Crossing safety state perception and diagnosis and treatment system and method based on unmanned aerial vehicle video

Abstract

The application discloses a kind of intersection safety state perception and diagnosis management system and method based on unmanned aerial vehicle video, system includes video input module, index extraction module, safety evaluation module, conflict discrimination module, hidden danger diagnosis module, measures management module and result output module.The application proposes an improved deceleration DRAC calculation method for avoiding conflict, which is suitable for various types of traffic conflict events in the database;The system can automatically detect the intersection safety level, and automatically generate management measures for problem intersections, realize the precise strategy and fine management of intersection, effectively reduce the intersection accident, and improve the intersection safety level.

Description

Technical Field

[0001] This invention relates to the field of intersection safety status perception and diagnosis management, and in particular to an intersection safety status perception and diagnosis management system and method based on UAV video. Background Technology

[0002] Intersections are the meeting points of various traffic modes, and also the points of conflict in traffic behavior and frequent traffic accidents. According to global statistics, intersection accidents account for 30%–80% of traffic accidents on urban roads. In recent years, intersection accidents in Chinese urban roads have shown a gradual upward trend. Therefore, the safety and order of traffic at intersections play a crucial role in the smooth operation of the entire road network. A reasonable assessment of the safety level of intersections is fundamental to reducing intersection accidents; therefore, perceiving the safety status of intersections and diagnosing potential hazards at intersections is of great practical significance.

[0003] Most existing research focuses on intersection safety assessments. However, after diagnosing a low level of intersection safety, improvement measures are often provided manually, which is subjective and inefficient. There is no discussion on how to intelligently improve and manage intersection safety. Summary of the Invention

[0004] The purpose of this invention is to provide a system and method for perceiving, diagnosing and managing the safety status of intersections based on drone video. This system can autonomously detect the safety level of intersections and automatically generate management measures for problematic intersections, thereby achieving precise policy implementation and refined management of intersections, effectively reducing the occurrence of intersection accidents and improving the safety level of intersections.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0006] In a first aspect, the present invention provides a system for perceiving, diagnosing, and managing the safety status of intersections based on drone video, comprising:

[0007] The video input module is used to input the collected video of the vehicle operating status at the intersection to be diagnosed into the system;

[0008] The indicator extraction module is used to perform target detection and trajectory tracking of vehicles in the video, and to extract vehicle speed and safety evaluation indicators.

[0009] The safety assessment module is used to evaluate the safety level of intersections and determine whether the hazard diagnosis needs to be carried out.

[0010] The conflict identification module is used to extract the driving trajectory of vehicles in the intersection to be diagnosed, compare it with the traffic conflict event type database, and obtain all conflict event types existing in the intersection to be diagnosed.

[0011] The hazard diagnosis module matches potential hazards at intersections with diagnostic capabilities based on the type of traffic conflict incident.

[0012] The measures and management module extracts management measures for intersections with diagnostic features based on potential hazards.

[0013] The results output module is used to output the intersection safety evaluation level, the types of traffic conflict events, potential hazards, and mitigation measures.

[0014] Secondly, the present invention provides a method for intersection safety status perception, diagnosis and management based on UAV video. This method is implemented based on the system described in the first aspect and specifically includes the following steps:

[0015] S1. Input the video of vehicle movement at the intersection to be inspected, collected by the drone, into the system;

[0016] S2. Use the YOLOv5x target detection algorithm and the JDE multi-target tracking algorithm to extract the vehicle's motion parameters and driving trajectory;

[0017] S3. Based on step S2, extract safety evaluation indicators and use the grey clustering evaluation method to classify the intersection into very safe, safe, critically safe and unsafe.

[0018] S4. Perform hazard diagnosis on intersections with an evaluation level of critical safety or unsafe; otherwise, directly output the safety evaluation level of the intersection.

[0019] S5. Extract the vehicle trajectory of the intersection to be diagnosed and compare it with the traffic conflict event type database to obtain all conflict event types existing at the intersection to be diagnosed.

[0020] S6. Based on the type of intersection conflict, perform intelligent matching with the database of potential hazards;

[0021] S7. Based on the matched intersection hazard factors, adaptively extract intersection mitigation measures from the safety mitigation measures library;

[0022] S8. Based on steps S3, S6, and S7, output the intersection safety evaluation level, the types of existing traffic conflict events, potential hazards, and mitigation measures.

[0023] Compared with the prior art, the significant advantages of the present invention are:

[0024] (1) This invention proposes an improved method for calculating the deceleration rate (DRAC) to avoid conflict, which is applicable to various traffic conflict event types in the database.

[0025] (2) This invention proposes a database of conflict event types, a database of traffic hazard factors, and a database of safety management measures for intersections;

[0026] (3) This invention proposes a new isomer decomposition-multi-feature matching (ID-MFM) method, which realizes multi-dimensional association between intersection traffic conflict event types and hidden danger factor database and governance measure database;

[0027] (4) The present invention can realize intelligent detection of the safety level of intersections and can adaptively generate potential hazards and safety management measures for intersections. Attached Figure Description

[0028] Figure 1 This is a flowchart of the process of the present invention.

[0029] Figure 2 Build a flowchart for the conflict type library.

[0030] Figure 3 This is a flowchart of the isomer decomposition-multi-feature matching algorithm.

[0031] Figure 4 This is a schematic diagram of an intersection in the embodiment. Detailed Implementation

[0032] Combined with appendix Figure 1 This invention provides a system for perceiving, diagnosing, and managing the safety status of intersections based on drone video, comprising:

[0033] Video input module: Inputs the collected video of the vehicle operating status at the intersection to be diagnosed into the system.

[0034] Indicator extraction module: Performs target detection and trajectory tracking on vehicles (including motor vehicles and non-motor vehicles) in the video, and extracts vehicle speed and safety evaluation indicators.

[0035] Safety assessment module: Evaluates the safety level of the intersection and determines whether the hazard to be diagnosed needs to be diagnosed.

[0036] Conflict identification module: Extracts the driving trajectory of vehicles in the intersection to be diagnosed, compares it with the traffic conflict event type database, and obtains all conflict event types existing in the intersection to be diagnosed.

[0037] Hazard Diagnosis Module: Matches potential hazards at intersections with diagnostic capabilities based on the type of traffic conflict incident.

[0038] Management Measures Module: Extract management measures for intersections with diagnostic features based on potential hazards.

[0039] Results output module: Outputs the intersection safety evaluation level, the types of traffic conflict events, potential hazards, and mitigation measures.

[0040] Furthermore, the present invention also provides a method for intersection safety status perception, diagnosis and management based on UAV video. This method is implemented based on the above system and specifically includes the following steps:

[0041] S1. Input the video of the vehicles running at the intersection to be inspected, collected by the drone, into the system.

[0042] S2. Use the YOLOv5x target detection algorithm and JDE multi-target tracking algorithm to extract the motion parameters and driving trajectory of vehicles (including motor vehicles and non-motor vehicles).

[0043] S3. Based on step S2, extract safety evaluation indicators such as intersection conflict rate (TC / MPCU), severe conflict rate (RSC), and average collision avoidance deceleration (ADRAC). Use the grey clustering evaluation method to classify intersections into very safe, safe, critically safe, and unsafe.

[0044] S4. Perform hazard diagnosis for intersections with an evaluation level of critical safety or unsafe; otherwise, directly output the intersection safety evaluation level.

[0045] S5. Extract the vehicle trajectory of the intersection to be diagnosed and compare it with the traffic conflict event type database to obtain all conflict event types existing at the intersection to be diagnosed.

[0046] S6. Based on the type of intersection conflict, perform intelligent matching with the database of potential hazards.

[0047] S7. Based on the matched intersection hazard factors, adaptively extract intersection management measures from the safety management measures library.

[0048] S8. Based on steps S3, S6, and S7, output the intersection safety evaluation level, the types of existing traffic conflict events, potential hazards, and mitigation measures.

[0049] Step S1 above includes:

[0050] S101, The drone has a frame rate of 30 frames / s and an image size of 4K.

[0051] Step S2 above includes:

[0052] S201. The YOLOv5x object detection algorithm is used to detect motor vehicles and non-motor vehicles, and the JDE multi-object tracking algorithm is used to track motor vehicles and non-motor vehicles and extract relevant motion parameters, as detailed below:

[0053] (1) The video of the intersection collected on the ground by drone was split into images frame by frame and saved in a folder to obtain several images at different times as a training set for target detection;

[0054] (2) Use LabelImg to draw frames around the vehicles and road markings in the captured images, and label them with their vehicle types;

[0055] (3) Convert the XML format file generated by LabelImg annotation into a TXT format file used by YOLOv5x;

[0056] (4) Modify the parameters of the YOLOv5x object detection algorithm, run the algorithm training part to train the dataset, and obtain the motor vehicle and non-motor vehicle detection algorithm;

[0057] (5) Run the detection algorithm in (4) to obtain the detection results, and use the JDE multi-target tracking algorithm for retraining to extract the trajectory;

[0058] (6) Use the trained tracking model to track motor vehicles and non-motor vehicles in the video of the UAV at the intersection and extract relevant motion parameters.

[0059] Step S3 above includes:

[0060] S301. Extract multiple safety evaluation indicators for intersections, as detailed below:

[0061] (1) Conflict rate (TC / MPCU) is the ratio of the number of traffic conflicts to the mixed equivalent traffic volume per hour.

[0062] (2) Severe Conflict Rate (RSC) is the ratio of the number of severe traffic conflicts (SC) at an intersection to the total number of conflicts (TC) at the intersection. A deceleration of 3.35 m / s² is selected to avoid conflicts. 2 This serves as a threshold for determining serious traffic conflicts.

[0063] (3) The average deceleration for avoiding collisions (ADRAC) is the ratio of the sum of all collision avoidance decelerations within the intersection to the number of collisions. The calculation formula is as follows:

[0064]

[0065] Where TC represents the number of traffic conflicts, and DRAC represents the number of traffic incidents. i Let be the DRAC value of the i-th traffic conflict.

[0066] S302. Using the grey clustering evaluation method, and combining the above three evaluation indicators, evaluate the safety level of the intersection.

[0067] Step S4 above includes:

[0068] S401. Based on the intersection safety evaluation level obtained in S3, if the safety level of the intersection to be inspected is very safe or safe, output the safety evaluation level; if the safety level of the intersection to be inspected is critically safe or unsafe, perform intersection hazard diagnosis.

[0069] Step S5 above includes:

[0070] S501. Construct a database of traffic conflict events, such as Figure 2 As shown, the steps are as follows:

[0071] (1) Use drones to collect video of vehicles running at intersections, and use YOLOv5x and JDE algorithms to extract the target's driving trajectory and motion parameters;

[0072] (2) Calculate the Deceleration Rate to Avoid a Crash (DRAC) for collisions between motor vehicles and between motor vehicles and non-motor vehicles at distances less than 10m. DRAC is defined as the relative deceleration of the two parties in a traffic conflict in order to avoid a collision. An improved DRAC calculation method is proposed for various types of traffic conflict events. The calculation formula is as follows:

[0073]

[0074] Among them, v A Let L be the velocity of vehicle A at time t, α be the angle between the direction of motion of vehicle A at time t and the vertical direction, and L be the velocity of vehicle A at time t. A v is the perpendicular distance between vehicle A and the point of collision (if no action is taken and a collision occurs) at time t. B Let L be the velocity of vehicle B at time t, β be the angle between the direction of motion of vehicle B at time t and the vertical direction, and L be the velocity of vehicle B at time t. B It is the vertical distance between vehicle B and the point of collision (if no action is taken and a collision occurs) at time t;

[0075] (3) Select a threshold k = 2.45 and filter traffic conflict events for the DRAC value calculated in step (2). If it is greater than the threshold k, it is determined that a traffic conflict has occurred; otherwise, it is determined that no conflict has occurred.

[0076] (4) Store the traffic conflict events selected in step (3) as conflict samples and extract the movement trajectories of the conflicting parties;

[0077] (5) Based on the motion trajectory extracted in step (4), the conflict events are classified into 15 types of traffic conflict events according to the conflicting parties, conflict type, and turning type. A traffic conflict event type database is constructed, and each type of traffic conflict event is denoted as TC. i(i = 1, 2, ..., 15). The database classification of traffic conflict event types is as follows:

[0078] Table 1. Database Classification of Traffic Conflict Incident Types

[0079]

[0080] Note: V represents motor vehicles, N represents non-motor vehicles, D represents diversion conflict, C represents merging conflict, O represents intersection conflict, S represents straight ahead, L represents left turn, R represents right turn, C′ represents crossing the street, and TC1 = V - VDS - L represents diversion conflict between straight-going motor vehicles and left-turning motor vehicles.

[0081] Step S6 above includes:

[0082] S601. Based on the conflict event type library in S5, construct a database of potential hazards corresponding to each conflict event type from four aspects: people, vehicles, roads, and the environment. The hazard sets are represented by 2, 3, and 5 correspondences. Each conflict event type corresponds to more than 10 potential hazards, and the database of potential hazards can be continuously enriched in the future.

[0083] S602. A novel feature matching method—Isomers Decompose-Multiple Feature Matching (ID-MFM)—is proposed to match traffic conflict event types with potential hazards. The input conflict event type TC is denoted as... i Let T be the number of features, and T′ be the number of features in the factor set N in the factor library. Different dimensions (T≠T′) are heterogeneous. For example Figure 3 As shown, the specific steps for matching are as follows:

[0084] (1) Input the type of traffic conflict event at the intersection (TC) i r = 1, and the number of N is x;

[0085] (2) For TC i Matching individual indicators P in the factor library yields indicator P′ with the smallest absolute error to P.

[0086] (3) Compare P with P′ and calculate the matching accuracy q of P′. The calculation formula is as follows:

[0087] m=n TP (n TP +n FP )

[0088] r = n TP (n TP +n FN )

[0089] q = m × r2

[0090] In the formula, m is the precision value, r is the recall value, and n TP n represents the number of true positives. FP n represents the number of false positives. FN The number of false negatives; positive indicates optimization, negative indicates no optimization.

[0091] (4) If q≠100%, iterate and rematch; otherwise, save the matched factor N to the factor pool, r=r+1;

[0092] (5) Based on actual needs, set the termination condition as r∈[2x, 5x];

[0093] (6) Decompose the evaluation set or factor set belonging to the isomer into homogeneity, and match the optimal solution factor N′ with the minimum mean square error in the factor pool. Denote its eigenvalue as T′. correct ;

[0094] (7) Calculate the matching reliability of factor N. If T′ correct ≠T,

[0095]

[0096] In the formula: q represents the matching precision, 0 < q < 1, and the larger the value, the higher the matching precision; P i，max P is the maximum value of the preset feature i; i，min The value of feature i for the current conflict event type is used as the initial iteration value; C i To ensure reliability.

[0097] (8) If the number of N′, Count(N′) < x, save N′ and its matching reliability;

[0098] (9) Otherwise, output all N′.

[0099] Step S7 above includes:

[0100] S701. Based on the hidden danger factor database in S6, a database of over a hundred intersection safety management measures is constructed, which is divided into two categories: short-term improvement measures and medium- and long-term improvement measures. Similarly, the database of management measures can be continuously enriched in the future.

[0101] S702. Set minimum support and minimum confidence. Connect a factor X in the hazard factor library with a governance measure Y in the governance measure library by using multidimensional Apriori association rules and fast pattern matching algorithm (KMP algorithm) to establish strong rules. Repeat this process to achieve an orderly connection between the intersection hazard factor library and the governance measure library.

[0102] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings and specific examples.

[0103] Example

[0104] Appendix Figure 4 In the scenario where this embodiment is applied, the intersection is a T-type signalized intersection. The following description of the implementation of the present invention will be based on this intersection as an example.

[0105] like Figure 1 As shown, the specific workflow is as follows:

[0106] S1. The drone collects video footage of vehicles operating at the intersection during the morning rush hour and inputs the collected video into the system.

[0107] S2. Use the YOLOv5x object detection algorithm and the JDE multi-object tracking algorithm to extract the driving trajectories and speeds of motor vehicles and non-motor vehicles in the video.

[0108] S3. Extract the safety evaluation indicators of the intersection and obtain TC / MPCU=0.0329, RSC=0.09, ADRAC=2.612. Using the grey clustering evaluation method, the safety level of the intersection is determined to be critically safe.

[0109] S4. The safety level of this intersection is critically safe, therefore, a hazard diagnosis of the intersection is required.

[0110] S5. Extract the driving trajectories of all vehicles at the intersection and compare them with the traffic conflict event type database to find that the conflict types at the intersection are V-VCS-R, V-VCR-L, V-VCS-L and V-VOS-L.

[0111] S6. Based on the type of intersection conflict, intelligently match the potential hazards of the intersection in the hazard factor database. The potential hazards of the intersection include irregular intersection shape and lack of separation measures between motor vehicles and non-motor vehicles.

[0112] S7. Based on the potential hazards at the intersection, the corresponding intersection mitigation measures are adaptively extracted from the safety mitigation measures library. The mitigation measures matched for this intersection include optimizing the intersection's horizontal alignment and setting up separation facilities for motor vehicles and non-motor vehicles.

[0113] S8. The system outputs the safety evaluation level of the intersection, the types of traffic conflict events, potential hazards, and mitigation measures.

[0114] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them; although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications can still be made to the specific implementation of the present invention or equivalent substitutions can be made to some technical features, such as the establishment of a database and multi-feature matching algorithms; the above should not depart from the spirit of the technical solutions of the present invention, and all of them should be covered within the scope of the technical solutions claimed in the present invention.

Claims

1. A method for intersection safety status perception, diagnosis, and management based on UAV video, characterized in that, Includes the following steps: S1. Input the video of the vehicle operation status at the intersection to be diagnosed, collected by the drone, into the system; S2. Use the YOLOv5x target detection algorithm and the JDE multi-target tracking algorithm to extract the vehicle's motion parameters and driving trajectory; S3. Based on step S2, extract safety evaluation indicators and use the grey clustering evaluation method to classify the intersection into very safe, safe, critically safe and unsafe. S4. Perform hazard diagnosis on intersections with a safety evaluation level of critical safety or unsafe; otherwise, directly output the safety evaluation level of the intersection. S5. Extract the vehicle driving trajectory of the intersection to be diagnosed and compare it with the traffic conflict event type database to obtain all traffic conflict event types existing at the intersection to be diagnosed. The steps to build a database of traffic conflict event types are as follows: (1) Use drones to collect video of vehicle operation status at intersections, and use YOLOv5x and JDE algorithms to extract the target's driving trajectory and motion parameters; (2) Calculate the DRAC value between motor vehicles and between motor vehicles and non-motor vehicles when the distance is less than 10m. DRAC is defined as the relative deceleration of the two parties in order to avoid a collision when a traffic conflict occurs. The DRAC calculation formula is as follows: ； in, yes The speed of vehicle A at that moment, yes The angle between the direction of motion of vehicle A and the vertical direction at any given moment. yes The perpendicular distance between vehicle A and the point of collision at any given time. yes The speed of vehicle B at that moment, yes The angle between the direction of motion of vehicle B and the vertical direction at any given moment. yes The perpendicular distance between vehicle B and the point of collision at any given moment; (3) Selecting a threshold The traffic conflict events are screened based on the DRAC values ​​calculated in step (2). If the value is greater than the threshold... If the traffic conflict occurs, it is determined that a traffic conflict has occurred; otherwise, it is determined that no traffic conflict has occurred. (4) Store the traffic conflict events selected in step (3) as conflict samples and extract the driving trajectories of the conflicting parties; (5) Based on the driving trajectory extracted in step (4), traffic conflict events are classified into 15 categories according to the conflicting parties, conflict type, and turning type. A traffic conflict event type database is constructed, and each type of traffic conflict event is denoted as [missing information]. The database categorizes traffic conflict incident types as follows: Table 1. Database Classification of Traffic Conflict Incident Types ； Wherein, V represents motor vehicles, N represents non-motor vehicles, D represents divergence conflict, C represents merging conflict, O represents intersection conflict, S represents straight ahead, L represents left turn, and R represents right turn. Indicates crossing the street. =V-VDS-L indicates a traffic flow conflict between vehicles going straight and vehicles turning left; S6. Based on the type of intersection conflict, perform intelligent matching with the database of potential hazards; S7. Based on the matched intersection hazard factors, adaptively extract intersection mitigation measures from the safety mitigation measures library; S8. Based on steps S3, S6, and S7, output the intersection safety evaluation level, the types of existing traffic conflict events, potential hazards, and mitigation measures.

2. The method for intersection safety status perception, diagnosis, and management based on UAV video according to claim 1, characterized in that, In step S1, the drone's frame rate is 30 frames / s, and the image size is 4K.

3. The method for intersection safety status perception, diagnosis, and management based on UAV video according to claim 1, characterized in that, Step S2 includes: S201. The YOLOv5x object detection algorithm is used to detect motor vehicles and non-motor vehicles, and the JDE multi-object tracking algorithm is used to track motor vehicles and non-motor vehicles and extract relevant motion parameters, as detailed below: (1) The video of the intersection collected on the ground was split into images frame by frame and saved in a folder to obtain several images at different times as a training set for target detection; (2) Use LabelImg to draw frames around the vehicles and road markings in the captured images, and label them with their vehicle models; (3) Convert the XML format file generated by LabelImg annotation into a TXT format file used by YOLOv5x; (4) Modify the parameters of the YOLOv5x object detection algorithm, run the algorithm training part to train the dataset, and obtain the motor vehicle and non-motor vehicle detection algorithm; (5) Run the motor vehicle and non-motor vehicle detection algorithm to obtain the detection results, and use the JDE multi-object tracking algorithm for retraining to extract the driving trajectory; (6) Use the trained tracking model to track motor vehicles and non-motor vehicles in the UAV video at the intersection and extract relevant motion parameters.

4. The method for intersection safety status perception, diagnosis, and management based on UAV video according to claim 3, characterized in that, Step S3 includes: S301. Extract multiple safety evaluation indicators for intersections, as detailed below: Conflict rate: The ratio of the number of traffic conflicts occurring within one hour to the mixed equivalent traffic volume; Severe Conflict Rate: The ratio of the number of severe traffic conflicts within an intersection to the total number of conflicts within the intersection; The average deceleration for avoiding collisions: the ratio of the sum of all collision-avoidance decelerations within the intersection to the number of collisions, calculated using the following formula: ； TC represents the number of traffic conflicts. For the first DRAC value of the secondary traffic conflict; S302. Using the grey clustering evaluation method, and combining the above three evaluation indicators, evaluate the safety level of the intersection.

5. The method for intersection safety status perception, diagnosis, and management based on UAV video according to claim 4, characterized in that, Step S4 includes: Based on the intersection safety evaluation level obtained in S3, if the safety evaluation level of the intersection to be diagnosed is very safe or safe, then the safety evaluation level is output; if the safety evaluation level of the intersection to be diagnosed is critically safe or unsafe, then the intersection hazard diagnosis is performed.

6. The method for intersection safety status perception, diagnosis, and management based on UAV video according to claim 1, characterized in that, Step S6 includes: S601. Based on the traffic conflict event type database in S5, construct a database of potential hazards corresponding to the traffic conflict event types. S602. Using the ID-MFM algorithm, the traffic conflict event types are matched with potential hazard factors. The input traffic conflict event type is recorded. The characteristic number is Factor set in the hidden danger factor database The characteristic number is Since different dimensions are heterogeneous, the specific steps for matching are as follows: (1) Types of traffic conflict events existing at the input intersection , , The number of ; (2) To single indicator Matching was performed to obtain data from the hazard factor database. The index with the smallest absolute error ; (3) and Compare, calculate Matching accuracy The calculation formula is as follows: ； ； ； In the formula, For precision values, For recall value, The number of true positives. The number of false positives, The number of false negatives; positive indicates optimization, negative indicates no optimization. (4) When If necessary, iterate and rematch; otherwise, discard the matched factors. Save to the factor pool. ; (5) Based on actual needs, set the termination condition as follows: ; (6) Decompose the evaluation set or factor set belonging to the heterogeneity, transform it into homogeneity, and match the optimal solution factor with the minimum mean square error in the factor pool. Let its characteristic number be . ; (7) Calculation factors The matching reliability, if, , ； In the formula: For matching accuracy, ; Preset features The maximum value; Preset features The minimum value; Characteristics of current traffic conflict incidents The value of is used as the initial iteration value; To ensure reliability; (8) If Number of ,save and its matching reliability; (9) Otherwise, output all .

7. The method for intersection safety status perception, diagnosis, and management based on UAV video according to claim 1, characterized in that, Step S7 above includes: S701. Based on the hidden danger factor database in S6, construct a database of safety management measures for intersections; S702. Set minimum support and minimum confidence, and establish strong rules between factor X in the hazard factor database and management measure Y in the safety management measure database through multidimensional Apriori association rules and fast pattern matching algorithm to achieve an orderly connection between the intersection hazard factor database and the safety management measure database.

8. A system for perceiving, diagnosing, and managing the safety status of intersections based on UAV video, used to implement the method described in claim 1, characterized in that the system... include: The video input module is used to input the collected video of the vehicle operating status at the intersection to be diagnosed into the system; The indicator extraction module is used to perform target detection and trajectory tracking of vehicles in the video, and to extract vehicle speed and safety evaluation indicators. The safety assessment module is used to evaluate the safety assessment level of the intersection and determine whether the hazard diagnosis needs to be carried out. The conflict identification module is used to extract the vehicle driving trajectory of the intersection to be diagnosed, compare it with the traffic conflict event type database, and obtain all traffic conflict event types existing at the intersection to be diagnosed. The hazard diagnosis module matches potential hazards at the intersection to be diagnosed based on the type of traffic conflict event. The measures and management module extracts management measures for the intersections to be diagnosed based on the potential hazards. The results output module is used to output the intersection safety evaluation level, the types of traffic conflict events, potential hazards, and mitigation measures.

Images