Road traffic incident identification and early warning method and system

Abstract

The invention relates to the technical field of intelligent traffic, in particular to a highway traffic incident recognition and early warning method and system, and the method comprises the steps: collecting traffic data in real time through sensing equipment arranged along a highway; performing preliminary identification and classification of conventional traffic events on the preprocessed traffic data through edge computing nodes; inputting a preliminary analysis result and unclassified traffic data into a cloud end, and carrying out secondary analysis and semantic understanding; and merging the logic relationship and the events of the same target to obtain a final event identification result. According to the traffic event attribute and the event identification result, generating early warning information; and inputting feedback data after each event processing into the recognition model for self-learning and optimization, and dynamically optimizing model parameters, event classification standards and early warning rules. According to the invention, the problems of false report and missing report in traffic event identification are effectively solved, the identification precision is improved, the repeated report is reduced, and the method can dynamically adapt to the problems of complex scenes and non-uniform standards.

Description

Technical Field

[0001] The present invention relates to the field of intelligent transportation technology, and in particular to a highway traffic incident recognition and early warning method and system. Background Art

[0002] With the expansion of the highway network, traffic volume has increased rapidly, and accidents and congestion problems have become increasingly serious. Although the existing traffic event detection system based on video surveillance and sensors has adopted algorithms such as RCNN and YOLO to make breakthroughs in target recognition, it still faces problems such as false alarms and missed alarms in practical applications, and the definition of algorithm events is inconsistent with that of business events. In addition, the lack of unified event detection standards leads to repeated reporting and frequent false alarms. The system lacks semantic understanding of complex traffic scenarios and is difficult to accurately handle special situations such as accidents and road maintenance.

[0003] The information disclosed in this background technology section is only intended to deepen the understanding of the overall background technology of the present disclosure, and should not be regarded as acknowledging or suggesting in any form that the information constitutes the prior art known to those skilled in the art. Summary of the invention

[0004] The present invention provides a highway traffic incident identification and warning method and system, thereby effectively solving the problems pointed out in the background technology.

[0005] In order to achieve the above object, the technical solution adopted by the present invention is: A highway traffic incident recognition and warning method, comprising: Collecting traffic data in real time through sensing devices installed along the highway, and preprocessing the traffic data; The pre-processed traffic data is passed through the edge computing node to perform preliminary identification and classification of conventional traffic events to obtain preliminary analysis results; Inputting the preliminary analysis results and unclassified traffic data into the cloud, the cloud performs secondary analysis and semantic understanding of the traffic event based on a large-scale pre-trained recognition model, and combines multimodal data sources to obtain secondary analysis results of the event; Confirm the events of the preliminary analysis results and the secondary analysis results, merge the events with logical relationships and the same target, and obtain the final event identification results; Generate warning information according to the attributes of the traffic event and the event recognition result, wherein the warning information includes key features and influencing factors related to the event; The feedback data after each event processing is input into the recognition model for self-learning and optimization, and the model parameters, event classification standards and early warning rules are dynamically optimized.

[0006] Furthermore, the pre-processed traffic data is used by edge computing nodes to perform preliminary identification and classification of conventional traffic events to obtain preliminary analysis results, including: Differentiate between real traffic events and background noise through dynamic algorithms, and automatically correct and compensate for sensor data anomalies based on data interaction with adjacent monitoring points; Based on historical traffic data and traffic trends, edge computing nodes conduct forward-looking identification and correlation reasoning of events, predict potential secondary events, and set priorities based on the severity and potential impact of the events; Through the multi-level event splitting and merging mechanism, similar and related traffic abnormal behaviors are integrated.

[0007] Furthermore, the preliminary analysis results and unclassified traffic data are input into the cloud, and the cloud performs secondary analysis and semantic understanding of the traffic event based on a large-scale pre-trained recognition model, and combines multimodal data sources to obtain secondary analysis results of the event, including: Through contextual semantic analysis of events, combined with historical data and current context information, the background and relevance of traffic events can be understood, and the complexity and potential impact of events can be judged; Combine multimodal data sources to cross-validate and supplement traffic events; For complex traffic incidents, decompose and reconstruct the incidents, identify the key triggering factors and possible secondary events, and conduct reconstruction analysis; Perform deep semantic recognition of abnormal behaviors, and analyze unresolved atypical events and abnormal behaviors in the preliminary recognition results based on large-scale pre-trained models; Based on multimodal data and historical data, the impact scope of traffic events is inferred, and the future development trend and potential traffic impact of the events are predicted.

[0008] Furthermore, event decomposition and reconstruction include: Identify the triggers and causes of incidents and break down the key components of incidents using multimodal data sources and historical traffic data; Based on the cause of the event, predict the secondary events that may be triggered by the event and analyze their potential impact in advance; Based on the decomposition results of the events, the time sequence of the events is reconstructed to reveal the causal relationship and development process of the events; Based on the traffic flow model, the impact of the incident on the current road section and surrounding areas is evaluated, and the traffic flow recovery time is predicted.

[0009] Furthermore, the preliminary analysis results and the secondary analysis results are used to confirm the events, and the logical relationships and events with the same target are merged to obtain the final event recognition results, including: Verify the consistency of the primary and secondary analysis results; Use correlation analysis to determine whether the events in the primary and secondary analysis are logically related, and identify potential connections between events based on time series and event causality; Remove redundant event information and merge the same events or events with the same logical target that are detected multiple times; Confirm and adjust the priority of events based on the results of secondary analysis, and prioritize high-impact events; Based on the confirmation and merging results, the final event identification results are generated, including event classification, relevance, impact scope and priority setting.

[0010] Furthermore, according to the attributes of the traffic event and the event recognition result, warning information is generated, and the warning information includes key features and influencing factors related to the event, including: Generate early warning information containing event characteristics based on the type, severity, location and impact range of the event; Analyze the time characteristics of events and the impact on traffic flow, and dynamically update warning information to reflect the real-time development of events; Generate different levels of warning information based on event priority and determine the scope of dissemination; Based on the future development trend of the event, early warning information including detour suggestions and possible affected areas is generated.

[0011] Furthermore, the feedback data after each event processing is input into the recognition model for self-learning and optimization, and the model parameters, event classification standards and warning rules are dynamically optimized, including: Collect feedback data after event processing, including event processing results, response time, accuracy of warning, response effect and traffic flow recovery; Input the collected feedback data into the recognition model, compare the actual event processing results with the model prediction results through a difference analysis algorithm, and adjust the parameters of the model; Retraining the event classification model according to the classification error rate in the feedback data, and iteratively updating the existing classification rules by using the new labeled data; Analyze the efficiency of event processing and the delay of warning response, adjust the warning trigger conditions, modify the warning rules by optimizing the threshold of triggering events and adjusting the response time window, and redefine the warning priority according to the scope of event impact.

[0012] Further, according to the classification error rate in the feedback data, the event classification model is retrained, and the existing classification rules are iteratively updated by using the new labeled data, including: By comparing the actual event type with the model classification result, the classification error rate is calculated and the misclassified event type is marked; Retraining the model using the collected feedback data, wherein the retraining includes re-labeling the misclassified events as correct categories and iteratively optimizing the recognition model based on the labeled data; Adjust the classification boundaries of the existing event classification model by modifying the thresholds in the classification criteria and reallocating feature weights; Based on the unclassified and new types of events detected in the feedback data, new event categories are dynamically generated to expand the category set of the classification model.

[0013] A highway traffic incident recognition and warning system, the system comprising: A traffic data collection module collects traffic data in real time through sensing devices installed along the highway and pre-processes the traffic data; A preliminary identification and classification module, which passes the pre-processed traffic data through the edge computing node to perform preliminary identification and classification of conventional traffic events to obtain preliminary analysis results; An event secondary analysis module inputs the preliminary analysis results and unclassified traffic data into the cloud, and the cloud performs secondary analysis and semantic understanding of the traffic event based on a large-scale pre-trained recognition model, and combines multimodal data sources to obtain a secondary analysis result of the event; The recognition result acquisition module confirms the events of the preliminary analysis results and the secondary analysis results, merges the logical relationships and the events with the same target, and obtains the final event recognition results; A warning information generation module, which generates warning information according to the attributes of the traffic event and the event recognition result, wherein the warning information includes key features and influencing factors related to the event; The self-learning and optimization module inputs the feedback data after each event processing into the recognition model for self-learning and optimization, and dynamically optimizes the model parameters, event classification standards and early warning rules.

[0014] Furthermore, the preliminary identification and classification module includes: Differentiate between real traffic events and background noise through dynamic algorithms, and automatically correct and compensate for sensor data anomalies based on data interaction with adjacent monitoring points; Based on historical traffic data and traffic trends, edge computing nodes conduct forward-looking identification and correlation reasoning of events, predict potential secondary events, and set priorities based on the severity and potential impact of the events; Through the multi-level event splitting and merging mechanism, similar and related traffic abnormal behaviors are integrated.

[0015] The technical solution of the present invention can achieve the following technical effects: The present invention effectively solves the problems of false alarms and missed alarms in traffic event identification, improves identification accuracy, reduces duplicate reporting, and can dynamically adapt to complex scenarios and inconsistent standards. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0017] Figure 1 A schematic diagram of a process for identifying and warning a highway traffic incident; Figure 2 Schematic diagram of the process to obtain preliminary analysis results; Figure 3 A schematic diagram of a process for obtaining secondary analysis results of an event; Figure 4 A flowchart for event decomposition and reconstruction; Figure 5 A flowchart for obtaining the final event recognition result; Figure 6 A schematic diagram of a process for generating warning information based on the attributes of a traffic event and the event recognition results; Figure 7 A schematic diagram of the process of inputting feedback data into the recognition model for self-learning and optimization; Figure 8 Schematic diagram of the process of retraining the event classification model based on the classification error rate. DETAILED DESCRIPTION

[0018] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0019] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art of the present invention. The terms used in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "and / or" used herein includes any and all combinations of one or more related listed items.

[0020] Embodiment 1 like Figure 1As shown, the present invention provides a highway traffic incident recognition and warning method, the method comprising: S1: collect traffic data in real time through sensing devices installed along the highway and pre-process the traffic data; Specifically, first, sensing devices such as cameras, radars and vehicle sensors are deployed along the highway to collect multimodal data including vehicle location, speed, traffic flow, road conditions and weather information in real time; then, the data is preprocessed, and the steps include data cleaning to remove noise and redundant information, as well as format conversion and standardization processing to ensure that data from different sensors can be used in a unified manner; then, the system performs feature extraction on the preprocessed data to extract key information such as vehicle trajectory, speed changes, and traffic density. These data are combined through multimodal fusion, combining video, radar and sensor data, which enhances the comprehensive understanding of traffic conditions and provides accurate and reliable data support for subsequent traffic event identification and early warning.

[0021] S2: The pre-processed traffic data is passed through the edge computing node to perform preliminary identification and classification of conventional traffic events to obtain preliminary analysis results; Specifically, edge computing nodes can quickly process and analyze real-time traffic data, and perform preliminary identification and classification of routine events close to the data source, such as traffic congestion, speeding, wrong-way driving, accidents, etc. By performing analysis locally, edge computing can significantly reduce the system's response time and latency, ensuring that effective judgments can be made in the early stages of an incident. This process is designed to provide a foundation for subsequent in-depth analysis and processing, enabling the system to more efficiently process large-scale, real-time traffic data, thereby optimizing the overall traffic management and early warning response speed.

[0022] S3: The preliminary analysis results and unclassified traffic data are input into the cloud. The cloud performs secondary analysis and semantic understanding of the traffic events based on a large-scale pre-trained recognition model, and combines multimodal data sources to obtain secondary analysis results of the events. Specifically, by utilizing large-scale pre-trained models in the cloud, we can conduct a deeper analysis and semantic understanding of traffic events, improve the accuracy of event recognition and the processing capabilities in complex scenarios. The cloud's powerful computing resources can perform in-depth semantic analysis of more complex traffic events (such as multi-vehicle accidents, abnormal traffic behaviors caused by weather conditions, etc.), and combine multimodal data from different sources (such as video, radar, and environmental sensor data) for cross-validation and supplementation to solve scenarios that cannot be handled by a single data source or edge computing node.

[0023] S4: Confirm the events based on the preliminary analysis results and the secondary analysis results, merge the events with logical relationships and the same target, and obtain the final event recognition results; Specifically, by comparing the preliminary analysis of edge computing and the in-depth analysis of the cloud, we ensure that the results of the two are logically consistent, avoid multiple reports of the same event, and determine whether different analysis results belong to the same event by analyzing key factors such as the time, location, and type of the event. Logically related events are merged and processed. This not only helps to reduce duplicate reports and false alarms, ensure the accuracy and uniqueness of the final event identification results, but also better understand the overall development of the event, especially the connection between multi-stage and complex events.

[0024] S5: Generate warning information based on the attributes of the traffic event and the event recognition results. The warning information includes key features and influencing factors related to the event. Specifically, based on the identified traffic events, accurate warning information is generated so that traffic managers or drivers can understand the current traffic conditions in a timely manner and take appropriate response measures. The generation of warning information mainly includes the type of traffic event (such as accidents, congestion, vehicle failures, etc.), the location of the event, the severity of the event, the scope of impact, and the estimated duration, etc. These key characteristics and influencing factors can help relevant parties quickly understand the nature and possible consequences of the event, so as to take appropriate measures (such as adjusting driving routes, issuing traffic control information, etc.). By generating accurate warning information, the response speed of traffic events and the efficiency of emergency handling can be improved, effectively reducing traffic delays and secondary accidents caused by accidents.

[0025] S6: Input the feedback data after each event processing into the recognition model for self-learning and optimization, and dynamically optimize the model parameters, event classification standards and warning rules.

[0026] Specifically, after the event is processed, the feedback data is used to correct deficiencies in the recognition model, and the model parameters are continuously optimized through self-learning to reduce false positives and missed positives. At the same time, the recognition model is dynamically adjusted according to different traffic scenarios and new types of events to adapt to the ever-changing traffic environment. In addition, the analysis of feedback data can also optimize the warning trigger conditions and response strategies to ensure that the warning information is sent at the appropriate time, thereby improving the accuracy of the warning and the timeliness of the response. In general, the core purpose of this step is to continuously optimize the event recognition and classification standards and improve the warning rules through the cyclic iteration of self-learning and feedback data, so that the recognition model can maintain efficient and stable performance in the long run.

[0027] As a preferred embodiment of the above, Figure 2 As shown, in step S2, the pre-processed traffic data is passed through the edge computing node to perform preliminary identification and classification of conventional traffic events to obtain preliminary analysis results, including: S21: Use dynamic algorithms to distinguish between real traffic events and background noise, and automatically correct and compensate for sensor data anomalies based on data interaction with adjacent monitoring points; S22: Based on historical traffic data and traffic trends, edge computing nodes perform forward-looking identification and correlation reasoning of events, predict potential secondary events, and set priorities based on the severity and potential impact of the events; S23: Through the multi-level event splitting and merging mechanism, similar and related traffic abnormal behaviors are integrated.

[0028] Specifically, first, a dynamic algorithm is used to process the collected traffic data to distinguish between real traffic events and background noise, such as filtering out short-term traffic stagnation or environmental interference. In order to ensure the accuracy of the data, the data from adjacent monitoring points is interacted with to automatically correct and compensate for abnormal conditions in the sensor data, such as sensor failure or signal loss. Then, historical traffic data and current traffic trends are combined to conduct forward-looking identification and association reasoning to predict possible secondary events, such as traffic congestion caused by traffic accidents, and set event priorities based on the severity and scope of the event to ensure that important events are given priority. In addition, through a multi-level event splitting and merging mechanism, similar or related traffic abnormal behaviors are integrated to avoid repeated reporting and information redundancy. The synergistic effect of each step makes the initial identification and classification of traffic events more efficient and accurate, providing a basis for subsequent more in-depth analysis and processing.

[0029] As a preferred embodiment of the above, Figure 3 As shown, in step S3, the preliminary analysis results and the unclassified traffic data are input into the cloud, and the cloud performs secondary analysis and semantic understanding of the traffic event based on a large-scale pre-trained recognition model, and combines the multimodal data source to obtain the secondary analysis results of the event, including: S31: Through contextual semantic analysis of events, combined with historical data and current context information, the background and relevance of traffic events are understood, and the complexity and potential impact of events are judged; S32: Combine multimodal data sources to cross-validate and supplement traffic events; S33: For complex traffic events, decompose and reconstruct the events, identify the key triggering factors and possible secondary events, and perform reconstruction analysis; S34: Deep semantic recognition of abnormal behaviors, analyzing unresolved atypical events and abnormal behaviors in the initial recognition results based on large-scale pre-trained models; S35: Based on multimodal data and historical data, infer the impact scope of traffic events and predict the future development trend and potential traffic impact of the events.

[0030] Specifically, first, through contextual semantic analysis, combined with historical data and current traffic conditions, a comprehensive understanding of the background and relevance of the event is carried out to determine the complexity and potential impact of the event; then, multimodal data sources, such as video, image, radar and environmental data, are used to cross-validate and supplement data of traffic events to ensure mutual verification among different data sources and improve the accuracy of analysis. For complex traffic events, such as multi-vehicle accidents or multi-section anomalies, the system will decompose the event, identify key triggering factors, and reconstruct the entire process of the event to predict possible secondary events. For atypical or abnormal traffic behaviors, deep semantic recognition is performed based on large-scale pre-trained models to resolve abnormal phenomena that were not fully understood in the initial identification; finally, multimodal data is combined with historical data to infer the scope of the impact of the traffic event, predict the future development trend of the event, and evaluate its potential impact on traffic flow and region.

[0031] As a preferred embodiment of the above, Figure 4 As shown, event decomposition and reconstruction include: A10: Identify the triggers and causes of events and decompose the key components of events through multimodal data sources and historical traffic data; A20: Based on the cause of the event, predict the secondary events that may be triggered by the event and analyze their potential impact in advance; A30: Based on the decomposition results of the events, reconstruct the time sequence of the events and reveal the causal relationship and development process of the events; A40: Based on the traffic flow model, evaluate the extended impact of the incident on the current road section and surrounding areas, and predict the traffic flow recovery time.

[0032] Specifically, the triggering factors and causes of traffic incidents are identified by combining multimodal data sources (such as cameras, radars, vehicle sensors, etc.) with historical traffic data. This process decomposes the key components of the incident, including the cause of the accident (such as vehicle failure, road conditions, weather, etc.) and external factors that affect the development of the incident. Through the detailed decomposition of the components of the incident, the essence of the incident can be more accurately identified, providing a basis for subsequent processing; next, based on the analysis of the cause of the incident, further evaluate the secondary events that may be caused by the incident and predict their potential impact in advance. For example, a traffic accident may cause subsequent congestion or even secondary accidents. Through the prediction of secondary events, it can help management departments take preventive measures in advance to reduce the probability of secondary events and reduce their further interference with traffic flow; then, based on the previously decomposed event components, reconstruct the time series of the event to reveal the causal relationship and development process of the event. This step helps to understand the entire process of the event from its cause to its development, analyze the interactions and impacts between the various components, and through time series reconstruction, not only can the dynamic changes of the event be displayed, but also the potential risk points in the event can be identified, so as to facilitate more effective early warning and response measures in similar events in the future; finally, combined with the traffic flow model, the extended impact of the current event on the road and surrounding areas is evaluated, and the degree of interference of the event on traffic flow is quantified. Based on the evaluation results, the scope of the event's impact on traffic and the estimated recovery time can be predicted.

[0033] As a preferred embodiment of the above, Figure 5 As shown, step S4 is to confirm the events of the preliminary analysis results and the secondary analysis results, merge the logical relationships and the events with the same target, and obtain the final event recognition results, including: S41: Verify the consistency of the primary analysis results and the secondary analysis results; S42: Use correlation analysis to determine whether the events in the primary and secondary analysis are logically related, and identify potential connections between events based on time series and event causality; S43: removing redundant event information, and merging the same events or events with the same logical target that are detected multiple times; S44: Confirm and adjust the priority of events based on the results of the secondary analysis, and prioritize high-impact events; S45: Based on the confirmation and merging results, generate the final event identification results, including event classification, relevance, impact scope and priority setting.

[0034] Specifically, first, consistency verification is performed to ensure that the preliminary analysis results are consistent with the secondary analysis results on the cloud in key attributes. Attributes such as time, location, and scope of impact are the core elements of the event. By verifying whether these elements match, it can be ensured that the events targeted by the two analysis results are the same, avoiding false alarms or misreporting problems caused by data bias. For example, if the time or location of the event identified by edge computing does not match the analysis results on the cloud, the system may mark the event as requiring further analysis or re-detection; next, correlation analysis is performed. This step reveals the potential causal relationship between events by judging the logical correlation between the preliminary analysis and the secondary analysis. For complex traffic events, multiple small events often have logical connections. For example, a car accident in front causes an emergency stop or a chain collision of the vehicle behind. Through time series analysis, the causal relationship between these events can be discovered, ensuring that related events can be handled uniformly by the system rather than separately, reducing information dispersion and management confusion; then, redundancy removal is performed. This step aims to merge multiple detected identical events or events with the same target. In a traffic monitoring network, multiple Multiple sensors or monitoring devices may report the same event at the same time, especially when cameras, radars and other devices overlap their coverage areas. At this time, if the system does not remove redundancy, multiple identical events will be reported repeatedly, increasing unnecessary processing burden and causing data redundancy. By merging identical events, duplicate reporting can be reduced, resource allocation can be optimized, and event processing efficiency can be improved. Priority confirmation and adjustment are the next key steps. Based on the secondary analysis results of the event, the system will confirm and adjust the priority of the event. This process involves re-evaluating the scope and severity of the event to ensure that those events with a greater impact on traffic (such as major accidents or emergencies) can be given priority. Finally, the final event identification result is generated, which is the summary step of the entire process. By integrating the aforementioned confirmation, merging and priority adjustment, a complete event identification result is generated. The result not only includes the classification, relevance, and scope of impact of the event, but also includes priority setting. Through this information, traffic managers can quickly understand the overall picture of the current event and its scope of impact, so as to formulate effective processing plans to ensure rapid response and minimize traffic disruptions and risks.

[0035] As a preferred embodiment of the above, Figure 6 As shown, step S5 generates warning information according to the attributes of the traffic event and the event recognition result. The warning information includes key features and influencing factors related to the event, including: S51: Generate warning information containing event characteristics based on the type, severity, location and impact range of the event; S52: Analyze the time characteristics and traffic flow impact of the event, and dynamically update the warning information to reflect the real-time development of the event; S53: Generate different levels of warning information based on event priority and determine the scope of dissemination; S54: Generate warning information including detour suggestions and possible affected areas based on the future development trend of the event.

[0036] Specifically, in the process of generating warning information based on the attributes and identification results of traffic events, warning information that is consistent with the event is first generated based on the characteristics of the event (such as type, severity, location, etc.). This information is intended to ensure that relevant personnel can quickly understand the key content and impact area of ​​the event; as the event develops, the warning information will be dynamically updated based on the real-time traffic flow and the duration of the event to ensure the accuracy and real-time nature of the information. In addition, the priority of the warning information will be divided according to the importance of the event so that high-priority events can quickly attract attention and receive a response. At the same time, the scope of information dissemination is determined to ensure that people within different impact ranges can receive corresponding reminders; finally, the warning information will also provide predictions of future traffic impacts, including possible detour suggestions and affected areas, to help drivers avoid congestion and reduce the occurrence of secondary accidents. The synergistic effect of these steps ensures that the warning information not only reflects current events, but can also make forward-looking predictions about possible future traffic conditions.

[0037] As a preferred embodiment of the above, Figure 7 As shown, step S6, inputting the feedback data after each event processing into the recognition model for self-learning and optimization, dynamically optimizing the model parameters, event classification standards and warning rules, including: S61: Collect feedback data after event processing, including event processing results, response time, accuracy of warning, response effect and traffic flow recovery status; S62: Input the collected feedback data into the recognition model, compare the actual event processing results with the model prediction results through the difference analysis algorithm, and adjust the parameters of the model; S63: Retrain the event classification model based on the classification error rate in the feedback data, and iteratively update the existing classification rules by using the new labeled data; S64: Analyze the efficiency of event processing and the delay of warning response, adjust the warning trigger conditions, modify the warning rules by optimizing the threshold of triggering events and adjusting the response time window, and redefine the warning priority according to the scope of event impact.

[0038] Specifically, first, feedback data is collected after each event is processed. These data include the specific results of event processing, event response time, warning accuracy, response effect, and traffic flow recovery. These feedback data provide a practical basis for subsequent model optimization. By tracking the event results, the performance of the existing model can be evaluated. Then, the collected feedback data is input into the recognition model, and the actual event processing results are compared with the model's prediction results through a difference analysis algorithm. Through this comparison, the deficiencies in the model can be discovered. For example, there is a deviation between the prediction and actual processing of certain events. According to the difference results, the model parameters are adjusted to improve its recognition and prediction accuracy. Then, according to the feedback data, The classification error rate is reduced and the event classification model is retrained. By using the new labeled data, the existing classification rules are iteratively updated, especially the previously misclassified events are corrected. Through continuous iterative optimization, the model can gradually reduce classification errors and improve the recognition accuracy of different types of traffic events. Finally, the efficiency of event processing and the delay of warning response are analyzed, and the warning trigger conditions are adjusted based on this. By optimizing the threshold of event triggering and adjusting the warning response time window, it can ensure that the warning information is issued at an appropriate time and respond to the event in a timely manner. In addition, combined with the scope of impact of the event, the priority of the warning is redefined to ensure that major events are handled first, while non-emergency events can be postponed or responded to slowly.

[0039] As a preferred embodiment of the above, Figure 8 As shown, step S63, based on the classification error rate in the feedback data, retrain the event classification model, and iteratively update the existing classification rules by using the new labeled data, including: S631: Calculate the classification error rate by comparing the actual event type with the model classification result, and mark the misclassified event type; S632: retraining the model using the collected feedback data, where the retraining includes re-labeling the misclassified events as correct categories, and iteratively optimizing the recognition model based on the labeled data; S633: Adjust the classification boundary of the existing event classification model by modifying the threshold in the classification standard and reallocating the feature weights; S634: Dynamically generate new event categories based on unclassified and new types of events detected in the feedback data, and expand the category set of the classification model.

[0040] Specifically, first, by comparing the actual event types with the model classification results, the classification error rate is calculated, and those misclassified events are marked. The purpose of this step is to identify the errors that occur in the model classification process, and to determine the specific misclassification situations by analyzing the differences between the actual events and the model prediction results. This provides a basis for subsequent model optimization and lays the foundation for correcting classification errors; secondly, the model is retrained using the collected feedback data. During the retraining process, the misclassified events are relabeled as the correct categories, and the model is iteratively optimized based on these new labeled data. This step introduces actual labeled data to allow the model to learn the correct classification method, gradually reduce errors in future classifications, and ensure that the model can It can continuously adapt to new data and improve classification accuracy; while retraining, it is also necessary to adjust the classification boundaries of the event classification model. By modifying the threshold in the classification standard and redistributing the weights of each feature, the model can handle boundary events more accurately. This adjustment ensures that the model has higher processing accuracy for events that are fuzzy or close to the classification boundary, reducing the possibility of misjudgment; finally, based on the unclassified events or new types of events detected in the feedback data, new event categories are dynamically generated to expand the category set of the model. This step improves the flexibility of the model by introducing new event types, enabling it to adapt to the ever-changing traffic environment and emerging events, ensuring that the system has higher adaptability and processing capabilities when dealing with complex or emergency situations.

[0041] Embodiment 2 Based on the same inventive concept as the highway traffic incident recognition and warning method in the aforementioned embodiment, the present invention further provides a highway traffic incident recognition and warning system, the system comprising: Traffic data collection module, which collects traffic data in real time through sensing devices installed along the highway and pre-processes the traffic data; The preliminary identification and classification module passes the pre-processed traffic data through the edge computing node to perform preliminary identification and classification of routine traffic events and obtain preliminary analysis results; The event secondary analysis module inputs the preliminary analysis results and unclassified traffic data into the cloud. The cloud performs secondary analysis and semantic understanding of traffic events based on a large-scale pre-trained recognition model, and combines multimodal data sources to obtain secondary analysis results of the events. The recognition result acquisition module confirms the events of the preliminary analysis results and the secondary analysis results, merges the logical relationships and the events with the same target, and obtains the final event recognition results; The warning information generation module generates warning information according to the attributes of the traffic event and the event recognition results. The warning information includes key features and influencing factors related to the event. The self-learning and optimization module inputs the feedback data after each event processing into the recognition model for self-learning and optimization, and dynamically optimizes the model parameters, event classification standards and warning rules.

[0042] The above-mentioned system in the present invention can effectively implement the highway traffic incident identification and warning method, and the technical effects that can be achieved are as described in the above-mentioned embodiments, which will not be repeated here.

[0043] As a preferred embodiment of the above, the preliminary identification and classification module includes: Differentiate between real traffic events and background noise through dynamic algorithms, and automatically correct and compensate for sensor data anomalies based on data interaction with adjacent monitoring points; Based on historical traffic data and traffic trends, edge computing nodes conduct forward-looking identification and correlation reasoning of events, predict potential secondary events, and set priorities based on the severity and potential impact of the events; Through the multi-level event splitting and merging mechanism, similar and related traffic abnormal behaviors are integrated Similarly, the above-mentioned optimization schemes for the system can also respectively achieve the optimization effects corresponding to the method in Example 1, which will not be repeated here.

[0044] Although the present application has been described in conjunction with specific features and embodiments thereof, it is obvious that various modifications and combinations may be made thereto without departing from the spirit and scope of the present application. Accordingly, this specification and the accompanying drawings are merely exemplary illustrations of the present application as defined therein, and are deemed to have covered any and all modifications, variations, combinations or equivalents within the scope of the present application. Obviously, a person skilled in the art may make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application is intended to include these modifications and variations.

Claims

1. A highway traffic incident identification and warning method, characterized in that: include: Collecting traffic data in real time through sensing devices installed along the highway, and preprocessing the traffic data; The pre-processed traffic data is passed through the edge computing node to perform preliminary identification and classification of conventional traffic events to obtain preliminary analysis results; Inputting the preliminary analysis results and unclassified traffic data into the cloud, the cloud performs secondary analysis and semantic understanding of the traffic event based on a large-scale pre-trained recognition model, and combines multimodal data sources to obtain secondary analysis results of the event; Confirm the events of the preliminary analysis results and the secondary analysis results, merge the events with logical relationships and the same target, and obtain the final event identification results; Generate warning information according to the attributes of the traffic event and the event recognition result, wherein the warning information includes key features and influencing factors related to the event; The feedback data after each event processing is input into the recognition model for self-learning and optimization, and the model parameters, event classification standards and early warning rules are dynamically optimized.

2. The highway traffic incident identification and warning method according to claim 1, characterized in that: The pre-processed traffic data is passed through the edge computing node to perform preliminary identification and classification of conventional traffic events to obtain preliminary analysis results, including: Differentiate between real traffic events and background noise through dynamic algorithms, and automatically correct and compensate for sensor data anomalies based on data interaction with adjacent monitoring points; Based on historical traffic data and traffic trends, edge computing nodes conduct forward-looking identification and correlation reasoning of events, predict potential secondary events, and set priorities based on the severity and potential impact of the events; Through the multi-level event splitting and merging mechanism, similar and related traffic abnormal behaviors are integrated.

3. The highway traffic incident identification and warning method according to claim 1, characterized in that: The preliminary analysis results and the unclassified traffic data are input into the cloud, and the cloud performs secondary analysis and semantic understanding of the traffic event based on a large-scale pre-trained recognition model, and combines multimodal data sources to obtain secondary analysis results of the event, including: Through contextual semantic analysis of events, combined with historical data and current context information, the background and relevance of traffic events can be understood, and the complexity and potential impact of events can be judged; Combine multimodal data sources to cross-validate and supplement traffic events; For complex traffic incidents, decompose and reconstruct the incidents, identify the key triggering factors and possible secondary events, and conduct reconstruction analysis; Perform deep semantic recognition of abnormal behaviors, and analyze unresolved atypical events and abnormal behaviors in the preliminary recognition results based on large-scale pre-trained models; Based on multimodal data and historical data, the impact scope of traffic events is inferred, and the future development trend and potential traffic impact of the events are predicted.

4. The highway traffic incident identification and warning method according to claim 3, characterized in that: Event decomposition and reconstruction include: Identify the triggers and causes of incidents and break down the key components of incidents using multimodal data sources and historical traffic data; Based on the cause of the event, predict the secondary events that may be triggered by the event and analyze their potential impact in advance; Based on the decomposition results of the events, the time sequence of the events is reconstructed to reveal the causal relationship and development process of the events; Based on the traffic flow model, the impact of the incident on the current road section and surrounding areas is evaluated, and the traffic flow recovery time is predicted.

5. The highway traffic incident identification and warning method according to claim 4, characterized in that: Confirm the events of the preliminary analysis results and the secondary analysis results, merge the logical relationships and events with the same target, and obtain the final event identification results, including: Verify the consistency of the primary and secondary analysis results; Use correlation analysis to determine whether the events in the primary and secondary analysis are logically related, and identify potential connections between events based on time series and event causality; Remove redundant event information and merge the same events or events with the same logical target that are detected multiple times; Confirm and adjust the priority of events based on the results of secondary analysis, and prioritize high-impact events; Based on the confirmation and merging results, the final event identification results are generated, including event classification, relevance, impact scope and priority setting.

6. The highway traffic incident identification and warning method according to claim 1, characterized in that: According to the attributes of the traffic event and the event recognition result, warning information is generated, and the warning information includes key features and influencing factors related to the event, including: Generate early warning information containing event characteristics based on the type, severity, location and impact range of the event; Analyze the time characteristics of events and the impact on traffic flow, and dynamically update warning information to reflect the real-time development of events; Generate different levels of warning information based on event priority and determine the scope of dissemination; Based on the future development trend of the event, early warning information including detour suggestions and possible affected areas is generated.

7. The highway traffic incident identification and warning method according to claim 1, characterized in that: Feedback data after each event processing is input into the recognition model for self-learning and optimization, and model parameters, event classification standards and warning rules are dynamically optimized, including: Collect feedback data after event processing, including event processing results, response time, accuracy of warning, response effect and traffic flow recovery; Input the collected feedback data into the recognition model, compare the actual event processing results with the model prediction results through a difference analysis algorithm, and adjust the parameters of the model; Retraining the event classification model according to the classification error rate in the feedback data, and iteratively updating the existing classification rules by using the new labeled data; Analyze the efficiency of event processing and the delay of warning response, adjust the warning trigger conditions, modify the warning rules by optimizing the threshold of triggering events and adjusting the response time window, and redefine the warning priority according to the scope of event impact.

8. The highway traffic incident identification and warning method according to claim 7, characterized in that: Retrain the event classification model based on the classification error rate in the feedback data, and iteratively update the existing classification rules by using the new labeled data, including: By comparing the actual event type with the model classification result, the classification error rate is calculated and the misclassified event type is marked; Retraining the model using the collected feedback data, wherein the retraining includes re-labeling the misclassified events as correct categories and iteratively optimizing the recognition model based on the labeled data; Adjust the classification boundaries of the existing event classification model by modifying the thresholds in the classification criteria and reallocating feature weights; Based on the unclassified and new types of events detected in the feedback data, new event categories are dynamically generated to expand the category set of the classification model.

9. A highway traffic incident recognition and warning system, characterized in that: The system comprises: A traffic data collection module collects traffic data in real time through sensing devices installed along the highway and pre-processes the traffic data; A preliminary identification and classification module, which passes the pre-processed traffic data through the edge computing node to perform preliminary identification and classification of conventional traffic events to obtain preliminary analysis results; An event secondary analysis module inputs the preliminary analysis results and unclassified traffic data into the cloud, and the cloud performs secondary analysis and semantic understanding of the traffic event based on a large-scale pre-trained recognition model, and combines multimodal data sources to obtain a secondary analysis result of the event; The recognition result acquisition module confirms the events of the preliminary analysis results and the secondary analysis results, merges the logical relationships and the events with the same target, and obtains the final event recognition results; A warning information generation module, which generates warning information according to the attributes of the traffic event and the event recognition result, wherein the warning information includes key features and influencing factors related to the event; The self-learning and optimization module inputs the feedback data after each event processing into the recognition model for self-learning and optimization, and dynamically optimizes the model parameters, event classification standards and early warning rules.

10. The highway traffic incident recognition and warning system according to claim 9, characterized in that: The preliminary identification and classification module comprises: Differentiate between real traffic events and background noise through dynamic algorithms, and automatically correct and compensate for sensor data anomalies based on data interaction with adjacent monitoring points; Based on historical traffic data and traffic trends, edge computing nodes conduct forward-looking identification and correlation reasoning of events, predict potential secondary events, and set priorities based on the severity and potential impact of the events; Through the multi-level event splitting and merging mechanism, similar and related traffic abnormal behaviors are integrated.

Images