Intelligent traffic management system based on multi-scene recognition and self-adaptive dispersion

Abstract

The invention belongs to the technical field of intelligent traffic management, and discloses an intelligent traffic management system based on multi-scene recognition and adaptive dispersion. Comprising the following steps: collecting accident video information, extracting and analyzing accident vehicle and personnel information based on a YOLOv5 model, and combining accident feature information and an accident type number sequence to construct different deep learning models which are respectively used for predicting damage / casualty evaluation values of accident vehicles and personnel in real time and identifying the accident type number sequence. The unmanned aerial vehicle participates in the accident evacuation process according to the accident type patient sequence identified in real time and a preset corresponding evacuation scheme and instruction, displays the accident type to both parties of the accident and provides an intelligent mediation instruction and responsibility determination service; if the two parties of the accident doubt the intelligent mediation instruction, the instruction is handled manually by a traffic management department.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent traffic management, and more specifically, to an intelligent traffic management system based on multi-scenario recognition and adaptive guidance. Background Art

[0002] Multi-scenario recognition is to identify different traffic scenarios through data collection and analysis to provide targeted solutions. Adaptive traffic control is to adjust the signal lights, lane allocation or induction strategies according to the real-time traffic status and vehicle scenarios, so as to optimize traffic flow and reduce congestion. Intelligent traffic management refers to the use of modern information technology, communication technology, control technology, electronic sensor technology, network technology, etc. to efficiently manage the traffic system. The intelligent traffic management system can monitor traffic conditions in real time, dynamically adjust traffic signals, provide traffic information services, etc., thereby improving the intelligence level of the entire traffic system.

[0003] The Chinese patent application publication number CN118470983A discloses an intelligent traffic management method and system based on vehicle identification, which implements real-time identification and data capture of vehicles in traffic flow to obtain relevant image information of vehicles; performs preliminary processing on the collected vehicle data; imports the pre-processed vehicle data into a database for systematic management; performs in-depth indicator analysis on the vehicle data in the database; and formulates corresponding traffic management strategies based on the results of data analysis.

[0004] The Chinese patent with authorization announcement number CN114582141B discloses a method and system for identifying sections with frequent heavy traffic. The patent is based on massive historical traffic big data, statistically analyzes the probability distribution of section traffic volume and traffic saturation, and calculates the probability that the section traffic volume and traffic saturation exceed the set threshold. In this way, traffic flow is identified and balanced distribution of traffic flow is guided.

[0005] However, the above patents have failed to effectively reduce manual involvement in traffic flow decision-making, and have failed to reduce labor costs while improving traffic safety in different complex traffic scenarios and improving driving efficiency under complex vehicle conditions.

[0006] In view of this, the present invention proposes an intelligent traffic management system based on multi-scenario recognition and adaptive guidance to solve the above problems. Summary of the invention

[0007] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides an intelligent traffic management system based on multi-scenario recognition and adaptive guidance.

[0008] To achieve the above object, the present invention provides the following technical solution: an intelligent traffic management method based on multi-scenario recognition and adaptive guidance, the method comprising:

[0009] Collect accident video information, accident types and accident section environmental assessment values; accident types are classified according to the severity, scale and cause of the accident;

[0010] Different numbers are set to correspond to the severity, scale and cause of the accident; a group of numbers for one type of accident is called an accident type number sequence; a corresponding diversion plan is preset for each type of accident;

[0011] Based on the accident video information, the yolov5 model is used to extract accident feature information; the accident feature information includes: the number of accident vehicles and the number of accident personnel;

[0012] The first machine learning model that can estimate the damage assessment value of the accident vehicle and the casualties assessment value of the accident personnel in real time based on the accident video information training;

[0013] A second machine learning model for real-time identification of accident types is trained based on accident video information, the number of accident vehicles, the number of accident personnel, the accident road section environmental assessment value, the accident type number sequence, the accident vehicle damage assessment value and the accident personnel casualty assessment value predicted by the first machine learning model;

[0014] Generate diversion instructions based on the accident type number sequence identified by the second machine learning model.

[0015] Preferably, the accident video information is a video clip of the scene after the accident occurs;

[0016] The severity of the accident refers to the personal injury and property loss caused by the accident; the scale refers to the number of vehicles involved in the accident and the process of the accident; the cause of the accident refers to the direct cause of the accident;

[0017] The damage assessment value of the accident vehicle is the overall damage of the vehicle as shown in the accident monitoring screen;

[0018] The casualty assessment value of the accident is the casualty assessment status of the personnel appearing on the accident monitoring screen;

[0019] The environmental assessment value of the accident section is the environmental risk level of the accident site; the environmental assessment value of the accident section is a rational number between 0 and 1.

[0020] Preferably, the yolov5 model acquisition process includes:

[0021] Extract frames from the accident video information to obtain key frames per second; denoise the key frames to enhance image quality; input the denoised key frames into the yolov5 model for detection and analysis; output the vehicle type in the frame and mark its position bounding box; adopt the target tracking algorithm based on the vehicle's position bounding box to delete the duplicate counts of the same vehicle; after deleting the duplicate counts, count the total number of different vehicle types;

[0022] The target tracking algorithm is DeepSORT; the process of DeepSORT target tracking algorithm includes:

[0023] Extract features; extract deep features for each detected target;

[0024] Trajectory management: create or update a trajectory for each target, assign a unique ID, and predict the trajectory state using a Kalman filter;

[0025] Matching trajectories: Matching detection results with existing trajectories by calculating the Mahalanobis distance and feature similarity of the same target in consecutive frames;

[0026] Update the target; delete the tracks that have not been matched successfully for a long time and update the target track at all times.

[0027] Preferably, the training process of the first machine learning model includes:

[0028] The accident video information is converted into a feature vector, the damage assessment value of the accident vehicle and the casualty assessment value of the accident personnel are used as labels corresponding to the feature vector, each set of feature vectors and the labels corresponding to each set of feature vectors are constructed as a sample, and multiple samples are collected to construct a data set; the data set is divided into a training set, a validation set and a test set, wherein the training set accounts for 60% of the data set, and the validation set and the test set each account for 20% of the data set;

[0029] The damage degree of each vehicle is manually labeled, and the damage degree is a rational number from 0 to 1. The damage assessment value of the accident vehicle is the sum of the damage degrees of all vehicles in the accident video information;

[0030] The methods for obtaining the accident casualty assessment value include:

[0031] Collect key point information of the human body; the key point information of the human body is the three-dimensional position coordinates (x, y, z) of the main parts of the human body; the origin of the three-dimensional coordinate system where the three-dimensional position coordinates are located is the location of the surveillance camera; the x-axis and y-axis are parallel to the ground, and the z-axis is a ray starting from the origin and perpendicular to the ground upward; collect key point information of the human body on the head and hips;

[0032] According to the formula Calculate the trunk inclination angle θ; where: x H ,y H , z H are the three-dimensional position coordinates of the head; L ,y L , z L is the three-dimensional position coordinate of the hip; the torso inclination angle is compared and analyzed with the preset angle threshold. When the torso inclination angle is less than or equal to the angle threshold, and at this time the z-axis coordinate of the head is zH When it is lower than 50cm, it is marked as a critically ill person. Each time a critically ill person is determined, the critically ill assessment value is increased by 1. The critically ill assessment value is initially 0. At the same time, professionals score the injuries of the people according to the images of the accident video information. The injury score is a rational number between 0 and 1. The larger the score, the greater the degree of casualties. The injury scores of all people and the critically ill assessment values ​​are accumulated as the accident casualties assessment value.

[0033] The training set is used as the input of the first machine learning model, and the first machine learning model uses the real-time accident vehicle damage assessment value and the accident casualty assessment value as the output; the accident vehicle damage assessment value and the accident casualty assessment value corresponding to a real-time set of feature vectors are used as prediction targets, and minimizing the loss function value of the first machine learning model is used as the training target; when the loss function value of the first machine learning model is less than or equal to the preset first target loss value, the training is stopped;

[0034] The first machine learning model is a deep neural network model; the first machine learning model loss function is a mean square error;

[0035] The calculation method of the hidden layer H of the deep neural network model includes:

[0036] H=G[b i +w i s(b i-1 +w i-1 x)]; where H is the output of the hidden layer; G is the activation function; b i is the bias term of the hidden layer; w i is the weight of the hidden layer; s is the activation function; b i-1 is the bias term of the previous layer; w i-1 is the weight of the previous layer; x is the input training set;

[0037] The deep neural network model consists of 4 fully connected layers. The first 3 hidden layers contain 32, 64 and 32 neurons respectively. One SoftMax classification layer contains 2 neurons. The neurons are connected to each other through weights and activated by the ReLU activation function.

[0038] Preferably, the training process of the second machine learning model includes:

[0039] The accident video information, the number of accident vehicles, the number of accident persons, the accident road section environmental assessment value, the accident vehicle damage assessment value and the accident person casualty assessment value are converted into a set of feature vectors as the input of the second machine learning model, and the second machine learning model uses the real-time accident type number sequence as the output; the accident type number sequence corresponding to a real-time set of accident video information, the number of accident vehicles, the number of accident persons, the accident road section environmental assessment value, the accident vehicle damage assessment value and the accident person casualty assessment value is used as the prediction target, and minimizing the second machine learning model loss function value is used as the training target; when the second machine learning model loss function value is less than or equal to the preset second target loss value, the training is stopped;

[0040] The second machine learning model loss function is a mean square error; the mean square error is calculated by Minimize to train the model; in the loss function, MSE is the loss function value, i is the feature vector group number; u is the number of feature vector groups; y i is the accident type number sequence corresponding to the i-th group of feature vectors, The accident type number sequence predicted by the i-th group of feature vectors.

[0041] Preferably, the diversion instruction includes executing any of the diversion plans described above; the diversion instruction also includes: setting a drone to go to the scene of the accident; the drone carries a terminal screen to display the accident type corresponding to the accident type number sequence; and gives instructions on whether both parties to the accident accept the intelligent mediation; when both parties to the accident accept the intelligent mediation instruction, the drone terminal determines the responsibility for the accident vehicle according to the type of accident; and confirms again whether both parties to the accident receive the responsibility determination instruction; when both parties to the accident accept the responsibility determination instruction, the electronic accident report is completed and returned to the traffic management department; when any party to the accident issues a questioning instruction regarding the intelligent mediation instruction or the responsibility determination instruction; the questioning instruction is returned to the traffic management department for manual processing.

[0042] Preferably, the traffic management department has full control over the diversion instructions and can control the drone terminal in real time.

[0043] An intelligent traffic management system based on multi-scene recognition and adaptive guidance, the system comprising:

[0044] The data collection module collects accident video information, accident types and accident section environmental assessment values; accident types are classified according to the severity, size and cause of the accident;

[0045] The pre-processing module sets different numbers to correspond to the severity, scale and cause of the accident; a group of numbers for one type of accident is called an accident type number sequence; and a corresponding diversion plan is preset for each type of accident;

[0046] Feature extraction module, which uses yolov5 model to extract accident feature information based on accident video information; accident feature information includes: number of accident vehicles and number of accident personnel;

[0047] A first model training module, which trains a first machine learning model for real-time estimation of the damage assessment value of the accident vehicle and the casualty assessment value of the accident personnel based on the accident video information;

[0048] The second model training module trains a second machine learning model for real-time identification of accident types based on accident video information, the number of accident vehicles, the number of accident personnel, the accident road section environmental assessment value, the accident type number sequence, the accident vehicle damage assessment value and the accident personnel casualty assessment value predicted by the first machine learning model;

[0049] The decision module generates diversion instructions based on the accident type number sequence identified by the second machine learning model.

[0050] An electronic device includes: a processor and a memory, wherein the memory stores a computer program that can be called by the processor; the processor executes the above-mentioned intelligent traffic management method based on multi-scenario recognition and adaptive guidance by calling the computer program stored in the memory.

[0051] A computer-readable storage medium stores instructions. When the instructions are executed on a computer, the computer executes the above-mentioned intelligent traffic management method based on multi-scenario recognition and adaptive guidance.

[0052] The technical effects and advantages of the intelligent traffic management system based on multi-scene recognition and adaptive guidance of the present invention are as follows:

[0053] The present invention collects accident video information and extracts and analyzes accident vehicle and personnel information based on the YOLOv5 model, combines accident feature information and accident type number sequence, and constructs two deep learning models, which are used to predict the damage / casualty assessment values ​​of accident vehicles and personnel in real time, and identify the accident type number sequence. After the drone identifies the accident type number sequence, it generates the corresponding diversion plan and instructions. The drone terminal participates in the diversion process, displays the accident type, and provides intelligent mediation instructions and liability determination services; if both parties to the accident question the intelligent mediation instructions, it will be handed over to the traffic management department for manual processing.

[0054] The present invention combines intelligent technology with drone assistance to achieve real-time identification, evaluation, guidance and rapid processing of road traffic accidents; its advantages are to improve the efficiency of accident scene processing, reduce manual participation, and reduce unnecessary human resource consumption; drone terminals can directly determine responsibility and complete electronic reports at the scene of the accident, greatly shortening the accident detention time, avoiding road congestion and secondary accidents. At the same time, the combination of deep learning-based prediction models and manual evaluation ensures the accuracy and flexibility of accident assessment and classification. The present invention can quickly respond to complex accident scenarios, provide efficient, intelligent, and automated decision-making support for traffic management departments, improve guidance safety in complex traffic scenarios, and reduce labor costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 A schematic diagram of an intelligent traffic management method based on multi-scenario recognition and adaptive guidance according to the present invention;

[0056] Figure 2 A schematic diagram of an intelligent traffic management system based on multi-scenario recognition and adaptive guidance according to the present invention;

[0057] Figure 3 It is a schematic diagram of the structure of an electronic device provided by an embodiment of the present application;

[0058] Figure 4 It is a schematic diagram of the structure of a computer-readable storage medium provided by an embodiment of the present application. DETAILED DESCRIPTION

[0059] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0060] Example 1

[0061] See also Figure 1 As shown, the intelligent traffic management system based on multi-scenario recognition and adaptive guidance described in this embodiment includes:

[0062] Collect accident video information; accident video information is the on-site video clip after the accident occurs; accident video information is obtained by the surveillance camera installed at the accident site.

[0063] Collect accident types; accident types are classified according to the severity, scale and cause of the accident, laying the foundation for subsequent multi-scenario identification; accident severity refers to personal injury and property loss caused by the accident; scale refers to the number of vehicles involved in the accident and the process of the accident; accident cause refers to the direct cause of the accident;

[0064] An example of an accident type includes: no casualties - a minor collision between two vehicles - the rear vehicle exceeding the speed limit; no casualties is the seriousness of the accident; and so on.

[0065] Different numbers are set to correspond to the severity, scale and cause of the accident. For example, in a certain accident, a means no casualties, b means a minor collision between two vehicles, and c means the rear vehicle is speeding. The numbering sequence of an accident is called abc; marked as the accident type numbering sequence;

[0066] Through the accident type number sequence, the different treatment measures required for different scenarios are reflected to the greatest extent possible. Complex scenarios can be preset and classified in advance, which is conducive to improving the traffic handling capabilities in complex situations at the temporary scene.

[0067] Extract accident feature information based on accident video information; accident feature information includes: number of accident vehicles; number of accident personnel; damage assessment value of accident vehicles; casualties assessment value of accident personnel; and environmental assessment value of accident road section;

[0068] The damage assessment value of the accident vehicle is the overall damage of the vehicle shown in the accident monitoring screen, which helps to analyze the accident status and intelligently recommend the corresponding treatment plan for the accident type;

[0069] The casualty assessment value of the accident is the casualty assessment status of the personnel appearing on the accident monitoring screen;

[0070] The environmental assessment value of the accident section is the environmental risk level of the accident site; the environmental assessment value of the accident section is preset by relevant professionals; the environmental assessment value of the accident section is a rational number between 0 and 1. For example, if the road section itself is narrow, visibility is low, and accidents are prone to occur, then the environmental assessment value of the accident section is high, and vice versa.

[0071] The number of accident vehicles is obtained through the yolov5 model; the acquisition process includes: extracting frames from the accident video information to obtain key frames per second; denoising the key frames to enhance the image quality; inputting the denoised key frames into the yolov5 model for detection and analysis; outputting the vehicle type in the frame and marking its position bounding box; adopting the target tracking algorithm based on the vehicle's position bounding box to delete the duplicate counts of the same vehicle; and counting the total number of different vehicle types after deleting the duplicate counts;

[0072] The target tracking algorithm may be one of DeepSORT or ByteTrack; an exemplary DeepSORT target tracking algorithm process includes:

[0073] Extract features; extract deep features for each detected target;

[0074] Trajectory management: create or update a trajectory for each target, assign a unique ID, and predict the trajectory state using a Kalman filter;

[0075] Matching trajectories: Matching detection results with existing trajectories by calculating the Mahalanobis distance and feature similarity of the same target in consecutive frames;

[0076] Update the target; delete the tracks that have not been successfully matched for a long time, and update the target track at all times to ensure that the same vehicle is not counted repeatedly.

[0077] It is understandable that the vehicle involved in the accident may not be a motor vehicle. Due to the current complex road conditions, there may be many different types of vehicles, such as motorcycles, bicycles, electric cars, wheelchairs, etc. The insurance coverage and accident liability involved in different types of vehicles are also different. Usually, manual judgment by traffic police is required. Presetting the vehicle type in advance can greatly alleviate the situation of limited traffic police resources. Therefore, it is necessary to output the vehicle type and mark the location boundary box.

[0078] YOLOv5 is an end-to-end target detection model that can predict the category and location bounding box of the target directly from the image. YOLOv5 is based on convolutional neural network (CNN) and can learn complex features in images, which enables it to identify different types of vehicles and even pedestrians. YOLOv5 optimizes the detection speed and can perform fast detection in real-time video streams; therefore, YOLOv5 can effectively assist traffic police departments in handling accidents and determining liability.

[0079] Number of accident persons: The process of obtaining the number of accident persons is the same as that of obtaining the number of accident vehicles, which will not be repeated here.

[0080] The first machine learning model that can estimate the damage assessment value of the accident vehicle and the casualties assessment value of the accident personnel in real time based on the accident video information training;

[0081] The training process of the first machine learning model includes:

[0082] The accident video information is converted into a feature vector, the damage assessment value of the accident vehicle and the casualty assessment value of the accident personnel are used as labels corresponding to the feature vector, each set of feature vectors and the labels corresponding to each set of feature vectors are constructed as a sample, and multiple samples are collected to construct a data set; the data set is divided into a training set, a validation set and a test set, wherein the training set accounts for 60% of the data set, and the validation set and the test set each account for 20% of the data set;

[0083] The damage degree of each vehicle is manually labeled. The damage degree is a rational number from 0 to 1. The greater the damage degree, the greater the damage to the vehicle. The damage assessment value of the accident vehicle is the sum of the damage degrees of all vehicles in the accident video information.

[0084] The methods for obtaining the accident casualty assessment value include:

[0085] Collect key point information of the human body; the key point information of the human body is the three-dimensional position coordinates (x, y, z) of the main parts of the human body; the origin of the three-dimensional coordinate system where the three-dimensional position coordinates are located is the location of the surveillance camera; the x-axis and y-axis are parallel to the ground, and the z-axis is a ray starting from the origin and perpendicular to the ground upward; collect key point information of the human body on the head and hips;

[0086] According to the formula Calculate the trunk inclination angle θ; where: x H ,y H , z H are the three-dimensional position coordinates of the head; L ,y L , z L is the three-dimensional position coordinate of the hip; the torso inclination angle is compared and analyzed with the preset angle threshold. When the torso inclination angle is less than or equal to the angle threshold, and at this time the z-axis coordinate of the head is z H When it is lower than 50cm, the person is marked as critically ill. Every time a person is determined to be in critical condition, the critical illness assessment value is increased by 1. The initial critical illness assessment value is 0. At the same time, professionals score the injuries of the people according to the pictures of the accident video information. The injury score is a rational number from 0 to 1. The larger the score, the greater the degree of casualties. The injury scores of all personnel and the critical illness assessment values ​​are added together as the accident casualties assessment value.

[0087] For example, if the emergency medical staff finds a person's legs bleeding heavily in the monitoring picture, the injury score is set to 0.9; if the person is also marked as critically ill, then 1 is added; if only one person is injured in the picture, the casualty assessment value of the accident is 1.9;

[0088] The system combines the trunk tilt angle and head height for automated critical illness detection, while also supplemented by manual scoring of injuries by professionals. This not only takes advantage of the efficiency and consistency of computers, but also incorporates the flexibility of human judgment, ensuring that the assessment results are more accurate and comprehensive. By combining multiple conditions (such as tilt angle, head height, and manual scoring), it avoids possible misjudgments caused by relying solely on a single factor, thus improving the accuracy of accident assessment.

[0089] The training set is used as the input of the first machine learning model, and the first machine learning model uses the real-time accident vehicle damage assessment value and the accident casualty assessment value as the output; the accident vehicle damage assessment value and the accident casualty assessment value corresponding to a real-time set of feature vectors are used as prediction targets, and minimizing the loss function value of the first machine learning model is used as the training target; when the loss function value of the first machine learning model is less than or equal to the preset first target loss value, the training is stopped;

[0090] The first machine learning model is a deep neural network model; the first machine learning model loss function is a mean square error;

[0091] The calculation method of the hidden layer H of the deep neural network model includes:

[0092] H=G[b i +w i s(b i-1 +w i-1 x)]; where H is the output of the hidden layer; G is the activation function; b i is the bias term of the hidden layer; w i is the weight of the hidden layer; s is the activation function; b i-1 is the bias term of the previous layer; w i-1 is the weight of the previous layer; x is the input training set.

[0093] The deep neural network model consists of 4 fully connected layers. The first 3 hidden layers contain 32, 64 and 32 neurons respectively. One SoftMax classification layer contains 2 neurons. The neurons are connected to each other through weights and activated by the ReLU activation function.

[0094] A second machine learning model for real-time identification of accident types is trained based on accident video information, the number of accident vehicles, the number of accident personnel, the accident road section environmental assessment value, the accident type number sequence, the accident vehicle damage assessment value and the accident personnel casualty assessment value predicted by the first machine learning model;

[0095] The accident video information, the number of accident vehicles, the number of accident persons, the accident road section environmental assessment value, the accident vehicle damage assessment value and the accident person casualty assessment value are converted into a set of feature vectors as the input of the second machine learning model, and the second machine learning model uses the real-time accident type number sequence as the output; the accident type number sequence corresponding to a real-time set of accident video information, the number of accident vehicles, the number of accident persons, the accident road section environmental assessment value, the accident vehicle damage assessment value and the accident person casualty assessment value is used as the prediction target, and minimizing the second machine learning model loss function value is used as the training target; when the second machine learning model loss function value is less than or equal to the preset second target loss value, the training is stopped;

[0096] The second machine learning model loss function is a mean square error; the mean square error is calculated by Minimize to train the model; in the loss function, MSE is the loss function value, i is the feature vector group number; u is the number of feature vector groups; y i is the accident type number sequence corresponding to the i-th group of feature vectors, is the accident type number sequence predicted by the i-th group of feature vectors;

[0097] Preset corresponding diversion plans for each accident type; for example, set up diversion plans for blocking road sections for multiple casualties, multiple vehicle serious collisions, and drunk driving;

[0098] Generate diversion instructions based on the accident type number sequence identified by the second machine learning model;

[0099] The diversion instruction includes executing any of the diversion plans; the diversion instruction also includes: setting a drone to go to the accident site; the drone carries a terminal screen to display the accident type corresponding to the accident type number sequence; and gives instructions to both parties of the accident on whether to accept intelligent mediation; when both parties of the accident accept the intelligent mediation instruction, the drone terminal determines the responsibility of the accident vehicle according to the accident type; and confirms again whether both parties of the accident accept the responsibility determination instruction; when both parties of the accident accept the responsibility determination instruction, the electronic accident report is completed and returned to the traffic management department; when any party of the accident issues a questioning instruction on the intelligent mediation instruction or the responsibility determination instruction; the questioning instruction is returned to the traffic management department for manual processing;

[0100] It should be noted that the traffic management department has full control over the diversion instructions and can control the drone terminal in real time to prevent drone terminal errors or inaccurate recognition of the accident type number by the second machine learning model.

[0101] When a low-level accident occurs, effective unmanned management can quickly and effectively clear the on-site roads to prevent long-term road blockages due to the accident, and the two parties involved in the accident occupying the road on the spot, affecting the traffic capacity of the road section; at the same time, the traffic management department has sufficient authority and response time to handle road accidents in real time, effectively reducing the waste of manpower resources in unnecessary situations; speeding up the overall response time of the accident and reducing the possibility of secondary accidents caused by staying on the spot after the accident.

[0102] Example 1 collects accident video information and extracts and analyzes accident vehicle and personnel information based on the YOLOv5 model, combines accident feature information and accident type number sequence, and constructs two deep learning models, which are used to predict the damage / casualty assessment values ​​of accident vehicles and personnel in real time, and identify the accident type number sequence. And generate corresponding diversion plans and instructions according to the preset number sequence. The drone terminal participates in the diversion process, displays the accident type, and provides intelligent mediation instructions and liability determination services; if both parties to the accident question the intelligent mediation instructions, it will be handed over to the traffic management department for manual processing.

[0103] Example 1 combines intelligent technology with drone assistance to achieve real-time identification, assessment, guidance and rapid processing of road traffic accidents; its advantages are to improve the efficiency of accident scene processing, reduce manual participation, and reduce unnecessary human resource consumption; drone terminals can directly determine responsibility and complete electronic reports at the scene of the accident, greatly shortening the accident detention time, avoiding road congestion and secondary accidents. At the same time, the combination of deep learning-based prediction models and manual evaluation ensures the accuracy and flexibility of accident assessment and classification. Example 1 can quickly respond to complex accident scenarios, provide efficient, intelligent, and automated decision-making support for traffic management departments, improve guidance safety in complex traffic scenarios, and reduce labor costs.

[0104] Example 2

[0105] See also Figure 2 As shown, the part not described in detail in this embodiment is described in Example 1, which provides an intelligent traffic management system based on multi-scene recognition and adaptive guidance, and the system includes:

[0106] The data collection module collects accident video information, accident types and accident section environmental assessment values; accident types are classified according to the severity, size and cause of the accident;

[0107] The pre-processing module sets different numbers to correspond to the severity, scale and cause of the accident; a group of numbers for one type of accident is called an accident type number sequence; and a corresponding diversion plan is preset for each type of accident;

[0108] Feature extraction module, which uses yolov5 model to extract accident feature information based on accident video information; accident feature information includes: number of accident vehicles and number of accident personnel;

[0109] A first model training module, which trains a first machine learning model for real-time estimation of the damage assessment value of the accident vehicle and the casualty assessment value of the accident personnel based on the accident video information;

[0110] The second model training module trains a second machine learning model for real-time identification of accident types based on accident video information, the number of accident vehicles, the number of accident personnel, the accident road section environmental assessment value, the accident type number sequence, the accident vehicle damage assessment value and the accident personnel casualty assessment value predicted by the first machine learning model;

[0111] The decision module generates diversion instructions based on the accident type number sequence identified by the second machine learning model.

[0112] The modules are connected via wired and / or wireless networks.

[0113] Example 3

[0114] See also Figure 3 As shown, according to another aspect of the present application, an electronic device 500 is also provided. The electronic device 500 may include one or more processors and one or more memories. The memories store computer readable codes, and when the computer readable codes are executed by one or more processors, an intelligent traffic management system based on multi-scenario recognition and adaptive diversion as described above may be executed.

[0115] The method or system according to the embodiment of the present application can also be Figure 3 The electronic device shown in the figure is implemented by the electronic device architecture. Figure 3 As shown, the electronic device 500 may include a bus 501, one or more CPUs 502, a read-only memory (ROM) 503, a random access memory (RAM) 504, a communication port 505 connected to a network, an input / output component 506, a hard disk 507, etc. The storage device in the electronic device 500, such as ROM 503 or hard disk 507, may store an intelligent traffic management system based on multi-scenario recognition and adaptive guidance provided by the present application. Furthermore, the electronic device 500 may also include a user interface 508. Of course, Figure 3 The architecture shown is only exemplary and can be omitted according to actual needs when implementing different devices. Figure 3 One or more components of an electronic device are shown.

[0116] Example 4

[0117] See also Figure 4As shown, a computer-readable storage medium 600 according to an embodiment of the present application is shown. Computer-readable instructions are stored on the computer-readable storage medium 600. When the computer-readable instructions are executed by the processor, an intelligent traffic management system based on multi-scene recognition and adaptive guidance according to an embodiment of the present application described with reference to the above figures can be executed. The storage medium 600 includes, but is not limited to, for example, volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and cache memory (cache), etc. Non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc.

[0118] In addition, according to the embodiments of the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the present application provides a non-transitory machine-readable storage medium, which stores machine-readable instructions, and the machine-readable instructions can be executed by a processor to execute instructions corresponding to the method steps provided by the present application. When the computer program is executed by a central processing unit (CPU), the above functions defined in the method of the present application are executed.

[0119] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented by software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present invention is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website site, computer, server or data center to another website site, computer, server or data center through a wired network or a wireless network. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that contains one or more available media sets. The available medium can be a magnetic medium (for example, a floppy disk, a hard disk, a tape), an optical medium (for example, a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state hard disk.

[0120] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed in the present invention can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0121] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0122] In the several embodiments provided by the present invention, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only an intelligent traffic management system based on multi-scene recognition and adaptive guidance. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0123] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0124] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0125] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

[0126] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. An intelligent traffic management method based on multi-scenario recognition and adaptive guidance, characterized in that: include: Collect accident video information, accident type and accident section environmental assessment value; Accident types classify accidents according to their severity, size and cause; Different numbers are set to correspond to the severity, scale and cause of the accident; a group of numbers for one type of accident is called an accident type number sequence; a corresponding diversion plan is preset for each type of accident; Based on the accident video information, the yolov5 model is used to extract accident feature information; Accident characteristic information includes: the number of accident vehicles and the number of accident personnel; The first machine learning model that can estimate the damage assessment value of the accident vehicle and the casualties assessment value of the accident personnel in real time based on the accident video information training; A second machine learning model for real-time identification of accident types is trained based on accident video information, the number of accident vehicles, the number of accident personnel, the accident road section environmental assessment value, the accident type number sequence, the accident vehicle damage assessment value and the accident personnel casualty assessment value predicted by the first machine learning model; Generate diversion instructions based on the accident type number sequence identified by the second machine learning model.

2. According to claim 1, the intelligent traffic management method based on multi-scene recognition and adaptive guidance is characterized in that: The accident video information is the on-site video clip after the accident occurs; The severity of the accident refers to the personal injury and property loss caused by the accident; the scale refers to the number of vehicles involved in the accident and the process of the accident; the cause of the accident refers to the direct cause of the accident; The damage assessment value of the accident vehicle is the overall damage of the vehicle as shown in the accident monitoring screen; The casualty assessment value of the accident is the casualty assessment status of the personnel appearing on the accident monitoring screen; The environmental assessment value of the accident section is the environmental risk level of the accident site; the environmental assessment value of the accident section is a rational number between 0 and 1.

3. The intelligent traffic management method based on multi-scenario recognition and adaptive guidance according to claim 2 is characterized in that: The yolov5 model acquisition process includes: Extract frames from the accident video information to obtain key frames per second; denoise the key frames to enhance image quality; input the denoised key frames into the yolov5 model for detection and analysis; output the vehicle type in the frame and mark its position bounding box; adopt the target tracking algorithm based on the vehicle's position bounding box to delete the duplicate counts of the same vehicle; after deleting the duplicate counts, count the total number of different vehicle types; The target tracking algorithm is DeepSORT; the process of DeepSORT target tracking algorithm includes: Extract features; extract deep features for each detected target; Trajectory management: create or update a trajectory for each target, assign a unique ID, and predict the trajectory state using a Kalman filter; Matching trajectories: Matching detection results with existing trajectories by calculating the Mahalanobis distance and feature similarity of the same target in consecutive frames; Update the target; delete the tracks that have not been matched successfully for a long time and update the target track at all times.

4. The intelligent traffic management method based on multi-scenario recognition and adaptive guidance according to claim 3 is characterized in that: The training process of the first machine learning model includes: The accident video information is converted into a feature vector, the damage assessment value of the accident vehicle and the casualty assessment value of the accident personnel are used as labels corresponding to the feature vector, each set of feature vectors and the labels corresponding to each set of feature vectors are constructed as a sample, and multiple samples are collected to construct a data set; the data set is divided into a training set, a validation set and a test set, wherein the training set accounts for 60% of the data set, and the validation set and the test set each account for 20% of the data set; The damage degree of each vehicle is manually labeled, and the damage degree is a rational number from 0 to 1. The damage assessment value of the accident vehicle is the sum of the damage degrees of all vehicles in the accident video information; The methods for obtaining the accident casualty assessment value include: Collect key point information of the human body; the key point information of the human body is the three-dimensional position coordinates (x, y, z) of the main parts of the human body; the origin of the three-dimensional coordinate system where the three-dimensional position coordinates are located is the location of the surveillance camera; the x-axis and y-axis are parallel to the ground, and the z-axis is a ray starting from the origin and perpendicular to the ground upward; collect key point information of the human body on the head and hips; According to the formula Calculate the trunk inclination angle θ; where: x H ,y H , z H are the three-dimensional position coordinates of the head; L ,y L , z L is the three-dimensional position coordinate of the hip; the torso inclination angle is compared and analyzed with the preset angle threshold. When the torso inclination angle is less than or equal to the angle threshold, and at this time the z-axis coordinate of the head is H When it is lower than 50cm, it is marked as a critically ill person. Each time a critically ill person is determined, the critically ill assessment value is increased by 1. The critically ill assessment value is initially 0. At the same time, professionals score the injuries of the people according to the images of the accident video information. The injury score is a rational number between 0 and 1. The larger the score, the greater the degree of casualties. The injury scores of all people and the critically ill assessment values ​​are accumulated as the accident casualties assessment value. The training set is used as the input of the first machine learning model, and the first machine learning model uses the real-time accident vehicle damage assessment value and the accident casualty assessment value as the output; the accident vehicle damage assessment value and the accident casualty assessment value corresponding to a real-time set of feature vectors are used as prediction targets, and minimizing the loss function value of the first machine learning model is used as the training target; when the loss function value of the first machine learning model is less than or equal to the preset first target loss value, the training is stopped; The first machine learning model is a deep neural network model; the first machine learning model loss function is a mean square error; The calculation method of the hidden layer H of the deep neural network model includes: H=G[b i +w i s(b i-1 +w i-1 x)]; where H is the output of the hidden layer; G is the activation function; b i is the bias term of the hidden layer; w i is the weight of the hidden layer; s is the activation function; b i-1 is the bias term of the previous layer; w i-1 is the weight of the previous layer; x is the input training set; The deep neural network model consists of 4 fully connected layers. The first 3 hidden layers contain 32, 64 and 32 neurons respectively. One SoftMax classification layer contains 2 neurons. The neurons are connected to each other through weights and activated by the ReLU activation function.

5. The intelligent traffic management method based on multi-scenario recognition and adaptive guidance according to claim 4 is characterized in that: The training process of the second machine learning model includes: The accident video information, the number of accident vehicles, the number of accident persons, the accident road section environmental assessment value, the accident vehicle damage assessment value and the accident person casualty assessment value are converted into a set of feature vectors as the input of the second machine learning model, and the second machine learning model uses the real-time accident type number sequence as the output; the accident type number sequence corresponding to a real-time set of accident video information, the number of accident vehicles, the number of accident persons, the accident road section environmental assessment value, the accident vehicle damage assessment value and the accident person casualty assessment value is used as the prediction target, and minimizing the second machine learning model loss function value is used as the training target; when the second machine learning model loss function value is less than or equal to the preset second target loss value, the training is stopped; The second machine learning model loss function is a mean square error; the mean square error is calculated by Minimize to train the model; in the loss function, MSE is the loss function value, i is the feature vector group number; u is the number of feature vector groups; y i is the accident type number sequence corresponding to the i-th group of feature vectors, The accident type number sequence predicted by the i-th group of feature vectors.

6. The intelligent traffic management method based on multi-scenario recognition and adaptive guidance according to claim 5 is characterized in that: The diversion instruction includes executing any of the diversion plans mentioned above; the diversion instruction also includes: setting a drone to go to the scene of the accident; the drone carries a terminal screen to display the accident type corresponding to the accident type number sequence; and gives instructions on whether the two parties to the accident accept the intelligent mediation; when the two parties to the accident accept the intelligent mediation instruction, the drone terminal determines the responsibility of the accident vehicle according to the type of accident; and confirms again whether the two parties to the accident receive the responsibility determination instruction; when the two parties to the accident accept the responsibility determination instruction, the electronic accident report is completed and returned to the traffic management department; when any party to the accident issues a questioning instruction regarding the intelligent mediation instruction or the responsibility determination instruction; the questioning instruction is returned to the traffic management department for manual processing.

7. The intelligent traffic management method based on multi-scenario recognition and adaptive guidance according to claim 6 is characterized in that: The traffic management department has full control over the diversion instructions and can control the drone terminal in real time.

8. An intelligent traffic management system based on multi-scenario recognition and adaptive guidance, characterized in that: The data collection module collects accident video information, accident types and accident section environmental assessment values; accident types are classified according to the severity, scale and cause of the accident; The pre-processing module sets different numbers to correspond to the severity, scale and cause of the accident; a group of numbers for one type of accident is called an accident type number sequence; and a corresponding diversion plan is preset for each type of accident; Feature extraction module, which uses yolov5 model to extract accident feature information based on accident video information; Accident characteristic information includes: the number of accident vehicles and the number of accident personnel; A first model training module, which trains a first machine learning model for real-time estimation of the damage assessment value of the accident vehicle and the casualty assessment value of the accident personnel based on the accident video information; The second model training module trains a second machine learning model for real-time identification of accident types based on accident video information, the number of accident vehicles, the number of accident personnel, the accident road section environmental assessment value, the accident type number sequence, the accident vehicle damage assessment value and the accident personnel casualty assessment value predicted by the first machine learning model; The decision module generates diversion instructions based on the accident type number sequence identified by the second machine learning model.

9. An electronic device, characterized in that: include: A processor and a memory, wherein the memory stores a computer program that can be called by the processor; The processor executes the intelligent traffic management method based on multi-scene recognition and adaptive traffic control as described in any one of claims 1 to 7 by calling the computer program stored in the memory.

10. A computer-readable storage medium, characterized in that: Instructions are stored, and when the instructions are executed on a computer, the computer executes an intelligent traffic management method based on multi-scene recognition and adaptive guidance as described in any one of claims 1 to 7.

Images