An AI agent-based intelligent traffic law enforcement collaboration method, system and medium

Abstract

The application provides a kind of based on AI Agent's wisdom traffic law enforcement cooperation method, system and medium, according to multi-source detection task information, through multimodal fusion model output illegal coefficient, combined with historical illegal records, obtain law enforcement task list;Based on dynamic radius algorithm, according to multi-source road condition information, get the radius of control circle, combined with law enforcement task list, generate dispatch list, for generating and pushing optimal path;Based on the image of law enforcement instrument, identify risk information and trigger risk warning;Combined with law enforcement instrument voice, judge opening speech and defense volume, trigger law enforcement early warning;Through AI Agent, analyze multi-source data, perform resource elastic scheduling, solve the problem of information silos, response lag, process flaws in traditional traffic law enforcement, realize from passive disposal to active management.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of traffic law enforcement, and more particularly, to a collaborative method, system and medium for intelligent traffic law enforcement based on AI Agent. BACKGROUND

[0002] The existing inspection control system, electronic monitoring and reporting platform operate independently, and illegal data is scattered in different processing units. The handling of illegal behaviors such as fake license plates requires manual comparison of multiple databases, which is time-consuming and has the risk of target vehicles escaping the control area. There is a lack of unified priority calculation model for multi-source tasks, and major escape cases are mixed with ordinary illegal parking, delaying the golden interception opportunity. Traditional law enforcement uses fixed radius control, such as a 15-minute driving circle, without considering real-time road network conditions. During peak congestion, law enforcement units cannot arrive on time, and during low-peak periods, police forces are idle. In addition, existing recorders only have basic recording functions and cannot identify dangerous goods or fugitives in real time.

[0003] Therefore, there is an urgent need for an intelligent traffic law enforcement technology based on AI Agent to realize passive disposal to active management. SUMMARY

[0004] In view of the above problems, the purpose of the present application is to provide a collaborative method, system and medium for intelligent traffic law enforcement based on AI Agent, which realizes intelligent triggering and dynamic decision-making of multi-source tasks, real-time calculation of illegal coefficients, generation of priority task lists combined with historical records, elastic control and precise scheduling, dynamic calculation of defense radius, matching of optimal law enforcement units, real-time verification of law enforcement compliance and evidence effectiveness through full-process compliance control and blockchain storage, and solves the problems of information silos, response lag and program flaws in traditional traffic law enforcement, realizing passive disposal to active management.

[0005] The first aspect of the present application provides a collaborative method for intelligent traffic law enforcement based on AI Agent, which comprises:

[0006] Obtaining multi-source detection task information, and outputting an illegal coefficient according to a preset multi-modal fusion model;

[0007] According to the illegal coefficient, combining historical illegal records, and based on a preset weighted fusion algorithm, a task list is obtained;

[0008] Obtaining multi-source road condition information, and obtaining a control circle radius based on a dynamic radius algorithm;

[0009] According to the task list and the control circle radius, a scheduling list is generated;

[0010] According to the scheduling list and the multi-source road condition information, a first path is pushed to a law enforcement unit;

[0011] An image of the law enforcement instrument is acquired, and a risk label is output and a risk warning is triggered according to a preset illegal record database and pattern recognition;

[0012] A voice of the law enforcement instrument is acquired, and an opening voice and a defense volume are recognized according to a preset semantic recognition model in combination with the image of the law enforcement instrument;

[0013] If the opening voice is not recognized or the defense volume is insufficient, a law enforcement warning is triggered.

[0014] In the scheme, the multi-source detection task information is acquired, and an illegal coefficient is output according to a preset multi-modal fusion model, specifically:

[0015] A real-time video stream of the first electronic system is acquired, and vehicle state information is recognized through a convolutional neural network;

[0016] An abnormal control signal transmitted by the vehicle-mounted unit and a public reporting information database are acquired;

[0017] According to the vehicle state information, the abnormal control signal and the reporting information database, in combination with data of a control system, after matching characteristic information of a vehicle violating laws and regulations, multi-source detection task information is formed;

[0018] The multi-source detection task information is input into the multi-modal fusion model, and an illegal coefficient is output.

[0019] In the scheme, the multi-source road condition information is acquired, and a control circle radius is obtained based on a dynamic radius algorithm, specifically:

[0020] A GPS trajectory of a target vehicle is acquired, and an average moving speed thereof is obtained;

[0021] A road network congestion index is obtained based on a traffic road condition platform interface;

[0022] Visibility data is obtained based on a meteorological platform interface;

[0023] The average moving speed, the road network congestion index and the visibility data are combined to obtain multi-source road condition information;

[0024] Based on a preset dynamic radius algorithm, a control circle radius is obtained according to the multi-source road condition information.

[0025] In the scheme, the scheduling list is generated according to the task list and the control circle radius, specifically:

[0026] Based on the control circle radius, a law enforcement unit within a preset radius threshold is acquired;

[0027] Obtaining a law enforcement resource image of the law enforcement unit, including a real-time position, equipment configuration and skill authentication state of the law enforcement unit;

[0028] Based on a preset law enforcement matching model, a distance score, an equipment score and a skill score are obtained according to the task list and the law enforcement resource image;

[0029] Based on a preset weight coefficient, a law enforcement scheduling score is obtained according to the distance score, the equipment score and the skill score;

[0030] The scheduling list is obtained by arranging the law enforcement scheduling scores from high to low.

[0031] In the scheme, the law enforcement instrument image is obtained, and a risk label is output and a risk warning is triggered according to a preset illegal record database and pattern recognition, specifically including:

[0032] According to the law enforcement instrument image, a public security personnel database is matched, and the face features of the person to be investigated are compared;

[0033] If the in-situ personnel library is hit, the in-situ risk label is marked, and the in-situ warning is triggered;

[0034] According to the law enforcement instrument image, a preset dangerous goods feature library is matched to obtain a dangerous goods category and a confidence degree;

[0035] If the confidence degree exceeds a preset confidence threshold, the weapon risk label is marked, and the weapon warning is triggered.

[0036] In the scheme, the law enforcement instrument voice is obtained, and the opening speech and the defense volume are identified according to a preset semantic recognition model in combination with the law enforcement instrument image, specifically including:

[0037] According to the law enforcement instrument voice, law enforcement personnel voice and party voice are obtained based on a preset voiceprint recognition separation;

[0038] According to the law enforcement personnel voice, the opening speech content is matched according to a preset semantic recognition model;

[0039] If the preset opening speech content is not identified, an opening speech prompt is generated;

[0040] According to the party voice, a voice decibel value change curve is obtained;

[0041] If the voice decibel value is lower than a preset loudness threshold, a standby voice collection channel is activated.

[0042] The second aspect of the application provides a collaborative system for intelligent traffic law enforcement based on an AI agent, comprising a collaborative method program for intelligent traffic law enforcement based on an AI agent, which realizes the following steps when executed by the processor:

[0043] Obtain multi-source detection task information, and output a law violation coefficient according to a preset multi-modal fusion model;

[0044] According to the law violation coefficient, combine historical law violation records, and obtain a task list based on a preset weighted fusion algorithm;

[0045] Obtain multi-source road condition information, and obtain a control circle radius based on a dynamic radius algorithm;

[0046] Generate a dispatch list according to the task list and the control circle radius;

[0047] According to the dispatch list and the multi-source road condition information, push a first path to a law enforcement unit;

[0048] Obtain law enforcement instrument images, output a risk label and trigger a risk warning according to a preset law violation record database and pattern recognition;

[0049] Obtain law enforcement instrument voice, combine the law enforcement instrument images, and identify an opening speech and a defense volume according to a preset semantic recognition model;

[0050] If the opening speech is not identified or the defense volume is insufficient, trigger a law enforcement warning.

[0051] In the scheme, the multi-source detection task information is obtained, and a law violation coefficient is output according to a preset multi-modal fusion model, specifically as follows:

[0052] Obtain real-time video streams of a first electronic system, and identify vehicle state information through a convolutional neural network;

[0053] Obtain abnormal control signals transmitted by a vehicle-mounted unit and a public reporting information database;

[0054] According to the vehicle state information, the abnormal control signals, and the reporting information database, combine control system data, match illegal vehicle feature information, and then combine to form multi-source detection task information;

[0055] Input the multi-source detection task information into the multi-modal fusion model, and output a law violation coefficient.

[0056] In the scheme, the multi-source road condition information is obtained, and a control circle radius is obtained based on a dynamic radius algorithm, specifically as follows:

[0057] Obtain the target vehicle GPS track to obtain its average moving speed;

[0058] Obtain the road network congestion index based on the traffic road condition platform interface;

[0059] Obtain the visibility data based on the meteorological platform interface;

[0060] Combine the average moving speed, the road network congestion index and the visibility data to obtain multi-source road condition information;

[0061] Obtain the control circle radius based on the preset dynamic radius algorithm according to the multi-source road condition information.

[0062] The third aspect of the present application provides a computer readable storage medium, the computer readable storage medium comprises an AI Agent-based intelligent traffic law enforcement collaboration method program, when the AI Agent-based intelligent traffic law enforcement collaboration method program is executed by a processor, the steps of the AI Agent-based intelligent traffic law enforcement collaboration method are realized.

[0063] The present application provides an AI Agent-based intelligent traffic law enforcement collaboration method, system and medium, according to multi-source detection task information, a multi-modal fusion model is used to output a violation coefficient, historical violation records are combined to obtain a law enforcement task list; based on a dynamic radius algorithm, a control circle radius is obtained according to multi-source road condition information, a dispatch list is generated by combining the law enforcement task list, which is used to generate and push an optimal path; based on the law enforcement instrument image, risk information is identified and a risk warning is triggered; the opening speech and the defense volume are judged by combining the law enforcement instrument voice, and a law enforcement early warning is triggered; through the AI Agent, multi-source data is analyzed, resource elastic scheduling is performed, the problems of information silos, response lag and program defects in traditional traffic law enforcement are solved, and passive disposal is changed to active management. BRIEF DESCRIPTION OF DRAWINGS

[0064] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments, and it should be understood that the following drawings only show some embodiments of the present application, and therefore should not be regarded as a limitation on the scope.

[0065] Figure 1 A flow chart of an AI Agent-based intelligent traffic law enforcement collaboration method is shown;

[0066] Figure 2 A running flow chart of a multi-modal fusion model is shown;

[0067] Figure 3A flow chart of a dynamic calculation process of a control circle radius is shown.

[0068] Figure 4 A block diagram of a collaborative system of intelligent traffic law enforcement based on an AI Agent is shown. DETAILED DESCRIPTION

[0069] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0070] Unless otherwise defined, all terms (including technical and scientific terms) used in the embodiments of the present application have the same meanings as commonly understood by one of ordinary skill in the art to which the present application belongs. It should also be understood that the terms, such as those defined in a generally used dictionary, should be interpreted as having a meaning consistent with their meaning in the context of the relevant art, and should not be interpreted in an idealized or overly formal sense, unless expressly so defined in the embodiments of the present application.

[0071] The terms "first", "second", and similar terms used in the embodiments of the present application do not denote any order, quantity, or importance, but are only used to distinguish different constituent parts. The terms "one", "a", or "the" and similar terms do not denote a quantity restriction, but denote the existence of at least one. Similarly, the terms "include" or "contain" and similar terms mean that the elements or objects before the terms encompass the elements or objects listed after the terms and their equivalents, and do not exclude other elements or objects. The terms "connect" or "connected" and similar terms are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. The steps before or after the methods in the embodiments of the present application do not necessarily proceed in an accurate order. On the contrary, various steps can be processed in reverse order or simultaneously. Meanwhile, other operations can be added to these processes, or a step or several steps can be removed from these processes.

[0072] In addition, the functional modules in each of the embodiments of the present application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0073] Figure 1 A flow chart of a collaborative method of intelligent traffic law enforcement based on an AI Agent is shown.

[0074] As Figure 1As shown, the first aspect of the present application discloses a collaborative method for intelligent traffic law enforcement based on AI Agent, which comprises:

[0075] S102, acquiring multi-source detection task information, and outputting a law violation coefficient according to a preset multi-modal fusion model;

[0076] S104, obtaining a task list based on a preset weighted fusion algorithm according to the law violation coefficient and historical law violation records;

[0077] S106, acquiring multi-source road condition information, and obtaining a control circle radius based on a dynamic radius algorithm;

[0078] S108, generating a dispatch list according to the task list and the control circle radius;

[0079] S110, pushing a first path to a law enforcement unit according to the dispatch list and the multi-source road condition information;

[0080] S112, acquiring law enforcement instrument images, and outputting a risk label and triggering a risk warning according to a preset law violation record database and pattern recognition;

[0081] S114, acquiring law enforcement instrument voice, and identifying opening voice and defense volume according to a preset semantic recognition model in combination with the law enforcement instrument images;

[0082] S116, triggering a law enforcement warning if the opening voice is not identified or the defense volume is insufficient.

[0083] It should be noted that the multi-source detection task information includes electronic monitoring video, vehicle control unit and mass reporting platform data; and the multi-source road condition information includes road congestion index, weather visibility and target vehicle speed. In this embodiment, first, the real-time video stream of the electronic monitoring camera, the vehicle control signal transmitted by the vehicle-mounted unit, the voice and text information of the mass reporting platform are accessed, and are analyzed through a multi-modal fusion engine; wherein the vehicle line pressing, reverse driving and other behaviors in the video are analyzed through a convolutional neural network, the license plate features in the reporting information are extracted through a speech recognition engine, and the high-risk vehicle information in the control library is compared through a data correlation module. Based on the real-time risk coefficient of illegal behavior and the weighted historical illegal times, a task list with at least three priority levels is generated, including priority, target vehicle and other information. The target vehicle GPS is synchronously acquired to convert speed, road congestion index and meteorological visibility data, and is substituted into a dynamic radius algorithm to output an elastic control range, for example, the control radius is set to 20 kilometers within the driving distance for a smooth road, and is contracted to 5 kilometers within the range in heavy rain and congestion. According to the task priority and the control ring, the law enforcement unit is screened, the position information, equipment type and skill authentication of the law enforcement unit are called, and a dispatch ranking is generated according to the set weight. After the law enforcement unit is selected, the optimal path is generated and pushed to the law enforcement unit based on the position of the law enforcement unit and the position of the target vehicle, and the dynamic road condition. Real-time recognition of the face and vehicle features in the law enforcement instrument picture, if matching the fugitive personnel library, mark the fugitive personnel icon, if detecting the weapon outline such as knives, trigger the weapon risk warning. Synchronously analyze the law enforcement voice stream, if the key words such as "show the certificate" are not recognized within 10 seconds or the volume of the party concerned is continuously lower than 30 decibels, automatically broadcast the compliance reminder and activate the standby recording device. The whole chain automation collaboration of illegal identification, task classification, resource scheduling and on-site disposal is realized.

[0084] Figure 2 A running flowchart of a multi-modal fusion model provided by an embodiment of the application is shown.

[0085] According to an embodiment of the application, as Figure 2 shown, the multi-source detection task information is acquired, and a violation coefficient is output according to a preset multi-modal fusion model, specifically:

[0086] S202, acquiring a real-time video stream of a first electronic system, and identifying vehicle state information through a convolutional neural network;

[0087] S204, acquiring an abnormal control signal transmitted by a vehicle-mounted unit, and a mass reporting information database;

[0088] S206, after matching the illegal vehicle feature information according to the vehicle state information, the abnormal control signal and the reporting information database, combining the control system data, the multi-source detection task information is formed by combining;

[0089] S208, input the multi-source detection task information into the multi-modal fusion model, and output a law violation coefficient.

[0090] It should be noted that the first electronic system is an electronic monitoring system, the abnormal control signal is a vehicle abnormal control signal, including rapid acceleration, rapid deceleration, rapid turning, etc., and the control system data is a vehicle violation feature vector. The road surface video stream is captured by a high-definition electronic monitoring camera, the violation situation is identified by a preset vehicle identification model, and the vehicle is identified by model, color, etc. The on-board unit OBU collects abnormal control signals through the CAN bus and sends them to the background. Based on the public reporting information, the text is converted and the key information is extracted, and the video identification result is cross-verified. The above data stream is composed of multi-source detection task information, which is input into the multi-modal fusion model to obtain the law violation coefficient corresponding to the illegal behavior. The output structured task information includes the illegal type, the illegal confidence, the vehicle information and the moving direction, etc. The coordination obstacles of multi-source data are broken through, and the missed detection or wrong detection problems of a single data system are avoided.

[0091] Figure 3 A dynamic calculation flowchart of a control circle radius is shown.

[0092] According to the embodiment of the present application, as shown in Figure 3 The multi-source road condition information is obtained, and the control circle radius is obtained based on a dynamic radius algorithm, specifically:

[0093] S302, obtaining the GPS trajectory of the target vehicle to obtain the average moving speed thereof;

[0094] S304, obtaining the road network congestion index based on the traffic road condition platform interface;

[0095] S306, obtaining the visibility data based on the meteorological platform interface;

[0096] S308, combining the average moving speed, the road network congestion index and the visibility data to obtain multi-source road condition information;

[0097] S310, obtaining the control circle radius based on the preset dynamic radius algorithm according to the multi-source road condition information.

[0098] It should be noted that the embodiment provides a logic of dynamically calculating the radius of the control circle. First, the target vehicle GPS track point is analyzed, the average moving speed is calculated after removing the signal drift point. The traffic condition platform interface is called, for example, the map database API is used to obtain real-time traffic data, and the congestion index within the set radius around the target vehicle is extracted. The interface data of the meteorological department is accessed, and the visibility data is read. The average moving speed, road network congestion index and visibility data are integrated to obtain multi-source traffic information. The above multi-source traffic information is input into the dynamic radius algorithm to obtain the reference radius and the floating radius, which are used to confirm the radius of the control circle of the target vehicle. The embodiment dynamically adjusts the radius of the control circle to avoid tracking failure or resource waste caused by fixed radius.

[0099] According to the embodiment of the present application, the dispatch list is generated according to the task list and the control circle radius, specifically:

[0100] Based on the control circle radius, the law enforcement unit within the preset radius threshold is obtained;

[0101] The law enforcement resource portrait of the law enforcement unit is obtained, including the real-time position, equipment type and skill authentication state of the law enforcement unit;

[0102] Based on the preset law enforcement matching model, the distance score, equipment score and skill score are obtained according to the task list and the law enforcement resource portrait;

[0103] Based on the preset weight coefficient, the law enforcement dispatch score is obtained according to the distance score, the equipment score and the skill score;

[0104] The dispatch list is obtained by arranging the law enforcement dispatch score from high to low.

[0105] It should be noted that in the embodiment, all online law enforcement units are screened in the dynamic control circle geographic fence. The law enforcement portrait of the law enforcement unit is called, which includes: the GPS position of the law enforcement unit for calculating the distance from the target point, the equipment state for adapting to illegal verification of illegal behavior, and the skill label for adapting to case handling requirements. Then, the law enforcement dispatch score is calculated according to the preset weight value, and the dispatch list is generated in descending order. The list is synchronously pushed to the mobile terminal of the execution unit, and the expected arrival time and the optimal travel path of each execution unit are displayed.

[0106] According to the embodiment of the present application, the law enforcement instrument image is obtained, the risk label is output and the risk warning is triggered according to the preset illegal record database and pattern recognition, specifically including:

[0107] According to the law enforcement instrument image, the public security personnel database is matched, and the face features of the person to be investigated are compared;

[0108] If the hit in the fugitive personnel library, mark the fugitive risk label, and trigger the fugitive warning;

[0109] According to the image of the law enforcement instrument, match the preset dangerous goods feature library to obtain the dangerous goods category and its confidence;

[0110] If the confidence exceeds the preset confidence threshold, mark the weapon risk label, and trigger the weapon warning.

[0111] It should be noted that the embodiment provides a high-risk task identification and early warning mechanism. In the embodiment, after capturing the face picture by the law enforcement recorder, the feature vector is extracted by the lightweight model and sent to the background; based on the real-time comparison result of the background and the fugitive personnel library, it is judged whether the person is a fugitive. If so, through the icon, voice or vibration warning, the warning is provided to the law enforcement unit. Synchronously scan the article contour, when detecting long strip metal article, etc., the matching mechanism of the dangerous goods feature library is started. When the dangerous goods confidence exceeds the set threshold, the icon, voice or vibration warning is provided to the law enforcement unit. The embodiment improves the judgment accuracy of the law enforcement unit to the dangerous situation on the scene through visual warning.

[0112] According to the embodiment of the application, the law enforcement instrument voice is acquired, combined with the law enforcement instrument image, and according to a preset semantic recognition model, the opening speech and the defense volume are recognized, specifically including:

[0113] According to the law enforcement instrument voice, the law enforcement personnel voice and the party voice are separated based on a preset voiceprint recognition;

[0114] According to the law enforcement personnel voice, the opening speech content is matched according to a preset semantic recognition model;

[0115] If the preset opening speech content is not recognized, the opening speech prompt is generated;

[0116] According to the party voice, the voice decibel value change curve is obtained;

[0117] If the voice decibel value is lower than the preset loudness threshold, the standby voice collection channel is activated.

[0118] It should be noted that the embodiment provides a law enforcement compliance monitoring. Based on the preset voiceprint matching model, the law enforcement personnel sound source and the party sound source are separated. Through the semantic recognition model, the law enforcement personnel voice content and the party voice content are recognized. If it is detected that the law enforcement personnel does not tell the opening speech content, such as the keywords of showing the certificate, checking, etc., the opening speech prompt is played to the law enforcement personnel earphone device. When it is detected that the party voice volume is insufficient, the standby microphone is activated to enhance the collection. The embodiment ensures the program justice zero omission, and solves the evidence collection problem caused by the incoordination of the party.

[0119] It is worth mentioning that, further comprising:

[0120] According to the time axis, the image and the voice of the law enforcement instrument are integrated, the GPS positioning, the device ID and the timestamp of the law enforcement instrument are combined, and an electronic evidence package is generated;

[0121] According to the preset blockchain contract generation algorithm, the encrypted evidence package is generated according to the electronic evidence package;

[0122] According to the preset distributed storage, the encrypted evidence package is stored to the first blockchain node.

[0123] It should be noted that the present embodiment provides an evidence preservation mechanism based on blockchain. In the present embodiment, multi-source evidence is integrated according to the time axis, including law enforcement instrument image, law enforcement instrument voice, law enforcement instrument GPS positioning and law enforcement instrument ID, and an electronic evidence package is obtained by combining the timestamp. Then, through the preset blockchain contract generation algorithm, the electronic evidence package is encrypted into an encrypted evidence package. Finally, through the distributed storage mode, the encrypted evidence package is stored in the multiple blockchain nodes set by the relevant departments, so as to ensure that the evidence is traceable and tamper-proof throughout the process.

[0124] It is worth mentioning that, further comprising:

[0125] Obtain historical task disposal data; based on a preset law enforcement energy efficiency analysis mode, obtain a high-law-violation area and a patrol law enforcement frequency, and generate a law enforcement heat difference map;

[0126] According to the law enforcement heat difference map, the skill authentication state of the law enforcement unit is matched to obtain a skill gap rate;

[0127] If the skill gap rate exceeds a preset gap rate threshold, a skill training course is pushed to the patrol law enforcement unit of the high-law-violation area;

[0128] Based on the skill training course learning situation, the law enforcement resource portrait of the law enforcement unit is updated.

[0129] It should be noted that, as an implementation manner, all task data within 30 days is collected, a high-law-violation heat map and a patrol frequency map are generated by kernel density estimation algorithm; the heat difference is calculated by superposition to generate a law enforcement heat difference map. The skill library of law enforcement personnel is associated, the skill deficiency rate of law enforcement personnel in the area where the heat difference exceeds the set threshold is calculated; when the skill deficiency rate exceeds the set threshold, the corresponding skill training course is pushed to the patrol law enforcement unit of the area. After completing the training, the system tracks the subsequent task timeliness, and if the timeliness improves to the required level, the skill portrait is updated, and the English matching score is increased in the dispatch weight. Further, the precise management driven by data is realized.

[0130] It is worth mentioning that, further comprising:

[0131] According to the historical illegal record, the real-time traffic flow information and the meteorological warning information, a pre-trained illegal prediction model is used to obtain a probability space distribution diagram of illegal behavior in a unit window time;

[0132] According to the illegal probability, a high-risk area is determined;

[0133] According to the high-risk area, the patrol law enforcement route of the law enforcement unit is adjusted, and the first electronic system is linked to improve the snapshot frequency.

[0134] It should be noted that the present embodiment provides a predictive defense strategy. As an implementation manner, historical illegal data in the past 5 days, real-time traffic density and meteorological warning are input, a pre-trained illegal prediction model is used to output the illegal probability in the next 3 hours with a grid unit of 1 square kilometer. For the high-risk area, the patrol unit is deployed in advance, and the snapshot frequency of the electronic monitoring system is improved to realize the early defense. Finally, all the prediction records are compared with the actual illegal data to correct the model weight.

[0135] Figure 4 A block diagram of an AI Agent-based intelligent traffic law enforcement collaborative system is shown.

[0136] As shown in Figure 4 The second aspect of the present application discloses an AI Agent-based intelligent traffic law enforcement collaborative system 4, which comprises a memory 41 and a processor 42, the memory comprising an AI Agent-based intelligent traffic law enforcement collaborative method program, and the AI Agent-based intelligent traffic law enforcement collaborative method program is executed by the processor to realize the following steps:

[0137] Obtain multi-source detection task information, and output an illegal coefficient according to a pre-set multi-modal fusion model;

[0138] According to the illegal coefficient, combine historical illegal records, and obtain a task list based on a pre-set weighted fusion algorithm;

[0139] Obtain multi-source road condition information, and obtain a control circle radius based on a dynamic radius algorithm;

[0140] Generate a dispatch list according to the task list and the control circle radius;

[0141] According to the dispatch list and the multi-source road condition information, push a first path to a law enforcement unit;

[0142] Obtain law enforcement instrument images, output a risk label and trigger a risk warning according to a pre-set illegal record database and pattern recognition;

[0143] Acquire the voice of the law enforcement instrument, combine the image of the law enforcement instrument, identify the opening speech and the defense volume according to a preset semantic recognition model;

[0144] If the opening speech or the defense volume is not identified, a law enforcement warning is triggered.

[0145] It should be noted that the multi-source detection task information includes electronic monitoring video, vehicle control unit and crowd reporting platform data; and the multi-source road condition information includes road congestion index, weather visibility and target vehicle speed. In this embodiment, first, the real-time video stream of the electronic monitoring camera, the vehicle control signal transmitted by the vehicle-mounted unit, the voice and text information of the crowd reporting platform are accessed, and are analyzed by a multi-modal fusion engine; wherein the vehicle line pressing, reverse driving and other behaviors in the video are analyzed by a convolutional neural network, the vehicle plate features in the reporting information are extracted by a voice recognition engine, and the high-risk vehicle information in the control library is compared by a data correlation module. Based on the real-time risk coefficient of illegal behavior and the weighted historical illegal times, a task list with at least three priority levels is generated, including priority, target vehicle and other information. The target vehicle GPS is synchronously acquired to calculate the speed, road congestion index and meteorological visibility data, and the dynamic radius algorithm is output to output the elastic control range, for example, the control radius is set to 20 kilometers for a smooth road, and is contracted to 5 kilometers in heavy rain and congestion. According to the task priority and the control circle, the law enforcement unit is screened, the position information, equipment type and skill authentication of the law enforcement unit are called, and a dispatch ranking is generated according to the set weight. After the law enforcement unit is selected, the optimal path is generated and pushed to the law enforcement unit based on the position of the law enforcement unit and the position of the target vehicle, and the dynamic road condition. The face and vehicle features in the law enforcement instrument picture are recognized in real time, if they match the wanted personnel library, the wanted personnel icon is marked, if weapons such as knives are detected, a weapon risk warning is triggered. The law enforcement voice stream is synchronously analyzed, if the keywords such as "present your ID card" are not recognized within 10 seconds or the volume of the party concerned is continuously lower than 30 decibels, a compliance reminder is automatically broadcast and a backup recording device is activated. The whole chain automation collaboration of illegal identification, task classification, resource scheduling and on-site disposal is realized.

[0146] According to the embodiment of the present application, the multi-source detection task information is acquired, and a law enforcement coefficient is output according to a preset multi-modal fusion model, specifically:

[0147] Acquire the real-time video stream of the first electronic system, and identify the vehicle state information by a convolutional neural network;

[0148] Acquire the abnormal control signal transmitted by the vehicle-mounted unit and the crowd reporting information database;

[0149] After matching the illegal vehicle feature information according to the vehicle state information, the abnormal control signal and the reporting information database, combined with the data of the control system, the multi-source detection task information is formed.

[0150] input the multi-source detection task information into the multi-modal fusion model, and output a law violation coefficient.

[0151] It should be noted that the first electronic system is an electronic monitoring system; the abnormal control signal is a vehicle abnormal control signal, including sudden acceleration, sudden deceleration, rapid turning, etc.; and the control system data is a vehicle violation feature vector. A high-definition electronic monitoring camera shoots a road video stream, a pre-set vehicle recognition model identifies a violation situation, and a vehicle is then identified in terms of vehicle type, color, etc. An on-board unit OBU collects abnormal control signals through a CAN bus and sends them to a background; based on mass reporting information, key information is extracted and cross-verified with video recognition results. The above data streams are combined into multi-source detection task information, which is input into a multi-modal fusion model to obtain a law violation coefficient corresponding to a law violation behavior. Structured task information is output, including law violation type, law violation confidence, vehicle information, and moving direction, etc. The coordination obstacles of multi-source data are broken down, and the missed detection or false detection problems of a single data system are avoided.

[0152] According to the embodiment of the present application, the multi-source road condition information is obtained based on a dynamic radius algorithm to obtain a control circle radius, specifically:

[0153] The GPS trajectory of the target vehicle is obtained to obtain an average moving speed thereof;

[0154] The road network congestion index is obtained based on a traffic road condition platform interface;

[0155] The visibility data is obtained based on a meteorological platform interface;

[0156] The average moving speed, the road network congestion index, and the visibility data are combined to obtain multi-source road condition information;

[0157] The control circle radius is obtained based on a pre-set dynamic radius algorithm according to the multi-source road condition information.

[0158] It should be noted that the embodiment provides a logic for dynamically calculating a control circle radius. First, the GPS trajectory points of the target vehicle are analyzed, and the average moving speed is calculated after removing signal drift points. The real-time road condition data is obtained by calling a traffic road condition platform interface, such as a map database API, and the congestion index within a set radius around the target vehicle is extracted. The interface data of the meteorological department is accessed to read the visibility data. The multi-source road condition information is integrated based on the average moving speed, the road network congestion index, and the visibility data. The above multi-source road condition information is input into a dynamic radius algorithm to obtain a reference radius and a floating radius, which are used to confirm the control circle radius of the target vehicle. The embodiment dynamically adjusts the control circle radius to avoid tracking failure or resource waste caused by a fixed radius.

[0159] According to the embodiment of the present application, the scheduling list is generated according to the task list and the control circle radius, specifically:

[0160] Based on the control circle radius, the law enforcement unit within the preset radius threshold is obtained;

[0161] The law enforcement resource portrait of the law enforcement unit is obtained, including the real-time position, equipment configuration and skill authentication status of the law enforcement unit;

[0162] Based on the preset law enforcement matching model, the distance score, equipment score and skill score are obtained according to the task list and the law enforcement resource portrait;

[0163] Based on the preset weight coefficient, the law enforcement scheduling score is obtained according to the distance score, the equipment score and the skill score;

[0164] The scheduling list is obtained by arranging the law enforcement scheduling score from high to low.

[0165] It should be noted that in the present embodiment, all online law enforcement units are screened within the dynamic control circle geographic fence. The law enforcement portrait of the law enforcement unit is called, which includes: the GPS position of the law enforcement unit for calculating the distance from the target point, the equipment state for adapting to illegal verification of illegal behavior, and the skill label for adapting to the case handling requirement. Then, the law enforcement scheduling score is calculated according to the preset weight value, and the scheduling list is generated in descending order. The list is synchronously pushed to the mobile terminal of the execution unit, and the expected arrival time and the optimal travel path of each execution unit are displayed.

[0166] According to the embodiment of the present application, the law enforcement instrument image is obtained, and the risk label is output and the risk warning is triggered according to the preset illegal record database and pattern recognition, specifically including:

[0167] According to the law enforcement instrument image, the police personnel database is matched, and the face features of the person to be investigated are compared;

[0168] If the wanted personnel library is hit, the wanted risk label is marked, and the wanted warning is triggered;

[0169] According to the law enforcement instrument image, the preset dangerous goods feature library is matched to obtain the dangerous goods category and its confidence;

[0170] If the confidence exceeds the preset confidence threshold, the weapon risk label is marked, and the weapon warning is triggered.

[0171] It should be noted that the embodiment provides a high-risk task identification and early warning mechanism. In the embodiment, after capturing the face picture by the law enforcement recorder, the feature vector is extracted by the lightweight model and sent to the background; based on the real-time comparison result of the background and the fugitive personnel library, it is judged whether the person is a fugitive. If so, the icon, voice or vibration warning is provided to the law enforcement unit. Synchronously scan the outline of the article, and when detecting a long strip of metal article, etc., the matching mechanism of the dangerous goods feature library is started. When the dangerous goods confidence exceeds the set threshold, the icon, voice or vibration warning is provided to the law enforcement unit. The embodiment improves the judgment accuracy of the law enforcement unit to the dangerous situation on the scene through visual warning.

[0172] According to the embodiment of the present application, the law enforcement instrument voice is acquired, combined with the law enforcement instrument image, and the opening speech and defense volume are identified according to a preset semantic recognition model, specifically including:

[0173] According to the law enforcement instrument voice, the law enforcement personnel voice and the party voice are separated based on a preset voiceprint recognition;

[0174] According to the law enforcement personnel voice, the opening speech content is matched according to a preset semantic recognition model;

[0175] If the preset opening speech content is not identified, an opening speech prompt is generated;

[0176] According to the party voice, a voice decibel value change curve is obtained;

[0177] If the voice decibel value is lower than a preset loudness threshold, a backup voice collection channel is activated.

[0178] It should be noted that the embodiment provides a law enforcement compliance monitoring. Based on a preset voiceprint matching model, the law enforcement personnel sound source and the party sound source are separated. The law enforcement personnel voice content and the party voice content are identified by a semantic recognition model. If it is detected that the law enforcement personnel does not tell the opening speech content, such as the keywords of showing the certificate, checking, etc., the opening speech prompt is played to the law enforcement personnel earphone device. When it is detected that the party voice volume is insufficient, the backup microphone is activated to enhance the collection. The embodiment ensures the program justice and solves the evidence collection problem caused by the party's non-cooperation.

[0179] It is worth mentioning that it also includes:

[0180] According to the time axis, the law enforcement instrument image and the law enforcement instrument voice are integrated, combined with the GPS positioning, device ID and timestamp of the law enforcement instrument, to generate an electronic evidence package;

[0181] Based on a preset blockchain contract generation algorithm, an encrypted evidence package is generated according to the electronic evidence package;

[0182] According to the preset distributed storage, the encrypted evidence package is stored to the first blockchain node.

[0183] It should be noted that the present embodiment provides an evidence preservation mechanism based on a blockchain. In the present embodiment, multi-source evidence is integrated according to a time axis, including law enforcement instrument images, law enforcement instrument voices, law enforcement instrument GPS positioning and law enforcement instrument IDs, and an electronic evidence package is obtained by integrating the timestamps. Then, the electronic evidence package is encrypted into an encrypted evidence package through a preset blockchain contract generation algorithm. Finally, the encrypted evidence package is stored in multiple blockchain nodes set by relevant departments in a distributed storage manner, so that the evidence is fully traceable and tamper-proof.

[0184] It is worth mentioning that it also includes:

[0185] Obtain historical task handling data; based on a preset law enforcement energy efficiency analysis mode, obtain a high-law-violation area and a patrol law enforcement frequency, and generate a law enforcement heat difference map;

[0186] According to the law enforcement heat difference map, match the skill authentication state of the law enforcement unit to obtain a skill gap rate;

[0187] If the skill gap rate exceeds a preset gap rate threshold, a skill training course is pushed to the patrol law enforcement unit of the high-law-violation area;

[0188] Based on the skill training course learning situation, update the law enforcement resource portrait of the law enforcement unit.

[0189] It should be noted that as an implementation manner, all task data within 30 days are collected, a law violation heat map and a patrol frequency map are generated through a kernel density estimation algorithm; a law enforcement heat difference map is generated by superimposing and calculating the heat difference; the skill library of law enforcement personnel is associated, the skill deficiency rate of law enforcement personnel in the area where the heat difference exceeds the set threshold is calculated; when the skill deficiency rate exceeds the set threshold, the corresponding skill training course is pushed to the patrol law enforcement unit of the area. After completing the training, the system tracks the timeliness of subsequent tasks, and if the timeliness improves to the required level, the skill portrait is updated, and the English matching score is increased in the dispatch weight. Thus, data-driven precise governance is realized.

[0190] It is worth mentioning that it also includes:

[0191] According to historical law violation records, real-time traffic information and weather warning information, a pre-trained law violation prediction model is used to obtain a law violation probability spatial distribution map within a unit window time;

[0192] According to the law violation probability, a high-risk area is determined;

[0193] According to the high-risk area, the patrol law enforcement route of the law enforcement unit is adjusted, and the first electronic system is linked to improve the snapshot frequency.

[0194] It should be noted that the embodiment provides a predictive deployment strategy. As an implementation, the historical illegal data in the past 5 days, real-time traffic density, and meteorological warning are input, and a pre-trained illegal prediction model is used to output the illegal probability in the next 3 hours in a 1 square kilometer grid unit. For high-risk areas, patrol units are deployed in advance, and the snapshot frequency of the electronic monitoring system is improved to realize early deployment. Finally, all prediction records are compared with the actual illegal data to correct the model weight.

[0195] The third aspect of the present application provides a computer readable storage medium, wherein the computer readable storage medium comprises an AI Agent-based intelligent traffic law enforcement collaboration method program, and the AI Agent-based intelligent traffic law enforcement collaboration method program is executed by a processor to implement the steps of the AI Agent-based intelligent traffic law enforcement collaboration method according to any one of the above.

[0196] In summary, the present application provides an AI Agent-based intelligent traffic law enforcement collaboration method, system and medium, which outputs an illegal coefficient through a multi-modal fusion model according to multi-source detection task information, obtains a law enforcement task list in combination with historical illegal records, obtains a control circle radius according to multi-source road condition information based on a dynamic radius algorithm, generates a scheduling list in combination with the law enforcement task list, generates and pushes an optimal path, identifies risk information based on images of a law enforcement instrument and triggers a risk warning, judges opening speech and defense volume in combination with a law enforcement instrument voice, triggers a law enforcement warning, and analyzes multi-source data through an AI Agent to perform resource elastic scheduling, solve the problems of information silos, response lag and program defects in traditional traffic law enforcement, and realize passive disposal to active governance.

[0197] If the functions are realized in the form of software function modules and sold or used as independent products, they can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application or parts of the present application that essentially contribute to the prior art or parts of the technical solutions can be embodied in the form of software products, which are stored in a storage medium and include instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the methods described in various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various program code storage media.

[0198] The above merely provides the preferred embodiments of the present application, and is not used to limit the present application. For those skilled in the art, the present application can have various modifications and changes. Any modifications, equivalent replacements, improvements, etc. made within the principles and technical scope of the present application shall fall into the scope of the present application.

Claims

1. A method for AI Agent-based intelligent traffic law enforcement collaboration, characterized in that, The method comprises: obtaining multi-source detection task information, and outputting a law violation coefficient according to a preset multi-modal fusion model; obtaining a task list based on a preset weighted fusion algorithm according to the law violation coefficient and historical law violation records; obtaining a control circle radius based on a dynamic radius algorithm according to multi-source road condition information; generating a scheduling list according to the task list and the control circle radius; pushing a first path to a law enforcement unit according to the scheduling list and the multi-source road condition information; obtaining law enforcement instrument images, and outputting a risk label and triggering a risk warning according to a preset law violation record database and pattern recognition; obtaining law enforcement instrument voice, and identifying opening speech and defense volume according to a preset semantic recognition model in combination with the law enforcement instrument images; if the opening speech is not identified or the defense volume is insufficient, triggering a law enforcement warning. 2.The AI Agent-based intelligent traffic law enforcement collaboration method of claim 1, wherein, The method comprises: obtaining real-time video streams of a first electronic system, and identifying vehicle state information through a convolutional neural network; obtaining abnormal control signals transmitted by a vehicle-mounted unit and a public report information database; combining control system data to match illegal vehicle feature information after obtaining the vehicle state information, the abnormal control signals and the report information database, and combining to form multi-source detection task information; inputting the multi-source detection task information into the multi-modal fusion model to output a law violation coefficient.

3. The collaborative method for AI Agent based smart traffic enforcement according to claim 1, wherein, The method comprises: obtaining a target vehicle GPS trajectory to obtain an average moving speed; obtaining a road network congestion index based on a traffic road condition platform interface; obtaining visibility data based on a meteorological platform interface; combining the average moving speed, the road network congestion index and the visibility data to obtain multi-source road condition information; obtaining a control circle radius based on a preset dynamic radius algorithm according to the multi-source road condition information.

4. The collaborative method for AI Agent based smart traffic enforcement according to claim 1, wherein, The method comprises: obtaining law enforcement units within a preset radius threshold based on the control circle radius; obtaining law enforcement resource portraits of the law enforcement units, including real-time positions, equipment types and skill authentication states of the law enforcement units; obtaining distance scores, equipment scores and skill scores based on a preset law enforcement matching model according to the task list and the law enforcement resource portraits; obtaining law enforcement scheduling scores based on preset weight coefficients according to the distance scores, the equipment scores and the skill scores; arranging the law enforcement scheduling scores from high to low to obtain a scheduling list.

5. The collaborative method for AI Agent based smart traffic enforcement according to claim 1, wherein, The method comprises: matching a public security personnel database according to the law enforcement instrument images to compare the facial features of the persons to be investigated; if the persons to be investigated are in the fugitive personnel database, marking them as fugitive risk labels and triggering a fugitive warning; matching a preset dangerous goods feature database according to the law enforcement instrument images to obtain dangerous goods categories and their confidence degrees; if the confidence degrees exceed a preset confidence threshold, marking them as weapon risk labels and triggering a weapon warning.

6. The collaborative method for AI Agent based smart traffic enforcement according to claim 1, wherein, The law enforcement instrument voice is acquired, the law enforcement instrument image is combined, a preset semantic recognition model is used for identifying opening speech and defense volume, and specifically, the method comprises the following steps: According to the law enforcement instrument voice, the voiceprint recognition is separated based on the preset to obtain the law enforcement personnel voice and the party voice; According to the law enforcement personnel voice, the opening speech content is matched according to the preset semantic recognition model; If the preset opening speech content is not identified, the opening speech prompt is generated; According to the party voice, the voice decibel value change curve is obtained; If the voice decibel value is lower than the preset loudness threshold, the standby voice collection channel is activated.

7. An AI Agent-based intelligent traffic law enforcement system, characterized in that, The system includes a memory and a processor, the memory includes an AI Agent-based intelligent traffic law enforcement collaboration method program, and the AI Agent-based intelligent traffic law enforcement collaboration method program is executed by the processor to realize the following steps: Acquire multi-source detection task information, and output a violation coefficient according to a preset multi-modal fusion model; According to the violation coefficient, combined with the historical violation record, a task list is obtained based on a preset weighted fusion algorithm; Obtain multi-source road condition information, and obtain the radius of the control circle based on the dynamic radius algorithm; According to the task list and the radius of the control circle, a scheduling list is generated; According to the scheduling list and the multi-source road condition information, a first path is pushed to the law enforcement unit; Acquire law enforcement instrument image, output risk label and trigger risk warning according to preset violation record database and pattern recognition; Acquire law enforcement instrument voice, combine the law enforcement instrument image, and identify opening speech and defense volume according to a preset semantic recognition model; If the opening speech is not identified or the defense volume is insufficient, the law enforcement warning is triggered.

8. The collaborative system for AI Agent based smart traffic enforcement of claim 7, wherein, The acquisition of multi-source detection task information, according to the preset multi-modal fusion model, outputs the violation coefficient, specifically: Obtain the real-time video stream of the first electronic system, and identify the vehicle state information through the convolutional neural network; Obtain the abnormal control signal transmitted by the vehicle unit and the public reporting information database; According to the vehicle state information, the abnormal control signal and the reporting information database, combined with the data of the control system, the characteristics information of the illegal vehicle is matched to form multi-source detection task information; The multi-source detection task information is input into the multi-modal fusion model to output the violation coefficient.

9. The collaborative system for AI Agent based smart traffic enforcement of claim 7, wherein, The acquisition of multi-source road condition information, based on the dynamic radius algorithm, obtains the radius of the control circle, specifically: Obtain the average moving speed of the target vehicle GPS track; Based on the traffic road condition platform interface, the road network congestion index is obtained; Based on the meteorological platform interface, the visibility data is obtained; The average moving speed, the road network congestion index and the visibility data are combined to obtain multi-source road condition information; Based on the preset dynamic radius algorithm, the multi-source road condition information is obtained to obtain the radius of the control circle.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer readable storage medium comprises an AI Agent-based intelligent traffic law enforcement collaboration method program, and the AI Agent-based intelligent traffic law enforcement collaboration method program is executed by the processor to realize the steps of the AI Agent-based intelligent traffic law enforcement collaboration method in any one of claims 1 to 6.

Images