Multi-dimensional dynamic control method of traffic incidents based on vehicle flow monitoring

Abstract

The present invention relates to the technical field of traffic data processing, and specifically to a method for continuously monitoring traffic data detected by artificial intelligence and judging corresponding traffic events based on real-time data, thereby realizing automatic and efficient judgment of various traffic events including accidents and congestion in a designated road section, and can effectively save manpower and improve event processing efficiency. The multi-dimensional dynamic control method of traffic events based on vehicle flow monitoring relies on artificial intelligence algorithms to obtain various characteristics of vehicles and personnel in the detection section, and then realizes efficient, rapid and automatic judgment of traffic events based on the characteristics, thereby achieving the purpose of improving detection efficiency and reducing manpower consumption.

Description

Technical field:

[0001] The present invention relates to the technical field of traffic data processing, and specifically to a method for continuously monitoring traffic data detected by artificial intelligence and judging corresponding traffic events based on real-time data, thereby realizing automatic and efficient judgment of various traffic events including accidents and congestion in a designated road section, and realizing a multi-dimensional dynamic control method for traffic events based on vehicle flow monitoring, which can effectively save manpower and improve event processing efficiency. Background technology:

[0002] Road transportation is currently widely used in my country. Due to its flexible and convenient characteristics, it plays an important role in freight and passenger transportation. With the improvement of people's living standards and the enrichment of their lifestyles, the total number of domestic vehicles and road travel volume have continued to rise, which has also led to an increasing frequency of traffic congestion and vehicle accidents. How to more quickly identify various abnormal traffic incidents, handle them in a timely manner, and ensure personnel safety and traffic efficiency has become an urgent problem to be solved.

[0003] Since there was a gap in this aspect of past technology, the acquisition of traffic incidents mostly relied on the parties to report to the police, and the traffic police relied on video cameras to observe. The disadvantages of this type of method are: 1. Lack of real-time performance. When the traffic police obtain the incident, the corresponding incident has often occurred for a period of time, and the work of dredging and rescue cannot be carried out in time, which is easy to cause secondary injuries or delay treatment. 2. Currently, videos are mostly used as post-event evidence, and directly rely on personnel to monitor the video to discover the incident, which consumes a lot of manpower, and is limited by the time and energy of the monitoring personnel, and it is difficult to achieve continuous observation throughout the day. In addition, the different judgment standards of personnel may also affect the effect of monitoring, which is difficult to quantify. Summary of the invention:

[0004] In view of the shortcomings and deficiencies in the prior art, the present invention proposes a method for multi-dimensional dynamic control of traffic events based on vehicle flow monitoring, which relies on artificial intelligence algorithms to obtain various characteristics of vehicles and personnel on the detected road section, and then realizes efficient, rapid and automated judgment of traffic events based on the characteristics, thereby achieving the purpose of improving detection efficiency and reducing manpower consumption.

[0005] The present invention is achieved by the following measures:

[0006] A multi-dimensional dynamic control method for traffic events based on vehicle flow monitoring, characterized by comprising the following contents:

[0007] Step 1: Obtain detection data within the detection area: Continuously obtain and store the speed (including direction), position, and vehicle head direction data of each vehicle within the detection range at the current moment. Thus, the speed sequence V of all vehicles can be obtained. V is composed of multiple single-vehicle speed sequences v, where v = (v1, v2, ..., v n ), where v n is the current speed, v1~v n-1 is the historical speed detected and stored before, and the value of n depends on the number of historical data to be retained. Similarly, the position sequence X of all vehicles and the head direction sequence H of all vehicles can also be obtained;

[0008] In addition, the license plates of all vehicles within the range are identified, the license plate information is obtained, and the correspondence between the speed, position, and vehicle head direction data and the license plate number is established. The data source is based on the image recognition algorithm;

[0009] Step 2: Get the information of each vehicle in the detection area:

[0010] Step 2-1: Calculate the maximum speed change, i.e., the maximum acceleration a, of a specified vehicle in the recent period of time. The formula is: δ is a coefficient that depends on the time interval between consecutive speed detections in step 1;

[0011] Step 2-2: Calculate the current cumulative parking time t for the specified vehicle n

[0012] The formula is: t0 = 0, 1≤i≤n, δ is a coefficient, which depends on the time interval of continuous speed detection in step 1;

[0013] Step 2-3: Determine whether the specified vehicle is waiting in the queue, and calculate its position in the queue of the current road section, which is called the queue sequence number s: specifically:

[0014] Step 2-3-1: Set the queue number s=1 for the specified vehicle and obtain its current vehicle head direction h n ;

[0015] Step 2-3-2: According to the current positions of all vehicles in the position sequence X of all vehicles, check whether there is a vehicle within a certain distance in front of the vehicle. If there is a vehicle in front, increase the queue number s by 1, and check the front direction of the vehicle in front according to the head direction sequence H of all vehicles. According to the current head direction of the vehicle in front and the current positions of all vehicles, continue to determine whether there is a vehicle in a certain range in front of it; loop until it is found that there is no vehicle in front within a certain range, or there is a stop line or image boundary in front. If the loop is terminated because there is no vehicle in front within a certain range, it is considered that the vehicle is not waiting in the queue. Otherwise, if the loop is terminated because there is a stop line or image boundary in front, it is considered that the vehicle is waiting in the queue, and the value of s is the queue number;

[0016] Step 2-4: Calculate the distance dr between the specified vehicle and the roadside at the current moment. The specific method is: according to the current position x of the vehicle n And the road boundary information R, find the minimum distance between the vehicle and all roadsides;

[0017] Step 2-5: Calculate the minimum distance d between a specified vehicle and other vehicles in the recent period of time. The formula is: d i It is the minimum distance between a specified vehicle and all other vehicles at each detection moment in the recent period of time;

[0018] Step 2-6: Detect whether the vehicle in the range has its doors open or has its hazard lights on. The data source is based on the image recognition algorithm.

[0019] Step 3: Obtain data related to each pedestrian in the detection area:

[0020] Step 3-1: Through the camera and its configured position detection and tracking algorithm, the current posture of each person in the detection range is determined, and the current position is continuously detected and stored. The posture of all people in the range and the current position data of all pedestrians are derived from the detection and positioning algorithm based on images and radars; thus, the position sequence Y of all people can be obtained. Y is composed of the position sequence y of multiple single people from history to the current moment, y=(y1,y2,...,y m ), where y m is the current position, y1~y m-1 is the historical location of the previous detection storage; the value of m depends on the number of historical data to be retained;

[0021] Step 3-2: Based on the position sequence y of the specified person, the current stationary time u of the person can be obtained n , the formula is: distance(y j ,y j-1 ) refers to the position y of a person at a certain timej The position y at the previous detection time j-1 The distance, γ is a coefficient, which depends on the time interval of continuous detection of the personnel position;

[0022] Step 3-3: Calculate the minimum distance dx between the specified vehicle and the pedestrian in the recent period of time. The formula is: dx i It is the minimum distance between the specified vehicle and all pedestrians at each detection moment in the recent period of time;

[0023] Step 3-4: Calculate the minimum distance dy between the specified pedestrian and the vehicle in the recent period of time. The formula is: dy j It is the minimum distance between a pedestrian and all vehicles at each detection moment in the recent period of time.

[0024] Step 4: Establish a vehicle-related traffic event judgment process, and perform logical judgment based on the information obtained in step 2. If it is judged as an abnormal event, the output device layout area, i.e. the event occurrence area, the timestamp, i.e. the event occurrence time, the license plate information, and the event type logical judgment result will be output.

[0025] The abnormal events in step 4 of the present invention include speeding, driving in the wrong direction, reversing, pulling over, waiting in line, suspected vehicle accidents, and vehicle accidents. The judgment process in step 3 continues, and each round judges a single vehicle within the range in turn. After completing the judgment of all vehicles within the range, the next round of judgment will be carried out based on the latest detection data.

[0026] In step 4 of the present invention, whether to stop the vehicle, and the speeding, reverse driving, and reversing conditions are determined according to the vehicle speed (including direction) as follows:

[0027] Step 4-1, obtain the vehicle's current speed v according to step 1 n :

[0028] Step 4-1-1: If the vehicle speed direction is the same as the road requirement, and the vehicle speed v n Greater than or equal to threshold N v , output events for the vehicle: speeding, speed value, and area, time, and license plate information;

[0029] Step 4-1-2: If the vehicle speed direction is opposite to the road requirement, then obtain the vehicle head direction h at the current moment according to step 1 n :

[0030] Step 4-1-2-1; if the vehicle head direction is h n If the direction is the same as the road requirement, the following events are output for the vehicle: reversing, as well as area, time, and license plate information;

[0031] Step 4-1-2-2: If the vehicle head direction is h n If the vehicle is not in the same direction as the road requirement, the following events will be output for the vehicle: wrong direction, as well as the area, time, and license plate information.

[0032] Step 4-1-3: If the vehicle speed is equal to 0, execute step 4-2;

[0033] Step 4-2: The vehicle entering this step has stopped and will be based on the vehicle position x in step 1. n , determine whether it is roadside parking, queue parking, and further confirm the situation;

[0034] When entering step 4-2, that is, when there is parking, step 2-4 is triggered to obtain the distance dr between the car and the roadside, and perform the following operations:

[0035] Step 4-2-1: If the distance to the roadside dr is less than the threshold N dr , then trigger step 2-2, obtain the accumulated parking time t of the vehicle n , output event for the vehicle: roadside parking, parking time t n , and area, time, and license plate information;

[0036] Step 4-2-2: If the distance to the roadside dr is not less than the threshold N dr , then step 2-3 is triggered to obtain the queue sequence number s and whether the car is waiting in line:

[0037] Step 4-2-2-1: If the car is waiting in a queue and the queue number s is greater than the threshold N s , then trigger step 2-2, obtain the accumulated parking time t of the vehicle n , then the output event for the vehicle is: waiting in line, parking time t n , and area, time, license plate information,

[0038] Step 4-2-2-2: If the vehicle is not waiting in line, it is considered that the possibility of an accident is high, and the process goes to step 4-3 for further confirmation;

[0039] Step 4-3: The vehicle entering this step is a stationary vehicle and the parking position is abnormal. It will be further confirmed based on the vehicle's historical status and the current scene to determine whether it is a vehicle accident. When entering step 4-3, step 2-1 is triggered to obtain the maximum speed change (maximum acceleration) a of the vehicle in the recent period; step 2-2 is triggered to obtain the accumulated parking time t of the vehicle. n; Trigger step 2-5: obtain the minimum distance d between it and other vehicles in the recent period; trigger step 3-3, obtain the minimum distance dx between it and pedestrians in the recent period; trigger opening step 2-6, obtain whether there is double flashing and opening the door;

[0040] Parking time alarm threshold N t There is a preset value that will adjust the parking time alarm threshold N according to the following information t :When the maximum speed change (maximum acceleration) a is greater than or equal to the threshold N a When the parking time threshold N t When the vehicle's minimum distance d from other vehicles and the minimum distance dx from pedestrians in the recent period is less than the threshold N d When the parking time threshold N t When the area has double flashing or door opening, the parking time threshold N t To make reductions;

[0041] Compare the accumulated parking time t of the vehicle n and the adjusted (reduced) parking time threshold N t , if t n Greater than or equal to N t , then the output event is: vehicle accident, parking time t n and area, time, license plate information; if t n Less than N t , then the output event is: suspected accident, parking time t n And area, time, license plate information.

[0042] The parking time threshold N in step 4 of the present invention t The determination is the result of comprehensive consideration of multiple factors. Even if there is no drastic speed change, too small distance between vehicles, double flash door opening and other behaviors, if a vehicle is stationary for an accumulated time t n Exceeding the threshold N t , the accident event will also be output. However, if the vehicle queries the historical movement abnormality (dramatic speed change, too small distance between vehicles) or triggers the image judgment to find a specific behavior (double flash, open door), it can be clarified based on this, which is manifested by reducing the threshold N t The value of is reflected in faster (even immediate) output of accident events.

[0043] The present invention also includes step 5, establishing a personnel-related traffic event judgment process, performing logical judgment in a certain order based on the information obtained in step 3, and if it is judged to be a personnel accident, the device layout area (event occurrence area), timestamp (event occurrence time), and event judgment result (personnel accident or suspected personnel accident) will be output. The judgment process continues, and after each round of judgment of all pedestrians within the range is completed, the next round of judgment will be performed based on the latest detection data.

[0044] Step 5 of the present invention specifically comprises the following steps:

[0045] Step 5-1: Obtain the posture of people in the area and the duration of their basic stillness at the current moment according to the image algorithm. If the posture of a person does not meet the requirements (such as lying flat), further analyze the person. If the posture of the person meets the requirements, it is considered normal.

[0046] Step 5-2: The person's current posture is abnormal, and step 3-4 is triggered to obtain the minimum distance dy between the person and the vehicle in the recent period of time. If the minimum distance dy between the person and the vehicle in the recent period of time is less than N q , then the output event is: personnel accident, and step 3-2 is triggered to obtain the static duration u n , accident area (equipment installation area), timestamp, and vehicles that are too close to it, license plate number and other information; if the minimum distance to the vehicle in the recent period of time dy ≥ N q , it is considered that the person has not been too close to the car for some time, and further judgment is made in step 5-3;

[0047] Step 5-3: Trigger step 3-2 and obtain the duration u of the person being basically still at this time n , with the threshold N u Compare and judge whether it is in a static state for a long time. If u n ≥N u , it is considered to be a personnel accident, and the output event is: personnel accident, the duration of basic stillness u n And area, timestamp, at this time, the person is stationary for too long, even if there is no car passing nearby, an alarm should be output. Otherwise, it is considered to be a suspected personnel accident, and the output event is: suspected personnel accident, the duration of basic inactivity u n The suspected personnel accident can be considered as a low-level alarm, that is, it cannot be identified at this time, but as its static time increases, it will be transformed into a personnel accident alarm.

[0048] The present invention also includes step 6, video data storage. When it is determined that there is an abnormality in a vehicle or a person, the characteristics of the abnormal period and the video image are persistently stored according to the timestamp of the determined data. The specific storage period, its start and end time can be extended before and after the abnormal period according to the setting. The image can be used as evidence and can also be used for manual review.

[0049] The present invention relies on artificial intelligence algorithms to obtain various features of vehicles and personnel on the detected road section, and then realizes efficient, rapid and automated judgment of traffic events based on the features, thereby achieving the purpose of improving detection efficiency and reducing manpower consumption. Description of the drawings:

[0050] Attached Figure 1 It is a flow chart of the present invention.

[0051] Attached Figure 2 It is a flow chart of vehicle abnormal event judgment in the present invention.

[0052] Attached Figure 3 It is a flow chart of personnel abnormal event judgment in the present invention.

[0053] Attached Figure 4 This is a schematic diagram of the regional situation at 11:00:10 seconds in Example 1 of the present invention.

[0054] Attached Figure 5 This is a schematic diagram of the regional situation at 11:00:11 seconds in Example 1 of the present invention.

[0055] Attached Figure 6 This is a schematic diagram of the regional situation at 11:00:12 seconds in Example 1 of the present invention.

[0056] Attached Figure 7 This is a schematic diagram of the regional situation at 11:00:10 seconds in Example 2 of the present invention.

[0057] Attached Figure 8 This is a schematic diagram of the regional situation at 11:00:11 seconds in Example 2 of the present invention.

[0058] Attached Fig. 9 This is a schematic diagram of the regional situation at 11:00:12 seconds in Example 2 of the present invention. Specific implementation method:

[0059] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0060] The present invention proposes a method for event judgment based on traffic data, which relies on artificial intelligence algorithms to obtain various features of vehicles and personnel on the detection section, and then realizes efficient, rapid and automatic judgment of traffic events based on the features, thereby achieving the purpose of improving detection efficiency and reducing manpower consumption. Specifically, it includes the following contents:

[0061] S1: According to the recording range of the dimensionality reduction and reconstruction data or the detection range of the camera, radar and other equipment, the detection area is set within its recording range or detection range;

[0062] S2: Obtain dimensionality reduction and reconstruction data of the detection area, or obtain data through detection equipment and related algorithm models. The required data are as follows, among which the data features of S2.1, 2.7, and 2.12 need to be continuously detected, acquired, and stored, and other S data are acquired by triggering calculations under specific circumstances;

[0063] S2.1: Continuously obtain and store the current speed (including direction), position, and vehicle head direction data of each vehicle within the detection range: The data sources of speed (including direction), position, and vehicle head direction can be the corresponding data in the road condition matrix and vehicle condition matrix after dimensionality reduction and reconstruction; or the data collected by the camera or radar are obtained through the corresponding speed, position detection, and tracking algorithms. For example, the source of speed and direction data can be the direct detection of the relevant radar, or the comparison of the current position and the historical position in two images, or the image-based vehicle crossing line detection algorithm;

[0064] Thus, the speed sequence V of all vehicles can be obtained. V is composed of the speed sequences v of multiple vehicles, v = (v1, v2, ..., v n ), where v n is the current speed, v1~v n-1 is the historical speed detected and stored before, and the value of n depends on the number of historical data to be retained. Similarly, the position sequence X of all vehicles and the head direction sequence H of all vehicles can also be obtained;

[0065] S2.2: For a specified vehicle, find its maximum speed change (maximum acceleration) a in the recent period of time. The formula is: δ is a coefficient, which depends on the time interval between consecutive speed detections in 2.1;

[0066] S2.3: Calculate the current accumulated parking time t for the specified vehicle n

[0067] The formula is: t0=0. 1≤i≤n. Same as S2.2, δ is a coefficient, which depends on the time interval of continuous speed detection in 2.1;

[0068] S2.4: Determine whether the specified vehicle is waiting in a queue and find its position in the queue of the current road section, which is called the queue sequence number s, specifically:

[0069] 1) For a specified vehicle, set its queue number s=1 and obtain its current vehicle head direction h n ;

[0070] 2) Based on the current positions of all vehicles in the position sequence X of all vehicles, check whether there is a vehicle within a certain distance in front of the vehicle;

[0071] 3) If there is a vehicle ahead, increase the queue number s by 1, and check the front direction of the vehicle ahead according to the front direction sequence H of all vehicles;

[0072] 4) Based on the current vehicle head direction and the current positions of all vehicles, continue to determine whether there are any vehicles within a certain range in front of the vehicle;

[0073] 5) The loop continues until it is found that there is no car ahead within a certain range, or there is a stop line or image boundary ahead. If the loop is terminated because there is no car ahead within a certain range, the car is considered not to be waiting in the queue. Otherwise, if the loop is terminated because there is a stop line or image boundary ahead, the car is considered to be waiting in the queue. At this time, the value of s is the queue sequence number, where the stop line position and image boundary information can come from a road condition matrix, an image recognition algorithm, or manual annotation.

[0074] S2.5: Calculate the distance dr between the specified vehicle and the roadside at the current moment by: n And the road boundary information R, find the minimum distance between the vehicle and all roadsides, the vehicle's current position x n The data source of can be the vehicle condition matrix after dimensionality reduction and reconstruction, or the image-based position detection algorithm; the data source of the road boundary information R can be the road condition matrix after dimensionality reduction and reconstruction, or the image-based detection algorithm, or manual annotation;

[0075] S2.6: Find the minimum distance d between a specified vehicle and other vehicles in the recent period of time.

[0076] The formula is: d i It is the minimum distance between a specified vehicle and all other vehicles at each detection moment in the recent period of time;

[0077] S2.7: Through the camera and its configured position detection and tracking algorithm, the current posture of each person within the detection range is determined, and their current position is continuously detected and stored. The posture of all people within the range and the current position data of all pedestrians are derived from the detection and positioning algorithm based on images and radars;

[0078] Thus, the position sequence Y of all personnel can be obtained. Y is composed of the position sequences y of multiple individuals from history to the current moment, y = (y1, y2, ..., y m ), where y m is the current position, y1~y m-1is the historical location of the previous detection storage; the value of m depends on the number of historical data to be retained;

[0079] S2.8: Based on the position sequence y of the specified person, the current stationary time u of the person can be obtained n , the formula is: u0=0. 1≤j≤m.

[0080] distance(y j ,y j-1 ) refers to the position y of a person at a certain time j The position y at the previous detection time j-1 The distance, γ is a coefficient, which depends on the time interval of continuous detection of the personnel position;

[0081] S2.9: Calculate the minimum distance dx between a specified vehicle and a pedestrian in the recent period of time.

[0082] The formula is: dx i It is the minimum distance between the specified vehicle and all pedestrians at each detection moment in the recent period of time.

[0083] S2.10: Calculate the minimum distance dy between a specified pedestrian and the vehicle in the recent period of time.

[0084] The formula is: dy j It is the minimum distance between a specified pedestrian and all vehicles at each detection moment in the recent period of time.

[0085] The data source of vehicle position X in S2.9 and 2.10 can be the vehicle condition matrix after dimension reduction and reconstruction, or the detection and positioning algorithm based on image or radar. The data source of personnel position Y can be the detection and positioning algorithm based on image or radar.

[0086] S2.11: Detect whether the vehicle in the range has its doors open or has its hazard lights on. The data source is based on the image recognition algorithm.

[0087] S2.12: Identify the license plates of all vehicles within the range, obtain the license plate information, and establish the correspondence between the speed, position, and vehicle head direction data in 2.1 and the license plate number. The data source is based on the image recognition algorithm.

[0088] S3: Establish a vehicle-related traffic event judgment process and perform logical judgment in a certain order based on the information obtained in S2.

[0089] If it is judged as an abnormal event (speeding, wrong-way driving, reversing, pulling over, waiting in line, suspected vehicle accident, vehicle accident), the device layout area (event occurrence area), timestamp (event occurrence time), license plate information (such as license plate type and license plate number) and event type (logical judgment result) will be output.

[0090] This judgment process continues, and each round judges each vehicle in the range in turn. After the judgment of all vehicles in the range is completed, the next round of judgment will be carried out based on the latest detection data.

[0091] S3.1: Determine whether to stop, speeding, driving in the wrong direction, or reversing based on the vehicle speed (including direction).

[0092] For a single vehicle, the current speed v n :

[0093] 1) If the vehicle speed direction is the same as the road requirement, and the vehicle speed v n Greater than or equal to threshold N v , output events for the vehicle: speeding, speed value, and area, time, and license plate information.

[0094] 2) If the vehicle speed direction is opposite to the road requirement, obtain the vehicle's current head direction h n :

[0095] 2.1) If the vehicle head direction h n If the direction is the same as the road requirement, the following events are output for the vehicle: reversing, as well as area, time, and license plate information;

[0096] 2.2) If the vehicle head direction h n If the vehicle is not traveling in the same direction as required by the road, the following events will be output for the vehicle: wrong direction, as well as area, time, and license plate information.

[0097] 3) If the vehicle speed is equal to 0, proceed to S3.2.

[0098] S3.2: The vehicle entering this S has stopped and will be judged whether it is roadside parking, queue parking, or further confirmation based on the vehicle position.

[0099] When entering S3.2, that is, when the car is parked, S2.5 is triggered to obtain the distance dr between the car and the roadside

[0100] 1) If the distance to the roadside dr is less than the threshold N dr , then trigger S2.3 to obtain the accumulated parking time t of the vehicle n , output event for the vehicle: roadside parking, parking time t n , and area, time, and license plate information.

[0101] 2) If the distance to the roadside dr is not less than the threshold N dr , then S2.4 is triggered to obtain the queue sequence number s and whether the car is waiting in line.

[0102] 2.1) If the car is waiting in a queue and the queue number s is greater than the threshold N s , then trigger S2.3 to obtain the accumulated parking time t of the vehicle n , then the output event for the vehicle is: waiting in line, parking time t n , and area, time, and license plate information.

[0103] 2.2) If the vehicle is not waiting in the queue, it is considered that the possibility of an accident is high and the process goes to S3.3 for further confirmation.

[0104] S3.3: The vehicle entering this S is a stationary vehicle and the parking position is abnormal. It will be further confirmed based on the vehicle's historical status and the current scene to determine whether it is a vehicle accident.

[0105] When entering S3.3, trigger S2.2 to obtain the maximum speed change (maximum acceleration) a of the vehicle in the recent period; trigger S2.3 to obtain the accumulated parking time t of the vehicle n ; Trigger S2.6: obtain the minimum distance d between it and other vehicles in the recent period; trigger S2.9, obtain the minimum distance dx between it and pedestrians in the recent period; trigger S2.11 to obtain whether there is double flash and car door opening behavior.

[0106] Parking time alarm threshold N t There is a preset value that will adjust the parking time alarm threshold N according to the following information t :

[0107] When the maximum speed change (maximum acceleration) a is greater than or equal to the threshold N a When the parking time threshold N t To reduce.

[0108] When the vehicle's minimum distance d from other vehicles and the minimum distance dx from pedestrians in the recent period is less than the threshold N d When the parking time threshold N t To reduce.

[0109] When there are double flashes and door opening in the area, the parking time threshold N t To reduce.

[0110] Compare the accumulated parking time t of the vehicle n and the adjusted (reduced) parking time threshold N t , if t nGreater than or equal to N t , then the output event is: vehicle accident, parking time t n And area, time, license plate information. If t n Less than N t , then the output event is: suspected accident, parking time t n And area, time, license plate information.

[0111] Parking time threshold N t The determination is the result of comprehensive consideration of multiple factors. Even if there is no drastic speed change, too small distance between vehicles, double flash door opening and other behaviors, if a vehicle is stationary for an accumulated time t n Exceeding the threshold N t , and accident events will also be output.

[0112] However, if the vehicle queries the historical movement abnormality (dramatic speed change, too small distance between vehicles) or triggers the image judgment to find a specific behavior (double flash, open door), it can be clarified based on this, which is manifested by reducing the threshold N t The value of is reflected in faster (even immediate) output of accident events.

[0113] S3 vehicle event judgment flow chart

[0114] If the judgment box is not marked, the event judgment is not output and it is considered normal.

[0115] S4: Establish a personnel-related traffic incident judgment process and perform logical judgment in a certain order based on the information obtained in S2.

[0116] If it is judged as a personnel accident, the device deployment area (the area where the incident occurred), timestamp (the time when the incident occurred), and the event judgment result (personnel accident or suspected personnel accident) will be output.

[0117] This judgment process continues, and after each round of judgment of all pedestrians within the range is completed, the next round of judgment will be carried out based on the latest detection data.

[0118] S4.1 The postures of people in the area and the duration of their basic stillness at the current moment are obtained based on the image algorithm. If the posture of a person does not meet the requirements (such as lying flat), further analysis is performed on the person. If the posture of the person meets the requirements, it is considered to be normal.

[0119] S4.2 The person's current posture is abnormal, and S2.9 is triggered to calculate the minimum distance dy between the person and the vehicle in the recent period of time.

[0120] If the minimum distance to the vehicle in the recent period of time dy<N q , then the output event is: personnel accident, and S2.8 is triggered to obtain the static duration u n, accident area (equipment installation area), timestamp, and vehicles that were too close to it, license plate numbers and other information.

[0121] If the minimum distance to the vehicle in the recent period of time dy ≥ N q , it is considered that the person has not been too close to the car for some time, and enters S4.3 for further judgment.

[0122] S4.3 triggers S2.8 to obtain the duration u of the person being basically still at this time n , with the threshold N u Compare and determine whether it is in a static state for a long time.

[0123] If u n ≥N u , it is considered to be a personnel accident, and the output event is: personnel accident, the duration of basic stillness u n and region, timestamp.

[0124] At this time, if the person has been stationary for too long, an alarm should be output even if there are no cars passing nearby.

[0125] Otherwise, it is considered to be a suspected personnel accident, and the output event is: suspected personnel accident, duration u of basic stillness n The suspected personnel accident can be considered as a low-level alarm, that is, it cannot be identified at this time, but as its static time increases, it will be transformed into a personnel accident alarm.

[0126] S5 video data storage

[0127] When it is determined that a vehicle or person is abnormal, the characteristics of the abnormal period and the video images will be persistently stored according to the timestamp of the determined data. The specific storage period can be extended before and after the abnormal period according to the settings. The images can be used as evidence and can also be used for manual review.

[0128] Embodiment 1:

[0129] This example provides vehicle accident detection based on traffic flow monitoring:

[0130] The speed, position and direction of the vehicles in the area are continuously detected and stored to form V, X and H data which are stored in a table. The three have the same format. The number of columns in the table depends on the number of vehicles in the detection range, and the number of rows depends on the number of historical data n retained. For example, detection is recorded once per second, δ=1, and the number of data n=10 (retaining 10 historical data) is used.

[0131] In this example, data collection starts at 11:00:01, and judgment begins when 10 records are collected (from 11:00:10 seconds), and no personnel appear during the entire process.

[0132] After pressing S1 to confirm the detection area,

[0133] At 11:00:10 seconds, V, X, and H are obtained according to S2.1 as follows:

[0134]

[0135] According to S3, all vehicles in the area are judged according to the design process. According to S3.1, the current speed of all vehicles is 4.30, and there is no stationary vehicle. No abnormal event is output.

[0136] The schematic diagram of the situation is as attached Figure 4 As shown;

[0137] At the next moment 11:00:11 seconds, S2.1 obtains V, X, and H as follows:

[0138]

[0139] Press S3 to judge each vehicle.

[0140] Perform S3.1 to check the speed. The speed of car 1 is 0. Enter S3.2 to further analyze car 1. Perform S3.2 and obtain its position with the roadside according to S2.5. According to its current position 7,184 and the road boundary, it is known that its current distance from the roadside is 7, which is greater than the preset threshold N. dr =1, indicating that it is not parked on the roadside. According to S2.4, its queue status and queue coefficient are calculated. At this moment, there is a vehicle 2 at a certain distance in front of it, but there are no other vehicles within a certain distance in front of vehicle 2, and it is not at the stop line or boundary. It is considered that it is not in the queue.

[0141] Go to S3.3, preset parking cumulative time threshold N t = 10 seconds. Depending on the acceleration, distance, and whether there is a double flash door, each item reduces the threshold by 4 seconds (the lowest threshold is 0 seconds)

[0142] According to S2.2, the acceleration sequence is calculated by differentially calculating the velocity sequence 22,22,22,22,22,20,15,15,4,0 to get the acceleration sequence 0,0,0,0,2,5,0,11,4, where the maximum acceleration is 11, which is greater than the threshold N. a =10, indicating that there is a drastic speed change, the parking cumulative time threshold N t Reduced by 4 seconds.

[0143] Calculate the minimum distance sequence between car 1 and other vehicles. The historical minimum distance sequence is obtained from the positions of car 1 and car 2. The minimum distance of 86, 78, 68, 56, 41, 26, 2 is 2, which is less than the threshold N. d=5, indicating that the vehicle is located at a time when other vehicles are close to it, and the parking time threshold N t Reduce it by another 4 seconds.

[0144] The hazard lights and door detection are turned on. If no hazard lights or door detection is detected, the accumulated parking time threshold will not be further reduced.

[0145] Therefore, the parking cumulative time threshold is N t =2 seconds.

[0146] Calculate the current parking time t of the vehicle according to the formula in S2.3 n = 1 second, threshold N is not reached t , it is determined as suspected accident information at this moment. (In this example, if there are double flashes and the car door is opened at this time, the parking cumulative time threshold N t will be reduced to 0 seconds, and the current parking time t n = 1 second, if the threshold is reached, it will be directly judged as a vehicle accident)

[0147] The schematic diagram of the situation is as attached Figure 5 As shown;

[0148] At the next moment, 11:00:12 seconds, V, X, and H are as follows:

[0149]

[0150] At this point, each vehicle is judged according to process 3:

[0151] Press S3.1 to check the speed. The speed of car 1 is 0. Further analyze car 1.

[0152] According to S3.2, its position with the roadside is obtained. According to its current position 7,184 and the road boundary, its current distance from the roadside is 7, which is greater than the preset threshold N. dr =1, indicating that it is not parked on the roadside.

[0153] Calculate its queue status and queue coefficient. At the current moment, there is no vehicle in front of it within a certain distance, and it is not at the stop line or boundary. It is considered that it is not in the queue.

[0154] According to S3.3, the preset parking cumulative time threshold is 10 seconds. Depending on the acceleration, distance, and whether there is a double flash door, the threshold is reduced by 4 seconds for each item (the minimum threshold is 0 seconds)

[0155] Calculate the acceleration sequence. The acceleration sequence is calculated by differentially calculating the velocity sequence 22,22,22,22,20,15,15,4,0,0 to be 0,0,0,2,5,0,11,4,0. The maximum acceleration is 11, which is greater than the threshold N. a=10, indicating that there is a drastic speed change, the parking cumulative time threshold N t Reduced by 4 seconds.

[0156] Calculate the minimum distance sequence between car 1 and other vehicles. The historical minimum distance sequence is obtained from the positions of car 1 and car 2. The minimum distance of 78, 68, 56, 41, 26, 2, 32 is 2, which is less than the threshold N. d =5, indicating that the vehicle is located at a time when other vehicles are close to it, and the parking time threshold N t Reduce it by another 4 seconds.

[0157] The hazard lights and door detection are turned on. If no hazard lights or door detection is detected, the accumulated parking time threshold will not be further reduced.

[0158] Therefore, the parking time threshold N t =2 seconds.

[0159] At the same time, calculate the current parking time t of the vehicle according to the formula in 2.3 n = 2 seconds, reaching the threshold N t , it is now determined to be a vehicle accident.

[0160] The schematic diagram of the situation is as attached Figure 6 As shown;

[0161] In this example, the two cars collide, and even if car 2 escapes after the collision, since car 1 is stationary, the vehicle accident judgment of car 1 will still be output. If both cars stop after the collision, the vehicle accident of both cars 1 and 2 will be output.

[0162] Embodiment 2:

[0163] This example provides vehicle congestion detection based on vehicle flow monitoring, and its conditions are the same as those of Example 1: the speed, position, and direction of vehicles in the area are continuously detected and stored to form V, X, and H data and store them in a table. The three have the same format. The number of columns in the table depends on the number of vehicles within the detection range, and the number of rows depends on the number of historical data items n retained. Take detection and recording once per second, δ=1, and the number of data items n=10 (retaining 10 historical data items) as an example.

[0164] In this example, data collection starts at 11:00:01, and judgment begins when 10 records are collected (from 11:00:10 seconds), and no personnel appear during the entire process.

[0165] At 11:00:10 seconds, V, X, and H are as follows:

[0166]

[0167] At this time, each vehicle is judged according to S3.

[0168] Car 1, with a speed of 0, continues to determine its position. At this point, the position is 6,200 and the distance from the roadside is 6, which is greater than the set threshold N. dr =1, it is considered to be stopped in the middle of the road.

[0169] Check its queue status. There is no car in front of it, but there is a stop line nearby. It is clear that it is in the queue. The queue sequence number s is 1, which is less than the set threshold N. s = 3. It is considered normal and no subsequent operation is performed.

[0170] Car 2, with a speed of 0, continues to determine its position. At this point, it is at 6,195 and is 6 meters from the roadside, which is greater than the set threshold N. dr =1, it is considered to be stopped in the middle of the road. It is judged to be normal and no subsequent operation is performed.

[0171] Check its queue status. There is a car in front of it. There is a stop line in front of the first car. It is clear that it is in the queue. The queue number s is 2, which is less than the set threshold N s = 3. No subsequent operation is performed.

[0172] Car 3, with a speed of 0, continues to determine its position. At this point, its position is 6,190, and its distance from the roadside is 6, which is greater than the set threshold N. dr =1, it is considered to be stopped in the middle of the road.

[0173] Check its queue status. There is a car in front of it. There is a stop line in front of the first car. It is clear that it is in the queue. The queue sequence number s is 3, which is greater than or equal to the set threshold N. s =3. Calculate the cumulative waiting time t of the vehicle n =1 second.

[0174] Output event, queue waiting, queue length 3, waiting time 1 second. (Duration of queue length reaching threshold 3), the situation at this time is shown in the attached figure Figure 7 As shown;

[0175] At 11:00:11 seconds, V, X, and H are as follows:

[0176]

[0177]

[0178] At this time, car 1 has started to move, but the queue still exists.

[0179] Press S3 to judge all cars in turn.

[0180] Car 1, with a speed of 5, is considered to be in normal condition.

[0181] Car 2, with a speed of 0, continues to determine its position. At this point, it is at 6,195 and is 6 meters from the roadside, which is greater than the set threshold N.dr =1, it is considered to be stopped in the middle of the road.

[0182] Check its queue status. There is a car in front of it. The first car (car 1) is in front of the stop line. It is clear that it is in the queue. The queue number s is 2, which is less than the set N. s = 3. No subsequent operation is performed.

[0183] Car 3, with a speed of 0, continues to determine its position. At this point, its position is 6,190, and its distance from the roadside is 6, which is greater than the set threshold N. dr =1, it is considered to be stopped in the middle of the road.

[0184] Check its queue status. There is a car in front of it. The first car (car 1) is in front of the stop line. It is clear that it is in the queue. The queue sequence number s is 3, which is greater than or equal to the set threshold N. s =3. Calculate the cumulative waiting time t of the vehicle n =2 seconds.

[0185] Output event: waiting in line, queue length 3, waiting time 2 seconds. (Duration of queue length reaching threshold 3), the situation is as follows Figure 8 As shown;

[0186] At 11:00:12 seconds, V, X, and H are as follows:

[0187]

[0188] Press S3 to judge all cars in turn.

[0189] Car 1 and Car 2 have speeds other than 0, which is considered normal.

[0190] Car 3, with a speed of 0, continues to determine its position. At this point, its position is 6,190, and its distance from the roadside is 6, which is greater than the set threshold N. dr =1, it is considered to be stopped in the middle of the road.

[0191] Check its queue status. There is a car in front of it. There is a stop line in front of the first car (car 2). It is clear that it is in the queue. The queue sequence number s is 2, which is less than the queue threshold N. s =3, judge it as normal, and do not perform subsequent operations.

[0192] At this time, the situation is as follows Fig. 9 As shown;

[0193] If subsequently, car 3 starts normally and has a non-zero speed, then all three cars are in normal condition.

[0194] If car 3 does not start for a long time, there will be no car in front of it (car 1 and car 2 have left), and the car is far away from the stop line, and it is considered to be in a non-queuing state. At this time, the speed of car 3 is 0, it is far away from the roadside, and it is not in a queue state. The process will continue to S3.2 to determine the vehicle accident.

[0195] Embodiment 3:

[0196] This example provides a personnel accident detection based on vehicle flow monitoring. It continuously detects and stores the speed, position, and direction of vehicles in the area, forming V, X, and H data and storing them in a table. It also continuously detects and stores the position and direction of people in the area, forming Y data and storing them in a table.

[0197] In this example, the number of data items of people and vehicles is the same, m=n=10. The detection intervals of people and vehicles are the same, one detection record per second, δ=λ=1.

[0198] In this example, data collection starts at 11:00:01, and judgment begins when 10 records are collected (from 11:00:10 seconds).

[0199] At 11:00:10 seconds, V, X, H, and Y are as follows:

[0200]

[0201] At this time, the judgment in S3 is performed on car 1. The speed of the car is not 0, so it is considered normal at this moment. The judgment in S4 is performed on pedestrian 1. Its current posture is standing, which meets the requirements, so it is considered normal at this moment. At 11:00:11 seconds, V, X, H, and Y are as follows:

[0202]

[0203]

[0204] At this time, the judgment in S3.1 is performed on vehicle 1, and the speed of the vehicle is 0.

[0205] Go to S3.2 and determine that the distance from the roadside is 5, which is greater than the threshold N dr =1, it is considered to be stopped in the middle of the road; judging its queue sequence number, it is found that there are no other vehicles around it and no stop line, so it is in a non-queuing state.

[0206] Go to S3.3, preset parking time threshold N t = 10 seconds, according to the formula, the maximum speed change 15 is greater than the threshold N a = 10, reducing 4 seconds, the minimum distance between the historical and human is 2, which is less than the distance threshold N d =5, reducing 4 seconds, the final parking time threshold is N t= 2 seconds. According to formula 2.3, the current parking time is t n = 1 second, threshold N is not reached t , it is temporarily believed that it is not a vehicle accident, but a suspected vehicle accident.

[0207] Pedestrian 1 is judged in S4, and its current non-standing state is judged in 4.1S. Then, the process goes to S4.2, and the minimum distance between the person and the car is calculated to be 2, which is less than the distance threshold N. q =5, output event: personnel accident, by calculating that the personnel is stationary for 0 seconds, and the vehicle that is too close to him is vehicle 1.

[0208] At 11:00:12 seconds, V, X, H, and Y are as follows:

[0209]

[0210] At this time, the judgment in S3.1 is performed on vehicle 1, and the speed of the vehicle is 0.

[0211] Go to S3.2 and determine that the distance from the roadside is 5, which is greater than the threshold N dr =1, it is considered to be stopped in the middle of the road; judging its queue sequence number, it is found that there are no other vehicles around it and no stop line, so it is in a non-queuing state.

[0212] Go to S3.3, preset parking time threshold N t = 10 seconds, according to the formula, the maximum speed change 15 is greater than the threshold N a = 10, reducing 4 seconds, the minimum distance between the historical and human is 2, which is less than the distance threshold N d =5, reducing 4 seconds, the final parking time threshold is N t = 2 seconds. According to formula 2.3, the current parking time is t n = 2 seconds, reaching the threshold N t , output vehicle accident. Pedestrian 1 is judged in S4, and the judgment process is the same as the last time, and a personnel accident event is still output. By calculating the time the person is stationary for 1 second, the vehicle that is too close to him is vehicle 1. Similar to Example 1, in this example, the vehicle stops after the accident, and the vehicle accident and the personnel accident will be output. Even if the vehicle escapes, the personnel accident will still be output, and the personnel accident has nothing to do with the vehicle situation.

Claims

1. A multi-dimensional dynamic control method for traffic events based on vehicle flow monitoring, characterized in that: Includes the following: Step 1: Obtain the detection data in the detection area: Continuously obtain and store the speed, direction, position, and head direction data of each vehicle in the detection range at the current moment. From this, the speed sequence V of all vehicles can be obtained. V is composed of multiple single-vehicle speed sequences v, v = (v1, v2, ..., v n ), where v n is the current speed, v1~v n-1 is the historical speed detected and stored before, and the value of n depends on the number of historical data to be retained. Similarly, the position sequence X of all vehicles and the head direction sequence H of all vehicles can also be obtained; Step 2: Perform event judgment on each vehicle in the detection area: Step 2-1: Calculate the maximum speed change, i.e., the maximum acceleration a, of a specified vehicle in the recent period of time. The formula is: δ is a coefficient that depends on the time interval between consecutive speed detections in step 1; Step 2-2: Calculate the current cumulative parking time t for the specified vehicle n The formula is: t0 = 0, 1≤i≤n, δ is a coefficient, which depends on the time interval of continuous speed detection in step 1; Step 2-3: Determine whether the specified vehicle is waiting in the queue, and calculate its position in the queue of the current road section, which is called the queue number s: specifically: Step 2-3-1: Set the queue number s=1 for the specified vehicle and obtain its current vehicle head direction h n ； Step 2-3-2: According to the current positions of all vehicles in the position sequence X of all vehicles, check whether there is a vehicle within a certain distance in front of the vehicle. If there is a vehicle in front, increase the queue number s by 1, and check the front direction of the vehicle in front according to the head direction sequence H of all vehicles. According to the current head direction of the vehicle in front and the current positions of all vehicles, continue to determine whether there is a vehicle in a certain range in front of it; loop until it is found that there is no vehicle in front within a certain range, or there is a stop line or image boundary in front. If the loop is terminated because there is no vehicle in front within a certain range, it is considered that the vehicle is not waiting in the queue. Otherwise, if the loop is terminated because there is a stop line or image boundary in front, it is considered that the vehicle is waiting in the queue, and the value of s is the queue number; Step 2-4: Calculate the distance dr between the specified vehicle and the roadside at the current moment. The specific method is: according to the current position x of the vehicle n And the road boundary information R, find the minimum distance between the vehicle and all roadsides; Step 2-5: Calculate the minimum distance d between a specified vehicle and other vehicles in the recent period of time. The formula is: d i It is the minimum distance between a specified vehicle and all other vehicles at each detection moment in the recent period of time; Step 3: Perform event judgment for each pedestrian in the detection area: Step 3-1: Through the camera and its configured position detection and tracking algorithm, the current posture of each person in the detection range is determined, and the current position is continuously detected and stored. The posture of all people in the range and the current position data of all pedestrians are derived from the detection and positioning algorithm based on images and radars; thus, the position sequence Y of all people can be obtained. Y is composed of the position sequence y of multiple single people from history to the current moment, y=(y1,y2,...,y m ), where y m is the current position, y1~y m-1 is the historical location of the previous detection storage; the value of m depends on the number of historical data to be retained; Step 3-2: Based on the position sequence y of the specified person, the current stationary time u of the person can be obtained n , the formula is: u0=0, distance(y j ,y j-1 ) refers to the position y of a person at a certain time j The position y at the previous detection time j-1 The distance, γ is a coefficient, which depends on the time interval of continuous detection of the personnel position; Step 3-3: Calculate the minimum distance dx between the specified vehicle and the pedestrian in the recent period of time. The formula is: dx i It is the minimum distance between the specified vehicle and all pedestrians at each detection moment in the recent period of time; Step 3-4: Calculate the minimum distance dy between the specified pedestrian and the vehicle in the recent period of time. The formula is: dy j It is the minimum distance between a pedestrian and all vehicles at each detection moment in the recent period of time. Step 3-5: Detect whether the vehicle in the range has its doors open or has its hazard lights on. The data source is based on the image recognition algorithm. Step 3-6: Identify the license plates of all vehicles within the range, obtain the license plate information, and establish the correspondence between the speed, position, and vehicle head direction data in step 1 and the license plate number. The data source is based on the image recognition algorithm; Step 4: Establish a vehicle-related traffic event judgment process, and perform logical judgment based on the information obtained in step 2. If it is judged as an abnormal event, the output device layout area, i.e., the event occurrence area, the timestamp, i.e., the event occurrence time, the license plate information, and the event type logical judgment result; The abnormal events in step 4 include speeding, driving in the wrong direction, reversing, pulling over, waiting in line, suspected vehicle accidents, and vehicle accidents. The judgment process in step 3 continues, and each round judges each vehicle in the range in turn. After completing the judgment of all vehicles in the range, the next round of judgment will be carried out based on the latest detection data.

2. According to claim 1, a multi-dimensional dynamic control method for traffic events based on vehicle flow monitoring is characterized in that: In step 4, the vehicle speed and direction are used to determine whether to stop the vehicle, as well as the speeding, reverse driving, and reverse driving conditions as follows: For a single vehicle, the current speed y n : Step 4-1-1: If the vehicle speed direction is the same as the road requirement, and the vehicle speed v n Greater than or equal to threshold N v , output events for the vehicle: speeding, speed value, and area, time, and license plate information; Step 4-1-2: If the vehicle speed direction is opposite to the road requirement, obtain the vehicle head direction h at the current moment n : Step 4-1-3; if the vehicle head direction is h n If the direction is the same as the road requirement, the following events are output for the vehicle: reversing, as well as area, time, and license plate information; Step 4-1-4: If the vehicle head direction is h n If the vehicle is not in the same direction as the road requirement, the event output for the vehicle is: reverse driving, as well as the area, time, and license plate information. If the vehicle speed is equal to 0, execute step 4-2; Step 4-2: The vehicle entering this step has stopped and will be judged whether it is roadside parking, queue parking, or further confirmed based on the vehicle position; When entering step 4-2, that is, when there is parking, step 2-5 is triggered to obtain the distance dr between the car and the roadside, and perform the following operations: If the distance to the roadside dr is less than the threshold N dr , then trigger step 2-2, obtain the accumulated parking time t of the vehicle n , output event for the vehicle: roadside parking, parking time t n , and area, time, and license plate information; If the distance to the roadside dr is not less than the threshold N dr , then trigger step 2-3, obtain the queue sequence number s and whether the car is waiting in line: if the car is waiting in line and the queue sequence number s is greater than the threshold N s , then trigger step 2-2, obtain the accumulated parking time t of the vehicle n , then the output event for the vehicle is: waiting in line, parking time t n , and area, time, and license plate information. If the vehicle is not waiting in line, it is considered that the possibility of an accident is high, and the process goes to step 4-3 for further confirmation; Step 4-3: The vehicle entering this step is a stationary vehicle and the parking position is abnormal. It will be further confirmed based on the vehicle's historical status and the current scene to determine whether it is a vehicle accident. When entering step 4-3, step 2-1 is triggered to obtain the maximum speed change of the vehicle in the recent period, that is, the maximum acceleration a; step 2-2 is triggered to obtain the accumulated parking time t of the vehicle. n ; Trigger step 2-5: obtain the minimum distance d between it and other vehicles in the recent period; trigger step 3-3, obtain the minimum distance dx between it and pedestrians in the recent period; trigger opening step 3-5, obtain whether there is double flashing and opening the door; Parking time alarm threshold N t There is a preset value that will adjust the parking time alarm threshold N according to the following information t :When the maximum speed changes, that is, the maximum acceleration a is greater than or equal to the threshold N a When the parking time threshold N t When the vehicle's minimum distance d from other vehicles and the minimum distance dx from pedestrians in the recent period is less than the threshold N d When the parking time threshold N t When the area has double flashing or door opening, the parking time threshold N t To reduce; Compare the accumulated parking time t of the vehicle n and the reduced parking time threshold N t , if t n Greater than or equal to N t , then the output event is: vehicle accident, parking time t n and area, time, license plate information; if t n Less than N t , then the output event is: suspected accident, parking time t n And area, time, license plate information.

3. According to claim 1, a multi-dimensional dynamic control method for traffic events based on vehicle flow monitoring is characterized in that: The parking time threshold N in step 4 t The determination is the result of comprehensive consideration of multiple factors. Even if there is no drastic speed change, too small distance between vehicles, double flash door opening and other behaviors, if a vehicle is stationary for an accumulated time t n Exceeding the threshold N t , it will also output the accident event, but if the vehicle queries the historical movement abnormality, the historical movement abnormality refers to the drastic speed change, the distance between vehicles is too small, or the image judgment is triggered to find a specific behavior, the specific behavior is double flash, open the door, which can be clarified based on this, which is manifested by reducing the threshold N t The value of is reflected in faster or immediate output of accident events.

4. According to claim 1, a multi-dimensional dynamic control method for traffic events based on vehicle flow monitoring is characterized in that: It also includes step 5, establishing a personnel-related traffic incident judgment process, and performing logical judgment in a certain order based on the information obtained in step 3. If it is judged to be a personnel accident, the output device layout area, i.e., the event occurrence area, the timestamp, i.e., the event occurrence time, and the event judgment result, i.e., a personnel accident or a suspected personnel accident, will be output. The judgment process continues, and after each round of judgment of all pedestrians within the range is completed, the next round of judgment will be performed based on the latest detection data.

5. The multi-dimensional dynamic control method for traffic events based on vehicle flow monitoring according to claim 1 is characterized in that: Step 5 specifically includes the following steps: Step 5-1: Obtain the posture of people in the area and the duration of their basic stillness at the current moment according to the image algorithm. If the posture of a person does not meet the requirements, further analyze the person. If the posture of the person meets the requirements, it is considered normal. Step 5-2: The person's current posture is abnormal, and step 3-4 is triggered to obtain the minimum distance dy between the person and the vehicle in the recent period of time. If the minimum distance dy between the person and the vehicle in the recent period of time is <N q , then the output event is: personnel accident, and step 3-2 is triggered to obtain the static duration u n , the accident area, i.e. the equipment installation area, timestamp, and vehicles that are too close to it, license plate number and other information; if the minimum distance to the vehicle in the recent period of time dy ≥ N q , it is considered that the person has not been too close to the car for some time, and further judgment is made in step 5-3; Step 5-3: Trigger step 3-2 and obtain the duration u of the person being basically still at this time n , and the threshold N u Compare and judge whether it is in a static state for a long time. If u n ≥N u , it is considered to be a personnel accident, and the output event is: personnel accident, the duration of basic stillness u n And area, timestamp, at this time, the person is stationary for too long, even if there is no car passing nearby, an alarm should be output, otherwise, it is considered to be a suspected personnel accident, and the output event is: suspected personnel accident, basic stationary duration u n And area, timestamp, suspected personnel accident can be considered as a low-level alarm, that is, it cannot be identified at this time, but as its static time increases, it will be transformed into a personnel accident alarm.

6. The multi-dimensional dynamic control method for traffic events based on vehicle flow monitoring according to claim 1 is characterized in that: It also includes step 6, video data storage. When it is judged that there is an abnormality in the vehicle or personnel, the characteristics of the abnormal period and the video image will be persistently stored according to the timestamp of the judged data. The specific start and end time of the storage period is extended before and after the abnormal period according to the setting. The image is used as evidence or for manual review.

Images