane intersection; the external weather environment is: sunny but not strong, no clouds, no fog, and light wind; take pictures and obtain drone videos. Then, use an aerial photography drone, the shooting position is 75m above the center of the plane intersection, and obtain the drone video sequence.

[0092] Step S2, using computer vision technology to process the plane intersection video sequence to extract vehicle trajectory information.

[0093] Here, computer vision technology is used to eliminate the vehicle trajectory deviation caused by the jitter of the drone video. The vehicle trajectory deviation phenomenon refers to an inevitable problem in the drone shooting process. The drone is affected by factors such as battery replacement, external wind and its own gimbal, and the shooting position often changes. The change of shooting position affects the change of the drone shooting range, resulting in the change of the reference coordinate system of the image in the drone video. The extracted vehicle trajectory will have a significant deviation phenomenon.

[0094] Step S3, using the spatial features and attribute features contained in the vehicle trajectory information, extracting the road structure information of the plane intersection; the road structure information includes the road network structure at the lane level, the length of the red light, the location information of the waiting area, and the traffic conflict point information.

[0095] Step S4, building an intersection simulation road network model according to vehicle trajectory information and road structure information, and obtaining traffic survey results of the flat intersection based on the intersection simulation road network model; the traffic survey results include the total traffic volume of the flat intersection, the traffic volume of each lane, the average vehicle speed, the average waiting time and the average delay time.

[0096] In one embodiment, the computer vision technology in step S2 includes an image registration algorithm and a vehicle target detection and tracking algorithm, the image registration algorithm includes a SIFT feature point detection algorithm and a RANSAC feature point matching algorithm, and the vehicle target detection and tracking algorithm includes YOLOv5 and Deepsort vehicle detection and tracking. Figure 2 A flowchart of extracting vehicle trajectory information is shown, see Figure 2 In step S2, the plane intersection video sequence is processed using computer vision technology to extract vehicle trajectory information, including:

[0097] Step S21, selecting a reference image in a plane intersection video sequence.

[0098] Here, the reference image is selected by selecting the center area of ​​the plane intersection contained in each frame of the drone video image as the reference image. In this embodiment, the center area of ​​the plane intersection of the first frame image in the drone video is selected as the reference image.

[0099] Step S22, performing image registration on the current frame image in the plane intersection video sequence and the reference image to obtain a homography transformation matrix; here, the image registration algorithm includes a SIFT feature point detection algorithm and a RANSAC feature point matching algorithm.

[0100] Step S23, extracting the vehicle track in the current frame image using a vehicle target detection and tracking algorithm; and processing the vehicle track using a homography transformation matrix to obtain a processed vehicle track. Here, the vehicle target detection and tracking algorithm includes YOLOv5 and Deepsort vehicle detection and tracking.

[0101] Step S24, repeating steps S22 and S23 until the plane intersection video sequence is traversed to obtain the final vehicle trajectory information.

[0102] In order to verify the effect of extracting vehicle trajectory information in this embodiment, the following analysis was performed:

[0103] First, the vehicle trajectory data of the intersection affected by the obvious jitter amplitude of the drone within 5 minutes was selected, and the correction success rate of the vehicle trajectory deviation phenomenon was calculated. The effect of extracting vehicle trajectory data by eliminating the jitter of drone video is evaluated as shown in the following formula:

[0104]

[0105] Among them, Nu I is the total number of vehicle trajectories affected by the UAV jitter during this time, Nu Cis the total number of vehicle trajectories after the vehicle trajectory deviation phenomenon is effectively corrected. The number of vehicle trajectories affected by the UAV jitter is Nu I It can be obtained from the homography transformation matrix. When the homography transformation matrix is ​​the unit matrix, the reference coordinate system of the detection image is consistent with that of the reference image, that is, all vehicle tracks in the detection image are not affected by the jitter of the drone. Figure 3 The flowchart of obtaining the total number of vehicle trajectories affected by drone jitter is shown. The overall steps include:

[0106] Step K1: Import the drone video. Assume that the current detection is the i-th frame of the drone video. Get the homography transformation matrix H between the i-th frame of the drone video image and the reference image. i ;

[0107] Step K2: Determine H i Is it a unit matrix? If not, store all vehicle IDs detected in the frame image into the array NU. If yes, continue to detect the next frame directly.

[0108] Step K3: Determine whether it is the last frame image. If not, continue to detect the next frame image. If yes, delete all repeated vehicle IDs in NU and count the number of elements contained in NU, which is Nu. I , output Nu I .

[0109] Then, it is necessary to obtain the vehicle trajectory number Nu in which the vehicle trajectory deviation phenomenon is effectively corrected C , count the vehicle IDs contained in the array NU finally obtained in step K3, and count and filter out the total number of vehicle trajectories after the vehicle trajectory deviation phenomenon is effectively corrected, and you can get Nu C In this embodiment, the vehicle trajectory data of a certain plane intersection with a significant jitter amplitude of the drone within 5 minutes is taken as an example. I , Nu C The values ​​are 479vehs and 443vehs respectively. The correcting success rate of vehicle trajectory deviation can be calculated as It can reach 92.48%, indicating that the combination of image registration algorithm and vehicle target detection and tracking algorithm can effectively eliminate the influence of UAV video screen jitter on vehicle trajectory data extraction, effectively correct the abnormal deviation of vehicle trajectory, and can extract and eliminate the vehicle trajectory information affected by UAV video screen jitter.

[0110] In one embodiment, in step S3, the road structure information of the plane intersection is extracted by using the spatial features and attribute features contained in the vehicle trajectory information, including:

[0111] Step S31, preprocessing the vehicle trajectory information to delete abnormal vehicle trajectories and obtain preprocessed vehicle trajectory information.

[0112] Step S32, a divisive hierarchical clustering method is used for each vehicle trajectory in the preprocessed vehicle trajectory information to determine the driving direction of the vehicle trajectory; the vehicle trajectories in the same driving direction constitute a vehicle trajectory data set in the driving direction, and an agglomerative hierarchical clustering method is used for the vehicle trajectory data set in each driving direction to determine a representative vehicle trajectory set in the driving direction, that is, a lane-level road network structure.

[0113] Step S33, based on the parking point information set at the intersection stop line, determine the vehicle trajectory in each driving direction at the stop line; calculate the vehicle's driving speed based on the vehicle trajectory; analyze the vehicle's driving speed to determine the red light duration and waiting area position information in each driving direction.

[0114] Step S34: determining the traffic conflict point information in each driving direction according to the representative vehicle trajectory set.

[0115] Specifically, in step S31, abnormal vehicle trajectories include redundant vehicle trajectories and vehicle trajectories with abnormal lengths. When a vehicle stays at the same location for too long due to a red light, pedestrians or other reasons, the vehicle trajectory data at that location will be redundant; in drone video, some vehicle trajectory data may be missing due to factors such as trees and buildings, or when the models, colors and other features of two vehicles are too similar, the two vehicles may be identified as one, which will cause a sudden change in vehicle trajectory data, causing the vehicle trajectory to disappear prematurely or continue too long.

[0116] Redundant vehicle trajectories are trajectories where the number of times the vehicle is repeatedly collected by the drone at the stop position is greater than the set number of times. The set number of times is calculated using the following formula:

[0117] C Δt =Δt×fps

[0118] Among them, C Δt is the set number of times, Δt is the sampling interval, and fps is the video frame rate obtained by the drone. In this embodiment, Δt = 2s, fps = 29.97 frames / s, then the calculated C Δt =60.

[0119] The vehicle trajectory with abnormal length is the vehicle trajectory whose trajectory length is outside the set range. The set range is calculated using the following formula:

[0120] f = fps × t

[0121] Among them, f is the set range, fps is the video frame rate obtained by the drone, and t is the duration of vehicle passage at the flat intersection.

[0122] In this embodiment, the extracted vehicle trajectory data is obtained from the drone video image after scaling by half, in the first frame of the drone video, Figure 4 The diagram shows the southeast, northwest and north-south direction signs of the plane intersection and the coordinate axis diagram of the pixel coordinate system. The resolution of the drone video image after scaling by half is R = 960 × 540. The ground resolution GSD of the drone video image is 10 cm. According to the ground resolution GSD, the actual distance length and width L × W of the drone video image can be calculated as shown in the following formula:

[0123] L×W=R×GSD×0.02=192m×108m

[0124] In this embodiment, the actual length of the drone video shooting range is 192m and the width is 108m. According to the Regulations for the Implementation of the Road Traffic Safety Law of the People's Republic of China, the maximum speed of vehicles passing through the intersection shall not exceed 30km / h, that is, v max =30km / h, in this embodiment, it is set is the average speed of vehicles passing through the intersection without waiting for the red light. The maximum time t for vehicles to pass through the intersection without waiting for the red light can be calculated. max With the minimum duration t min As shown below:

[0125]

[0126] in, for The minimum value is 20km / h, v max for The maximum value is 25km / h.

[0127] In this embodiment, considering the constraints between vehicles in the actual traffic flow and the influence of pedestrians and non-motor vehicles on the speed of vehicles at the intersection, the vehicle travel time through the plane intersection is t∈[8s,35s], and the video frame rate fps obtained by the drone used in this embodiment is 29.97 frames / s. Therefore, the range of normal vehicle trajectory data collection times in this embodiment is as follows:

[0128] f∈fps×t≈[240,1050]

[0129] Wherein, f is the length range of normal vehicle trajectory data in this embodiment. According to the above formula, in this embodiment, it is set to eliminate the vehicle trajectory data with the number of collections, that is, the trajectory length range outside [240,1050].

[0130] In one embodiment, Figure 5 A flowchart for determining the direction of travel of a vehicle trajectory is shown in FIG. Figure 5In step S32, a split hierarchical clustering method is used for each vehicle trajectory in the preprocessed vehicle trajectory information to determine the driving direction of the vehicle trajectory, including:

[0131] First, the difference △x between the horizontal coordinate of the terminal pixel and the horizontal coordinate of the starting pixel of the vehicle trajectory, and the difference △y between the vertical coordinate of the terminal pixel and the vertical coordinate of the starting pixel;

[0132] Here, the starting pixel coordinates are (x0, y0) and the ending pixel coordinates are (x t ,y t ), then Δx=x t -x0, Δy=y t -y0.

[0133] Then, determine whether |Δx|<χ or |Δy|<χ is true, χ is the threshold value. If so, the vehicle trajectory is in the straight direction, and then the direction of the vehicle trajectory is determined according to the positive or negative value of △x or △y. If not, calculate the concavity θ of the vehicle trajectory, and then determine the direction of the vehicle trajectory according to the positive or negative value of △x, △y and θ. The concavity θ of the vehicle trajectory is determined by the following formula:

[0134]

[0135] Where f(x) represents the function of the vehicle trajectory, f(x0) is the trajectory function value when x=x0, and f(x t ) is x=x t The trajectory function value at .

[0136] In this embodiment, since the drone shooting position is located above the center of the plane intersection, one of the △x and △y of the straight vehicle trajectory must be much smaller than the △x and △y of the turning vehicle trajectory. In this embodiment, a threshold value χ is set to determine whether the vehicle trajectory is straight. The threshold value χ is calculated using the following formula:

[0137]

[0138] Lane width at intersection v It is generally 3m, and the maximum number of lane changes for straight-moving vehicles is 3.

[0139] In this embodiment, the ground resolution GSD of the drone video image is 10 cm, and θ is the concavity of the vehicle's driving trajectory. When θ<0, the vehicle trajectory is convex, otherwise it is concave.

[0140] To determine the specific driving direction of the vehicle trajectory, please refer to Table 1.

[0141] Table 1 Division of vehicle trajectories in each driving direction

[0142]

[0143] In one embodiment, Figure 6 A flowchart of determining a representative set of vehicle trajectories in a driving direction is shown, see Figure 6 In step S32, for each vehicle trajectory data set in each driving direction, an agglomerative hierarchical clustering method is used to determine a representative vehicle trajectory set in the driving direction, including:

[0144] Step S321, input an empty set Re for storing output results;

[0145] Step S322: Input a vehicle trajectory data set TR in the driving direction = {T1, T2, ... T I …T N} and the empty set Rs, T I is the I-th vehicle trajectory, 1≤I≤N, N is the number of vehicle trajectories in the vehicle trajectory dataset; and the symmetric segmented path distance d between any two vehicle trajectories is calculated, and all symmetric segmented path distances d constitute the symmetric segmented path distance set D = (d 12 ,d 13 ,…d 1N ,d 23 ,d 24 ,…d (N-1)N ); Here, the process of determining the symmetrical segmented path distance d between any two vehicle trajectories is a prior art and will not be described in detail. Table 2 shows the two vehicle trajectories (T 1 ,T 2 ) between the symmetric segmented paths SSPD (T 1 ,T 2 ) calculation results.

[0146] Table 2D SSPD (T 1 ,T 2 )

[0147]

[0148] In this embodiment, α=5, β=40 in Table 2.

[0149] Step S323: Input the vehicle trajectory T in the vehicle trajectory dataset TR I The corresponding symmetric segmented path distance set D I , that is, D I =(d 1I ,d 2I ,…d (I-1)I ,d (I+1)I …d NI ).

[0150] Step S324: Determine D I Does d exist in VI ≤5, where V∈[1,N] and V≠I, V represents the set of vehicle trajectories that meet the conditions, V=(V1,V2,…V m ), m is the number of track labels in V. If so, replace V1, V2, ... V m And the vehicle trajectory corresponding to I Merge into new clusters Add the new cluster to the empty set Rs and delete the cluster from TR If not, delete T in TR directly. I ;

[0151] Step S325: Determine whether the vehicle trajectory data set TR is an empty set. If not, select a new vehicle trajectory from TR as T. I , return to step S323; if yes, proceed to step S326;

[0152] Step S326: Determine whether there is only one cluster or an empty set in Rs. If not, take Rs as the new Re and take Rs as the new vehicle trajectory data set TR, and return to step S323; if so, output the current Re as the final clustering result, that is, the current Re is the representative vehicle trajectory set in the driving direction. Figure 7 A schematic diagram of representative vehicle driving trajectory information is shown.

[0153] In one embodiment, Figure 8 A flowchart of determining the duration of a red light in each direction of travel is shown, see Figure 8 In step S33, determining the duration of the red light in each driving direction includes:

[0154] Step S331: Input the parking point information set K={K1, K2…K i ...K8}, K i Represents the parking point information at the stop line in the i-th driving direction, and at the same time inputs an empty set R for storing the results;

[0155] Step S332: Input the parking point information K at a stop line in a driving direction in K i =(x i ,y i ,id i ), (x i ,y i ) indicates the parking location information, id i The vehicle number indicating the i-th driving direction;

[0156] Step S333: Find the ID iThe corresponding vehicle trajectory T, and calculate the vehicle's speed v;

[0157] Step S334: Determine whether the parking point corresponding to v=0 is unique. If so, it means that there is no waiting area in the driving direction. Calculate the corresponding red light duration R according to the parking time of the vehicle at the stop line. i If not, it means that there is a waiting area in the driving direction, and the sum of the two stop times needs to be calculated to get the corresponding red light duration R i ;

[0158] Step S335: Set the red light duration R i Add to R, determine K i Have you traversed all K? If not, delete K from K. i , and select a new K from K i As input, return to step S332, and if so, output the current R, that is, the set of red light durations in each driving direction.

[0159] In this embodiment, a certain plane intersection is taken as the investigation object, and the red light duration in each driving direction is obtained as shown in Table 3, with a total of 8 driving directions.

[0160] Table 3 Red light duration information table

[0161]

[0162] In one embodiment, Fig. 9 A flowchart of determining the position information of the waiting area in each driving direction is shown. Fig. 9 In step S33, the position information of the waiting area in each driving direction is determined, including:

[0163] Step a: Input the parking point information set K = {K1, K2…K i ...K8}, K i Represents the parking point information at the stop line in the i-th driving direction, and inputs an empty set W for storing the results;

[0164] Step b: Input the parking point information K at the parking line in one driving direction in K i (x i ,y i ,id i ), (x i ,y i ) indicates the parking location information, id i The vehicle number indicating the i-th driving direction;

[0165] Step c: Find the id i The corresponding vehicle trajectory T, and calculate the vehicle's travel speed v;

[0166] Step d: Determine whether the parking point position corresponding to v = 0 is unique. If so, it means that there is no waiting area in this direction. If not, it means that there is a waiting area in the driving direction. Then add the two parking point position information (x1, y1) and (x2, y2) to the set W of the waiting area position information in the i-th driving direction. i , and W i Add to W;

[0167] Step e: Determine K i Have you traversed all K? If not, delete K from K. i , and select a new K from K i As input, return to step b, if yes, then output the current W, that is, the set of waiting area position information in each driving direction.

[0168] In one embodiment, Fig.10 A flowchart of determining traffic conflict point information in each driving direction is shown, see Fig.10 In step S34, traffic conflict point information in each driving direction is determined according to the representative vehicle trajectory set, including:

[0169] Step S341: Input representative vehicle trajectory set RE = {T′1, T′2…T′ i …T′ g}, T′ i is the i-th representative vehicle trajectory, 1≤i≤g, g is the number of representative vehicle trajectories, and the input is the empty set I used to store the output results;

[0170] Step S342: Input representative vehicle trajectory T′ i , calculate and obtain T′ i Traffic conflict point information set I i ;

[0171] Step S343: Determine T′ i Is the driving direction right? If so, directly turn I i Add to I, if not, first extract the trajectory T′ i The representative vehicle trajectories that cannot intersect in time are recorded as time-independent representative vehicle trajectories, that is, i Different phases and trajectory T′ i Representative vehicle trajectory information of non-adjacent phases, and delete I i The traffic conflict point information related to the time-disjoint representative vehicle trajectory in the updated I i Add to I;

[0172] Step S344: Determine whether all representative vehicle trajectories in RE have been traversed. If not, select the untraversed representative vehicle trajectory as the new T′ i Input, return to step S342, if yes, output the current I, that is, the set of traffic conflict point information including each driving direction.

[0173] In this embodiment, step S343 deletes I i The traffic conflict point information related to the time-disjoint representative vehicle trajectory in the , that is, the traffic conflict point information set I i Because each representative trajectory is distributed differently in time, some representative trajectories do not actually have traffic conflict points with each other. East-west straight and north-south straight traffic do not actually exist. As shown in Table 4, the phase states of the two are relatively long and will not cause traffic conflict. Therefore, in step S343, this embodiment first determines the representative trajectory T′ i Is it a right turn? The right turn direction is not affected by the phase state, so there must be a traffic conflict with the representative trajectories in each direction. If not, it is necessary to adjust the direction according to the trajectory T′. i The phase state deletes the abnormal traffic conflict point information.

[0174] In this embodiment, taking the investigation of a certain plane intersection as an example, the extracted traffic conflict point information is plotted in the representative vehicle trajectory, and the traffic conflict point is the corresponding black circular part in the figure. Fig.11 A schematic diagram of traffic conflict point information at a flat intersection is shown.

[0175] Table 4 Traffic signal timing table

[0176]

[0177] In one embodiment, in step S4, traffic simulation software SUMO is used to investigate traffic at the intersection, including:

[0178] Step S41: using the obtained planar intersection road structure information to customize the road network in the planar intersection simulation model.

[0179] Here, the overall process of customizing the road network in the plane intersection simulation model is as follows Fig.12 As shown, Fig.12 The flowchart for defining a road network file is shown; in this embodiment, a corresponding road network is built in SUMO, traffic flow is set, and a plane intersection simulation model is built to restore the actual road scene.

[0180] In this embodiment, the road section under study is a plane intersection, and the drone shooting position is located above the center point of the plane intersection, so five nodes are selected by default to build a simulated road network model. The five nodes are respectively located at the center point of the drone image and the midpoints of the four boundaries of the image. In this embodiment, a plane intersection simulation road network model with the same proportion as the drone video image is built in SUMO, so the horizontal and vertical coordinates of the pixels of these points in the image are the horizontal and vertical coordinates of the nodes, among which the node at the center point is specially defined as a traffic light type node, and the traffic signal timing corresponding to the number is set; in this embodiment, the number of lanes obtained can be used to connect the nodes to set the number of lanes. The edge definition file, each representative driving trajectory information is a set of several position coordinate points in each driving direction. The lane index corresponding to the entry and exit sections is found by using these position coordinate point sets, and the entry and exit lane definition connection files corresponding to each representative driving trajectory information can be connected; in addition, the connection priority needs to be set in the connection file, and the level intersection traffic conflict information obtained above is used for setting; in this embodiment, the obtained waiting area coordinate point information set is used to set the shape and length of the waiting area, and the entry and exit lane index information is set according to the driving route information corresponding to the waiting area.

[0181] Step S42: using the obtained intersection vehicle trajectory information to customize the vehicle flow in the intersection simulation model.

[0182] Step S43: Build a plane intersection simulation model, restore the actual road scene, conduct a traffic survey on the plane intersection, and obtain the traffic survey results of the plane intersection; the traffic survey results include the total traffic volume of the plane intersection, the traffic volume of each lane, the average speed, the average waiting time and the average delay time.

[0183] Fig.13 The constructed plane intersection simulation road network model is shown. The traffic volume in each driving direction in the actual situation and the simulation situation is shown in Table 5 below.

[0184] Table 5 Traffic volume in each direction at Nanshaomen intersection

[0185]

[0186] In this embodiment, the average driving time t of all vehicles in the simulation can be obtained through the data collected in the simulation. avg is 65.91s, and the average waiting time of all vehicles is t wait The average driving time of all vehicles and the average waiting time of all vehicles can be calculated based on the average driving time of all vehicles and the average waiting time of all vehicles. sim , as shown below:

[0187] tsim =t avg -t wait =19.28s

[0188] In this embodiment, it is also necessary to calculate the actual average travel time t of the vehicles at the intersection without waiting for the traffic lights. rea For comparison, the actual average driving time is obtained by multiplying the actual average driving time in each driving direction by the actual traffic volume proportion in each driving direction. The actual average driving time t of vehicles in each driving direction without waiting for traffic lights dir The calculation method is shown in the following formula:

[0189]

[0190] Among them, L dir is the actual driving distance of the vehicle in the driving direction, in meters; is the average speed of the vehicle in the driving direction, in m / s. In this embodiment, representative vehicle trajectory information in each driving direction has been obtained. The representative vehicle trajectory is a set of pixel coordinates of several trajectory points. These trajectory points are connected one by one with a line. The length of this line is converted using the ground resolution of the drone video image to obtain the actual driving distance corresponding to the driving direction. Suppose the representative vehicle trajectory in a driving direction is T dir {(x i ,y i )}, 1≤i≤n, n is the number of vehicle trajectory points, (x i ,y i ) is the coordinate of the i-th trajectory point, then L dir The calculation formula is as follows:

[0191]

[0192] Furthermore, the actual driving distance L of each driving direction at the intersection is obtained. dir Then, the collected free-flow vehicle trajectory data is used to calculate the average speed of free-flow vehicles at the intersection without waiting for traffic lights. After calculation, L dir ,t dir The values ​​are shown in Table 6:

[0193] Table 6 Actual average travel time in each direction at Nanshaomen intersection

[0194]

[0195] After obtaining the actual average travel time t of each driving direction at the intersection dirThen, the actual average travel time t of the vehicles at the intersection without waiting for the traffic lights can be calculated. rea , as shown below:

[0196] t rea =∑ψ dir ×t dir ≈18.09s

[0197] Among them, t rea is the actual average travel time of vehicles at the intersection without waiting for traffic lights, ψ dir is the ratio of the traffic volume in each driving direction to the total traffic volume in actual situations, t dir is the actual average travel time in each direction of the intersection.

[0198] The actual road scene restored in step S43 is evaluated by the average travel time accuracy Φ of the intersection, as shown in the following formula:

[0199]

[0200] Among them, t rea is the actual average travel time of vehicles at the intersection without waiting for traffic lights, t sim It is the average driving time of vehicles passing through the intersection without waiting for traffic lights in the simulation. Taking the Nanshaomen intersection as an example, the accuracy of the average driving time of the intersection under simulation can reach 93.42%, that is, the simulation of the intersection has better restored the traffic volume in each driving direction in the actual situation and the average driving time of vehicles passing through the intersection without waiting for traffic lights, and has better restored the actual road scene of the Nanshaomen intersection.

[0201] Using the same inventive concept as the method for surveying plane intersection traffic based on drone video, this embodiment also provides a corresponding plane intersection traffic survey device based on drone video. Fig.14 The structural block diagram of the plane intersection traffic survey device based on drone video is shown, including:

[0202] The video sequence acquisition module 141 is used to acquire the plane intersection video sequence photographed by the drone;

[0203] The vehicle trajectory information extraction module 142 is used to process the plane intersection video sequence using computer vision technology to extract vehicle trajectory information;

[0204] The road structure information extraction module 143 is used to extract the road structure information of the plane intersection by using the spatial features and attribute features contained in the vehicle trajectory information; the road structure information includes the road network structure at the lane level, the duration of the red light, the location information of the waiting area, and the information of the traffic conflict point;

[0205] The traffic survey module 144 is used to build an intersection simulation network model based on vehicle trajectory information and road structure information, and implement traffic survey on the plane intersection based on the intersection simulation network model. The traffic survey results include the total traffic volume, lane traffic volume, average vehicle speed, average waiting time and average delay time of the plane intersection.

[0206] The drone video-based planar intersection traffic survey device of this embodiment has the same inventive concept as the drone video-based planar intersection traffic survey method described above. Therefore, the specific implementation method of the device can be seen in the implementation example part of the drone video-based planar intersection traffic survey method described above, and its technical effect corresponds to the technical effect of the above-mentioned method, which will not be repeated here.

[0207] In summary, the present application has the following technical effects: 1. Compared with traditional manual traffic surveys and sensor detection traffic surveys, it is more convenient to use drone videos for traffic surveys, which reduces labor costs and equipment maintenance costs, has greater flexibility and mobility, and high resource utilization. 2. The image registration and vehicle target detection and tracking algorithms in computer vision technology are used to extract vehicle trajectory information from drone videos, which can eliminate the impact of drone video screen jitter on vehicle trajectory information extraction and effectively correct the vehicle trajectory deviation phenomenon caused by drone jitter. 3. For the high-frequency and dense vehicle trajectory information obtained from drone videos that can be divided into lane levels, the spatial characteristics and attribute characteristics of the vehicle trajectory information are used to effectively extract the road structure information of the plane intersection. 4. The present application is automated and intelligent, and has strong flexibility, low consumption of manpower and material resources, and effectively realizes traffic surveys of plane intersections.

[0208] The above descriptions are merely various implementations of the present application, but the protection scope of the present application is not limited thereto.

Claims

1. A planar intersection traffic survey method based on drone video, characterized in that: include: Step S1, obtaining a plane intersection video sequence shot by a drone; Step S2, processing the plane intersection video sequence using computer vision technology to extract vehicle trajectory information; Step S3, using the spatial features and attribute features contained in the vehicle trajectory information to extract road structure information of the plane intersection; the road structure information includes the road network structure at the lane level, the duration of the red light, the location information of the waiting area, and the traffic conflict point information; Step S4, building an intersection simulation road network model according to the vehicle trajectory information and the road structure information, and implementing a traffic survey of the plane intersection based on the intersection simulation road network model.

2. The method according to claim 1, characterized in that In step S2, the planar intersection video sequence is processed using computer vision technology to extract vehicle trajectory information, including: Step S21, selecting a reference image in the plane intersection video sequence; Step S22, performing image registration on the current frame image in the plane intersection video sequence and the reference image to obtain a homography transformation matrix; Step S23, extracting the vehicle track in the current frame image using a vehicle target detection and tracking algorithm; and processing the vehicle track using the homography transformation matrix to obtain a processed vehicle track; Step S24, repeating step S22 and step S23 until the plane intersection video sequence is traversed to obtain the final vehicle trajectory information.

3. The method according to claim 1, characterized in that In step S3, the road structure information of the plane intersection is extracted by using the spatial features and attribute features contained in the vehicle trajectory information, including: Step S31, preprocessing the vehicle trajectory information to delete abnormal vehicle trajectories and obtain preprocessed vehicle trajectory information; Step S32, using a divisive hierarchical clustering method for each vehicle trajectory in the preprocessed vehicle trajectory information to determine the driving direction of the vehicle trajectory; the vehicle trajectories in the same driving direction constitute a vehicle trajectory data set in the driving direction, and using an agglomerative hierarchical clustering method for each vehicle trajectory data set in the driving direction to determine a representative vehicle trajectory set in the driving direction, that is, a lane-level road network structure; Step S33, determining the vehicle trajectory in each driving direction at the stop line according to the stop point information set at the intersection stop line; calculating the vehicle's driving speed according to the vehicle trajectory; analyzing the vehicle's driving speed to determine the red light duration and waiting area position information in each driving direction; Step S34: determining traffic conflict point information in each driving direction according to the representative vehicle trajectory set.

4. The method according to claim 3, characterized in that In the step S31, the abnormal vehicle trajectory includes a redundant vehicle trajectory and a vehicle trajectory with an abnormal length; The redundant vehicle trajectory is a trajectory where the number of times the vehicle is repeatedly collected by the drone at the stop position is greater than the set number of times, and the set number of times is calculated using the following formula: C Δt =Δt×fps Among them, C Δt is the set number of times, Δt is the sampling interval, and fps is the video frame rate obtained by the drone; The vehicle trajectory with abnormal length is a vehicle trajectory whose trajectory length is outside a set range, and the set range is calculated using the following formula: f = fps × t Among them, f is the set range, fps is the video frame rate obtained by the drone, and t is the duration of vehicle passage at the flat intersection.

5. The method according to claim 3, characterized in that In step S32, a splitting hierarchical clustering method is used for each vehicle trajectory in the preprocessed vehicle trajectory information to determine the driving direction of the vehicle trajectory, including: Calculate the difference △x between the horizontal coordinate of the terminal pixel and the horizontal coordinate of the starting pixel of the vehicle trajectory, and the difference △y between the vertical coordinate of the terminal pixel and the vertical coordinate of the starting pixel; Determine whether |Δx|<χ or |Δy|<χ holds, where χ is a threshold value. If so, the vehicle trajectory is in a straight direction, and then the driving direction of the vehicle trajectory is determined according to the positive or negative value of △x or △y. If not, calculate the concavity θ of the vehicle trajectory, and then determine the driving direction of the vehicle trajectory according to the positive or negative value of △x, △y and θ.

6. The method according to claim 3, characterized in that In step S32, for each vehicle trajectory data set in each driving direction, an agglomerative hierarchical clustering method is used to determine a representative vehicle trajectory set in the driving direction, including: Step S321, input an empty set Re for storing output results; Step S322: Input a vehicle trajectory data set TR in the driving direction = {T1, T2, ... T I …T N } and the empty set Rs, T I is the I-th vehicle trajectory, 1≤I≤N, N is the number of vehicle trajectories in the vehicle trajectory dataset; and the symmetric segmented path distance d between any two vehicle trajectories is calculated, and all symmetric segmented path distances d constitute the symmetric segmented path distance set D = (d 12 ,d 13 ,…d 1N ,d 23 ,d 24 ,...d (N-1)N ); Step S323: Input the vehicle trajectory T in the vehicle trajectory dataset TR I The corresponding symmetric segmented path distance set D I , that is, D I =(d 1I ,d 2I ,…d (I-1)I ,d (I+1)I …d NI ); Step S324: Determine D I Does d exist in VI ≤5, where V∈[1,N] and V≠I, V represents the set of vehicle trajectories that meet the conditions, V=(V1,V2,…V m ), m is the number of track labels in V. If so, replace V1, V2, … V m And the vehicle trajectory corresponding to I T I Merge into new clusters Add the new cluster to the empty set Rs and delete the cluster from TR If not, delete T in TR directly. I ; Step S325: Determine whether the vehicle trajectory data set TR is an empty set. If not, select a new vehicle trajectory from TR as T. I , return to step S323; if yes, proceed to step S326; Step S326: Determine whether there is only one cluster or an empty set in Rs. If not, take Rs as the new Re and take Rs as the new vehicle trajectory data set TR, and return to step S323; if so, output the current Re as the final clustering result, that is, the current Re is the representative vehicle trajectory set in the driving direction.

7. The method according to claim 3, characterized in that In step S33, determining the duration of the red light in each driving direction includes: Step S331: Input the parking point information set K={K1, K2…K i ...K8}, K i Represents the parking point information at the stop line in the i-th driving direction, and at the same time inputs an empty set R for storing the results; Step S332: Input the parking point information K at a stop line in a driving direction in K i =(x i ,y i ,id i ), (x i ,y i ) indicates the parking location information, id i The vehicle number indicating the i-th driving direction; Step S333: Find the ID i The corresponding vehicle trajectory T, and calculate the vehicle's speed v; Step S334: Determine whether the parking point corresponding to v=0 is unique. If so, it means that there is no waiting area in the driving direction. Calculate the corresponding red light duration R according to the parking time of the vehicle at the stop line. i If not, it means that there is a waiting area in the driving direction, and the sum of the two stop times needs to be calculated to obtain the corresponding red light duration R i ; Step S335: Set the red light duration R i Add to R, determine K i Have you traversed all K? If not, delete K from K. i , and select a new K from K i As input, return to step S332, and if so, output the current R, that is, the set of red light durations in each driving direction.

8. The method according to claim 3, characterized in that In step S33, determining the position information of the waiting area in each driving direction includes: Step a: Input the parking point information set K = {K1, K2…K i ...K8}, K i Represents the parking point information at the stop line in the i-th driving direction, and inputs an empty set W for storing the results; Step b: Input the parking point information K at the parking line in one driving direction in K i (x i ,y i ,id i ), (x i ,y i ) indicates the parking location information, id i The vehicle number indicating the i-th driving direction; Step c: Find the id i The corresponding vehicle trajectory T, and calculate the vehicle's travel speed v; Step d: Determine whether the parking point position corresponding to v=0 is unique. If so, it means that there is no waiting area in this direction. If not, it means that there is a waiting area in the driving direction. Then add the two parking point position information (x1, y1) and (x2, y2) to the set W of the waiting area position information in the i-th driving direction. i , and W i Add to W; Step e: Determine K i Have you traversed all K? If not, delete K from K. i , and select a new K from K i As input, return to step b, if yes, then output the current W, that is, the set of waiting area position information in each driving direction.

9. The method according to claim 3, characterized in that In step S34, determining the traffic conflict point information in each driving direction according to the representative vehicle trajectory set includes: Step S341: Input representative vehicle trajectory set RE = {T′1, T′2…T′ i …T′ g }, T′ i is the i-th representative vehicle trajectory, 1≤i≤g, g is the number of representative vehicle trajectories, and the input is the empty set I used to store the output results; Step S342: Input representative vehicle trajectory T i ′ , calculate and obtain T i ′ Traffic conflict point information set I i ; Step S343: Determine T i ′ Is the driving direction right? If so, directly turn I i Add to I, if not, first extract the trajectory T i ′ The representative vehicle trajectories that cannot intersect in time are recorded as time-independent representative vehicle trajectories, and I is deleted. i The traffic conflict point information related to the representative vehicle trajectory that does not intersect with the time is obtained, and then the updated I i Add to I; Step S344: Determine whether all representative vehicle trajectories in RE have been traversed. If not, select the untraversed representative vehicle trajectory as the new T i ′ Input, return to step S342, if yes, output the current I, that is, the set of traffic conflict point information including each driving direction.

10. The method according to claim 1, characterized in that In the step S4, a simulation road network model of an intersection is built according to the vehicle trajectory information and the road structure information, and a traffic survey of the plane intersection is carried out based on the simulation road network model of the intersection, including: Step S41, using the road structure information to customize the road network in the plane intersection simulation model; Step S42, using the vehicle trajectory information to customize the vehicle flow in the plane intersection simulation model; Step S43, building a plane intersection simulation model, restoring the actual road scene, realizing the traffic survey of the plane intersection, and obtaining the traffic survey results of the plane intersection.

11. A plane intersection traffic survey system based on drone video, characterized in that: include: A video sequence acquisition module is used to acquire the plane intersection video sequence taken by the drone; A vehicle trajectory information extraction module is used to process the plane intersection video sequence using computer vision technology to extract vehicle trajectory information; A road structure information extraction module is used to extract the road structure information of the plane intersection by using the spatial features and attribute features contained in the vehicle trajectory information; the road structure information includes the road network structure at the lane level, the duration of the red light, the location information of the waiting area, and the traffic conflict point information; The traffic survey module is used to build an intersection simulation road network model according to the vehicle trajectory information and the road structure information, and to implement a traffic survey of the plane intersection based on the intersection simulation road network model.

Images