Expressway toll station special situation event processing system based on virtual digital human

Abstract

The invention relates to the technical field of image recognition, in particular to a highway toll station special situation event processing system based on virtual digital humans, which comprises an image annotation training module, a situation reasoning module, a dynamic image generation module, an event response module and a correction marking module. According to the method, the toll station image is labeled in a structured mode, the target category, position, time and environment information are integrated, high-risk areas and time periods are accurately divided, early recognition of key events is achieved, multi-dimensional data and time sequence changes are fused, dynamic understanding of traffic density and behavior characteristics is enhanced, the situation recognition capability is improved, and the method is suitable for popularization and application. Virtual prompt content is dynamically adjusted according to people flow density and urgency, interactive feedback personalization is achieved, a response strategy focuses on high-priority events, resource dispersion is avoided, a deviation recognition and re-classification mechanism is constructed through state backtracking and feedback correction, and an efficient processing closed loop integrating perception, decision making, response and feedback is formed.

Description

Technical Field

[0001] The present invention relates to the technical field of image recognition, and in particular to a special event processing system for a highway toll station based on a virtual digital human. Background Art

[0002] The field of image recognition technology includes a collection of technologies that use computer vision algorithms to extract features and determine the status of dynamic targets in traffic scenes. The core content of this technology field focuses on real-time analysis of video streams, multi-target behavior pattern recognition, and abnormal event detection model deployment. The systematic technical architecture consists of a front-end image acquisition unit, edge computing nodes, model inference acceleration modules, and event response interfaces. It specifically involves technical links such as video codec optimization, target detection network pruning, multi-threaded parallel processing, and detection result confidence calibration, forming a full-link technology system from raw data input to structured event output.

[0003] The highway toll station special event processing system refers to a processing system for special events within highway toll stations. It is specifically designed to detect and handle special events such as pedestrians entering the road, two-wheeled vehicles crossing the road, and hazardous chemical vehicles entering during prohibited driving hours. It collects video data through cameras, uses image recognition technology to process and analyze video frames, identifies abnormal events, and combines digital human image display and voice generation technology to provide real-time event prompts or warnings to on-site personnel to ensure that abnormal situations are handled promptly. Specifically, it includes multiple cameras for video data collection, GPU servers for image recognition and processing, and management clients and virtual digital human displays for event display and voice reminders. Based on a trained image recognition model, it can automatically detect and identify abnormal situations in video frames.

[0004] Existing event recognition relies on frame-level object detection and fixed rules for anomaly determination. It lacks multidimensional analysis of environmental variables and behavioral dynamics, resulting in consistent recognition results for the same object in different contexts and failing to achieve a differentiated understanding of the event's essence. During the event screening phase, decisions are made directly based on model confidence, without incorporating complex contextual relationship modeling. This leads to high misjudgment rates in situations where multiple objects intersect and behaviors overlap. For example, crowds during rush hour can be misidentified as intrusions, generating false alarms. On-site feedback mechanisms primarily rely on static text or voice prompts, lacking coordinated adjustments to on-site conditions, weakening the timeliness and relevance of warnings. Post-event processing, there is a lack of review and evaluation of execution effectiveness and deviation correction, making it prone to model results deviating from actual traffic conditions and failing to perceive them, impacting the robustness of the overall operation. The overall process exhibits limitations due to its single recognition logic, rigid response formats, and lack of feedback mechanisms, making it difficult to adapt to the complex dynamics of multi-source, heterogeneous events in dynamic traffic scenarios. Summary of the Invention

[0005] In order to solve the technical problems existing in the prior art, the embodiment of the present invention provides a highway toll station special event processing system based on virtual digital human. The technical solution is as follows:

[0006] The highway toll station special event processing system based on virtual digital human includes:

[0007] The image annotation training module obtains road image data from highway toll booths, extracts traffic object categories, locations, and time information, and combines weather and time period environment descriptions to determine the event type interval within the current sample set. It then selects event blocks distributed in risky time periods and regions, and generates an annotated event recognition map.

[0008] The contextual reasoning module collects traffic density, vehicle type, and behavior characteristics of the corresponding block based on the annotated event recognition map, compares the change direction and correlation differences of the multi-dimensional data according to the sampling time series, identifies the dynamic characteristics of events in the block, and generates a contextual reasoning state;

[0009] The dynamic image generation module dynamically adjusts the facial expressions, body movements, and voice intonation of the virtual digital human based on the contextual reasoning state and the real-time traffic flow and urgency of the incident in the perception area to optimize the interactive experience and response effect, and generates a dynamic image configuration table;

[0010] The event response module calls the dynamic image configuration table, classifies the event block map, eliminates areas overlapping with non-emergency time periods, compares the severity of the event with the response level, marks the time periods and areas that require priority processing, and generates an event intervention interval configuration table.

[0011] As a further solution of the present invention, the labeled event recognition map includes event type classification labels, block time series indexes, risk period and area identification information; the situational reasoning state includes traffic density change trends, vehicle behavior pattern characteristics, and event correlation analysis results; the dynamic image configuration table includes digital human appearance adjustment parameters, voice tone change range, and interactive behavior description; the event intervention interval configuration table includes intervention block number, target response level, and processing time window.

[0012] As a further solution of the present invention, the image annotation training module includes:

[0013] The object annotation submodule obtains highway toll booth road image data, identifies traffic object change sequences through the same block time series, combines category and location information, extracts event control blocks that meet dynamic change characteristics, and generates a set of annotation feature response block numbers;

[0014] The situation description submodule calls the annotated feature response block number set, extracts weather, time period and traffic flow data, identifies the correlation between environmental variables and event occurrence, filters the block numbers and time periods that are both within the risk period and area, and generates a risk section identification set;

[0015] The event block determination submodule compares the traffic density and the changing trend of the time series nodes in the block according to the risk section identification set, matches the abnormal event benchmark range, extracts the valid blocks and spatial positions, and generates a marked event recognition map.

[0016] As a further solution of the present invention, the contextual reasoning module includes:

[0017] The traffic feature acquisition submodule collects traffic density and vehicle behavior feature readings based on the annotated event recognition map, unifies the sampling time series, removes anomalies and repairs missing segments of the block data, extracts valid feature samples, and generates a basic set of block feature monitoring;

[0018] The behavior pattern comparison submodule calls the block feature monitoring basic set, extracts the traffic density and vehicle behavior change values ​​in adjacent time periods, compares the multi-dimensional feature change directions and correlation differences, identifies coupling features, and generates a coupling feature coefficient set;

[0019] The situational reasoning modeling submodule calls the coupling characteristic coefficient set, selects the stable trend block, extracts the traffic density change amplitude and vehicle behavior change rate, combines the sampling interval, sequence position and number of sample points, calculates the traffic density change amplitude, and maps it with the spatial range and time node of the event block to generate the situational reasoning state.

[0020] As a further solution of the present invention, the traffic density variation range is calculated using the formula:

[0021]

[0022] Calculation is performed, where ΔD represents the change in traffic density, ρ i represents the traffic density of the data point at time i, ρ i-1 represents the traffic density of the data point at the i-1th moment, Q represents the sampling interval, and N represents the number of consecutive data points at each moment.

[0023] As a further solution of the present invention, the dynamic image generation module includes:

[0024] The environmental perception submodule extracts traffic feature data at each layer based on the contextual reasoning state, combines real-time traffic flow with the urgency of the incident, analyzes the changing rhythm of traffic features and the fluctuation amplitude of environmental variables, determines whether they are within the matching threshold range, selects the number sequence that meets the conditions, and generates an image adjustment warning number group;

[0025] The appearance adjustment submodule selects the traffic characteristics and event urgency information of the corresponding time period based on the image adjustment warning number group, identifies the overlapping time period of the two types of data, and generates an appearance adjustment plan;

[0026] The voice feedback derivation submodule calls the appearance adjustment scheme, compares the traffic characteristics of the section with the event urgency data, extracts the rhythm overlap positions, groups and summarizes the voice feedback differences of each combination, and classifies them into response levels, summarizes the level change trajectory, and generates a dynamic image configuration table.

[0027] As a further solution of the present invention, the event response module includes:

[0028] The trend classification submodule calls the dynamic image configuration table, extracts the traffic change direction and amplitude value of the area in the event block map, classifies the trend type according to the continuity and gradient change judgment, eliminates the block numbers that overlap with the non-emergency period, and generates a responsive trend block set;

[0029] The priority comparison submodule calls the respondable trend block set, collects the current event severity and response level standard range of the block, compares the event severity and response lower limit value by block, sorts the block numbers, and obtains the priority response block sequence;

[0030] The intervention interval identification submodule extracts the corresponding response time interval according to the block number in the priority response block sequence, sets the response level and processing time interval configuration according to the continuous deviation of the event, and generates an event intervention interval configuration table.

[0031] As a further solution of the present invention, the event intervention interval configuration value adopts the formula:

[0032]

[0033] Calculation is performed, where I represents the event intervention interval configuration value, T1 represents the end time corresponding to the block number, T0 represents the start time corresponding to the block number, n represents the number of intervals in the response period, and δ j represents the continuous offset of the jth response period, P represents the basic response offset, α j Represents the weight of the j-th response period.

[0034] As a further solution of the present invention, the system further includes a correction marking module:

[0035] The correction marking module detects the traffic state recovery extent and the change in the performance feedback of the virtual digital human after the event processing according to the event intervention interval configuration table, screens and groups the event processing results that deviate from the expected block, marks the corresponding processing settings of the deviated block, and generates a toll station event processing offset label;

[0036] The toll booth event processing offset tag includes the traffic state recovery deviation level, the virtual digital human performance change range, and the processing setting offset mark.

[0037] As a further solution of the present invention, the correction marking module includes:

[0038] The traffic fluctuation screening submodule extracts the block event processing end time according to the event intervention interval configuration table, identifies the traffic status data after the corresponding time, analyzes the difference between the recovery amplitude and the processing target value, screens the blocks that exceed the recovery deviation threshold, and generates a traffic status deviation block set;

[0039] The performance difference determination submodule extracts the performance feedback change readings of the virtual digital human in the corresponding blocks based on the traffic state deviation block set, analyzes the performance change amplitude, and classifies them according to the ratio of the performance change value to the traffic state recovery deviation value, filters out the response inconsistent blocks, and generates a traffic-performance imbalance distribution group;

[0040] The processing offset identification submodule extracts corresponding processing parameters according to the traffic-performance imbalance distribution group and compares them with the processing targets in the configuration table, marks the blocks and setting parameters for setting offsets, and generates a toll booth event processing offset label.

[0041] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0042] By performing structured annotation on toll booth road images and integrating target categories, locations, time, and environmental factors, we can accurately delineate high-risk areas and time periods, ensuring that key events are identified at an early stage. During data processing, we introduce multi-dimensional information fusion and temporal trend analysis to capture the dynamic linkage between traffic density and behavioral characteristics, giving event recognition a stronger understanding of the situation. This capability is no longer limited to static detection of single-frame targets. By combining crowd density and event urgency, we dynamically adjust virtual display content, shifting event prompts from static template output to expressive, personalized interactions, improving information transmission efficiency and on-site response accuracy. A response intervention mechanism is constructed based on event importance and time block priority to avoid interference caused by equal resource allocation and achieve a focused response to key events. By combining the actual traffic status and virtual prompt feedback after event response, a closed-loop verification process is constructed to form a mechanism for identifying and reclassifying deviation events, thereby improving the recovery capability and model adaptability in complex scenarios. The overall process consists of data labeling, situational understanding, dynamic interaction, priority response to closed-loop feedback, forming a highly coupled and highly flexible abnormal event processing closed loop, which greatly enhances the abnormal perception, classification decision-making and interactive intervention capabilities in highway toll station scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0044] Figure 1 Schematic diagram of a highway toll station special event processing system based on a virtual digital human provided by an embodiment of the present invention;

[0045] Figure 2 Schematic diagram of the system framework of the present invention;

[0046] Figure 3 This is a flowchart of the image annotation training module in the present invention;

[0047] Figure 4 This is a flowchart of the situational reasoning module in the present invention;

[0048] Figure 5 This is a flow chart of the dynamic image generation module in the present invention;

[0049] Figure 6 This is a flow chart of the event response module in the present invention;

[0050] Figure 7 This is a flowchart of the modified marking module in the present invention. DETAILED DESCRIPTION

[0051] The technical solution of the present invention is described below in conjunction with the accompanying drawings.

[0052] In the embodiments of the present invention, words such as "exemplarily" and "for example" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as an "exemplary" in the present invention should not be interpreted as being preferred or advantageous over other embodiments or designs. Rather, the use of the word "exemplary" is intended to present concepts in a concrete manner. Furthermore, in the embodiments of the present invention, "and / or" can mean both or either of the two.

[0053] In the embodiments of the present invention, the terms "image" and "picture" may be used interchangeably. It should be noted that, when the distinction between them is not emphasized, their intended meanings are the same. The terms "of," "corresponding," and "corresponding" may be used interchangeably. It should be noted that, when the distinction between them is not emphasized, their intended meanings are the same.

[0054] In the embodiments of the present invention, sometimes a subscript such as W1 may be written as a non-subscript such as W1. When the difference is not emphasized, the meanings to be expressed are the same.

[0055] In order to make the technical problems, technical solutions and advantages to be solved by the present invention clearer, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0056] The embodiment of the present invention provides a highway toll station special event processing system based on virtual digital human, such as Figure 1-2 The schematic diagram of the highway toll station special event processing system based on virtual digital human is shown, and the system includes:

[0057] The image annotation training module obtains road image data from highway toll booths, extracts traffic object categories, locations, and time information, and combines weather and time period environment descriptions to determine the event type interval within the current sample set. It then selects event blocks distributed in risky time periods and regions, and generates an annotated event recognition map.

[0058] The contextual reasoning module collects traffic density, vehicle type, and behavior characteristics of the corresponding block based on the annotated event recognition map. It compares the change direction and correlation differences of multi-dimensional data according to the sampling time series, identifies the dynamic characteristics of events within the block, and generates a contextual reasoning state.

[0059] The dynamic image generation module is based on the situational reasoning state and dynamically adjusts the facial expressions, body movements, and voice intonation of the virtual digital human according to the real-time traffic flow and the urgency of the incident in the perception area, optimizing the interactive experience and response effect, and generating a dynamic image configuration table;

[0060] The event response module calls the dynamic image configuration table to classify the event block map, eliminate areas overlapping with non-emergency periods, compare the severity of the event with the response level, mark the time periods and areas that require priority, and generate the event intervention interval configuration table;

[0061] The correction marking module detects the changes in the traffic status recovery extent and virtual digital human performance feedback after event processing based on the event intervention interval configuration table, screens and groups event processing results that deviate from the expected blocks, marks the corresponding processing settings of the deviated blocks, and generates toll station event processing offset labels.

[0062] The annotated event recognition map includes event type classification labels, block time series indexes, risk period and area identification information; the situational reasoning status includes traffic density change trends, vehicle behavior pattern characteristics, and event correlation analysis results; the dynamic image configuration table includes digital human appearance adjustment parameters, voice intonation change range, and interactive behavior description; the event intervention interval configuration table includes intervention block number, target response level, and processing time window; the toll station event processing offset label includes traffic status recovery deviation level, virtual digital human performance change range, and processing setting offset mark.

[0063] Specifically, if Figure 2 、 3 As shown in Figure 2, the image annotation training module includes:

[0064] The object annotation submodule obtains highway toll booth road image data, identifies traffic object change sequences through the same block time series, combines category and location information, extracts event control blocks that meet dynamic change characteristics, and generates a set of annotation feature response block numbers;

[0065] First, real-time image data from highway toll booths is obtained. High-definition road surveillance video images of the area can be obtained using cameras deployed along the toll booths. The video images' timestamps accurately identify the temporal sequence of the image data. The captured images are then fed into an image recognition system, where a pre-trained model is used to identify traffic objects, such as vehicles, pedestrians, and traffic signs. After image data processing, the class of each traffic object (e.g., car, bus, truck) and its location within the image (e.g., center point coordinates, bounding box coordinates, etc.) are determined. Sequencing techniques can be used to analyze changes in objects, such as increases or decreases in the number of vehicles or changes in vehicle speed, based on image data over consecutive time periods. For example, suppose the number of vehicles in the image data increases from 10 to 20, with 15 cars and 5 trucks. The location coordinate data can be used to determine the distribution of vehicles at different locations on the road. By comparing the changes in data between the previous and next time periods, event control blocks that reflect dynamic changes can be extracted. The blocks are marked as annotated feature response blocks. The annotations will include the time and location of the event, as well as the types and changes of the traffic objects involved. By combining time series analysis with location data, a set of annotated feature response block numbers can be generated. The set of numbers represents the area and type of event that occurred within a specific time period. The set of numbers can provide data support for subsequent environmental factor analysis and risk identification.

[0066] The situation description submodule calls the labeled feature response block number set, extracts weather, time period, and traffic flow data, identifies the correlation between environmental variables and event occurrence, and selects block numbers and time periods that are both within the risk period and region to generate a risk section identification set;

[0067] Weather, time period, and traffic flow data are obtained from multiple data sources. The data comes from traffic monitoring systems, weather forecast services, and road sensors. Weather data includes temperature, humidity, precipitation, wind speed, etc. Time period data refers to the current time period, such as the morning rush hour, evening rush hour, or holidays. Traffic flow data is obtained through sensors or GPS devices, including the number of vehicles on the road and their speed during a specific time period. The correlation between different environmental variables and the occurrence of events is calculated through algorithms. For example, suppose that during a certain time period (such as the morning rush hour), the traffic flow at a highway toll station suddenly increases and a congestion incident occurs at the same time. The data will be correlated and it will be identified whether the specific environmental conditions during the time period are closely related to the occurrence of the incident. Through correlation analysis, specific block numbers in risky time periods and areas can be screened out. Assuming that during a specific period of time (such as the morning rush hour from 7 to 8 a.m.), weather conditions deteriorate (precipitation exceeds 5 mm), traffic flow reaches its maximum, congestion is severe, and the traffic accident rate is high, then this period and area will be identified as a risk period, and its corresponding block number and time period will be included in the risk section identification set. Generating a risk section identification set will provide a basis for subsequent traffic forecasting and risk control.

[0068] The event block determination submodule compares the traffic density and the changing trend of the time series nodes within the block based on the risk section identification set, matches the abnormal event benchmark range, extracts the valid blocks and spatial locations, and generates a marked event identification map;

[0069] Traffic density data is obtained for each block. Traffic density can be calculated by dividing the number of vehicles in a block by the block's area. For example, if the traffic density in a block reaches 10 vehicles per square meter during a certain period, significantly higher than the 3 vehicles per square meter during the same period, the current traffic density is compared with historical data to analyze the changing trends at the time series nodes. For example, if the traffic density in a block continues to rise and exceeds the historical maximum, this is considered an abnormal event. The current traffic density data is compared with a set abnormal event baseline range, which is statistically derived from historical data. For example, if the maximum allowable traffic density is set at 8 vehicles per square meter, any increase in traffic density beyond this range is marked as an abnormal event. This process extracts valid blocks and their spatial locations, and generates a labeled event identification map for them. For example, if the traffic density in a block reaches the warning value during a certain period, a labeled event identification map is generated containing information such as the block number, the time of occurrence, the traffic density, and the changing trend, helping decision makers quickly identify potential traffic problems and risk areas.

[0070] Specifically, if Figure 2 、 4 As shown, the situational reasoning module includes:

[0071] The traffic feature collection submodule collects traffic density and vehicle behavior feature readings based on the annotated event recognition map, unifies the sampling time series, removes anomalies and repairs missing segments in the block data, extracts valid feature samples, and generates a basic set of block feature monitoring;

[0072] By collecting readings of traffic density and vehicle behavior characteristics and performing data processing, the collected data is unified into a time series, time-aligned, and synchronized using sampling intervals. This allows accurate capture of traffic data for each time period. To ensure data accuracy, the collected raw data is subjected to anomaly removal and missing segments repair. Anomalous data can be identified by calculating the mean and standard deviation of the data, setting a certain deviation threshold (for example, data values ​​greater than ±3 standard deviations from the mean are considered anomalous). During processing, missing data segments are repaired through interpolation. Common methods include linear interpolation or interpolation based on adjacent data points. This fills in the missing data. For example, if the sampling interval is 10 minutes and traffic density is missing for a certain period, the data for the missing period can be obtained by linear interpolation using the traffic density values ​​of the preceding and following periods (e.g., 10 vehicles / minute in the preceding period and 20 vehicles / minute in the following period). This repaired dataset is used to generate the basic set of block feature monitoring data for subsequent analysis.

[0073] The behavior pattern comparison submodule calls the basic set of block feature monitoring, extracts the traffic density and vehicle behavior change values ​​in adjacent time periods, compares the multi-dimensional feature change directions and correlation differences, identifies coupling features, and generates a coupling feature coefficient set;

[0074] A multi-dimensional analysis of traffic density and vehicle behavior change values ​​is conducted. Two adjacent time periods are selected and the changes in traffic density and vehicle behavior data for these two periods are calculated. Assuming that the traffic density in the first period is 15 vehicles / minute and the second period is 20 vehicles / minute, the change in traffic density is 5 vehicles / minute. The behavior change value can be obtained by calculating the changes in vehicle driving status (such as changes in acceleration and braking frequency). The direction of multi-dimensional feature changes and associated differences are compared to analyze how the changes affect traffic flow. For example, if the increase in traffic density is accompanied by an increase in vehicle acceleration frequency, it indicates good road conditions, while the opposite indicates traffic congestion. In this way, coupling features are identified, which affect traffic flow or driving speed, thereby generating a set of coupling feature coefficients. This coefficient set is used to characterize the strength of the relationship between different features.

[0075] The contextual reasoning modeling submodule uses the coupling characteristic coefficient set, selects stable trend blocks, extracts the traffic density change amplitude and vehicle behavior change rate, combines the sampling interval, sequence position and number of sample points, calculates the traffic density change amplitude, and maps it with the spatial range and time node of the event block to generate the contextual reasoning state;

[0076] Select blocks with stable trends, extract the amplitude of traffic density changes and the rate of change of vehicle behavior, select the calculation method of the amplitude of traffic density changes, and set the initial value and change value of traffic density. For example, assuming that the traffic density is 10 vehicles / minute and 25 vehicles / minute in two time periods, the amplitude of change is 15 vehicles / minute. Combined with the number of sample points and the sampling interval, calculate the amplitude of traffic density changes in this period, and then analyze it through the rate of change of vehicle behavior data. Assuming that the rate of change of vehicle speed is 5 kilometers / hour in the same time period, the rate of change of behavior in this period is 5 kilometers / hour. Based on the data, combined with the number of sample points and the interval time, calculate the amplitude of change, and map the results with the spatial range and time nodes of the event block to ensure that the background before and after the event can be accurately inferred. The final generated situational reasoning state can be used to predict future changes in traffic patterns and provide support for decision-making;

[0077] The change range of traffic density is calculated using the formula:

[0078]

[0079] Calculation is performed, where ΔD represents the change in traffic density, ρ i represents the traffic density of the data point at time i, ρ i-1 represents the traffic density of the data point at time i-1, Q represents the sampling interval, and N represents the number of consecutive data points at each time;

[0080] Traffic density fluctuation refers to the magnitude of fluctuation or change in traffic density within a certain time interval. It reflects how the density of vehicles on a specific road or area changes over time, and is expressed as the change in traffic density per unit time. By calculating the absolute value of the difference between the traffic density at each sampling point and the previous moment, summing the differences, and dividing by the sampling interval and the number of sample points, the traffic density fluctuation can be calculated. This indicator helps assess traffic flow instability and reveals changing trends in traffic flow, providing an important reference for traffic management and scheduling.

[0081] Traffic density change (ΔD) calculation process:

[0082] Traffic density (ρ i ): Traffic density data obtained at each sampling time i by traffic flow monitoring equipment (such as geomagnetic sensors, video surveillance, radar detectors, etc.), in units of vehicles / km;

[0083] Sampling interval (Q): set by the data acquisition system, 1 second, 5 seconds or 10 seconds, indicating the time interval between two consecutive data collections;

[0084] Number of sample points (N): The number of consecutive data points selected within the stable trend block, usually 30, 60, 120, etc. The specific number is determined based on data stability and analysis requirements;

[0085] Taking a certain stable trend block as an example, assuming that the sampling interval Q = 5 seconds and the number of sample points N = 60, the traffic density data collected in this block are: ρ1 = 20 vehicles / km, ρ2 = 22 vehicles / km, ..., ρ 60 =18 vehicles / km;

[0086] Calculate the change in traffic density:

[0087]

[0088] Here, ρ0 is set to the value of ρ1;

[0089] Assume that the calculation is:

[0090] The results show that the average change in traffic density per unit time is 0.4 vehicles / km / second;

[0091] In the formula, the unit of traffic density is vehicles / km, the unit of sampling interval is seconds, and the number of sample points is dimensionless. To ensure the consistency of the units in the formula calculation, the traffic density needs to be converted to vehicles / km / second to match the unit of sampling interval. Through the above calculation process, the unit of traffic density change range is obtained as vehicles / km / second, ensuring the consistency of dimensions.

[0092] The weight parameters and adjustment coefficients are set based on:

[0093] Sampling interval (Q): The sampling interval is set based on the capabilities of the data collection system and analysis requirements. A shorter sampling interval can capture more detailed traffic changes but increases the complexity of data processing, while a longer sampling interval ignores some short-term traffic fluctuations.

[0094] Number of sample points (N): The number of sample points is selected based on the stability of the data and the analysis requirements. A larger number of sample points can improve the stability and accuracy of the calculation, but requires longer calculation time. A smaller number of sample points will lead to unstable calculation results.

[0095] Traffic density (ρ i Traffic density changes are affected by many factors, such as road conditions, traffic signal control, and weather conditions. Through real-time monitoring and historical data analysis, key factors affecting traffic density can be identified, providing a basis for traffic management and optimization.

[0096] Through the above process, the accuracy and rationality of the formula calculation are ensured, which provides a basis for the subsequent calculation of situational reasoning indicators.

[0097] Specifically, if Figure 2 、 5 As shown, the dynamic image generation module includes:

[0098] The environmental perception submodule extracts traffic feature data at each layer based on contextual reasoning. Combining real-time traffic flow with incident urgency, it analyzes the changing rhythm of traffic features and the fluctuation amplitude of environmental variables, determines whether they are within the matching threshold range, selects matching number sequences, and generates image adjustment warning number groups.

[0099] First, traffic characteristic data for each layer is extracted and combined with real-time traffic flow and incident urgency to analyze the rhythm of traffic characteristic changes and the magnitude of fluctuations in environmental variables. In practice, data from each layer, such as traffic density, vehicle speed, and traffic volume, is first extracted and dynamically adjusted for analysis using real-time data streams. This process involves calculating the magnitude of traffic density fluctuations within a specific time period. For example, if traffic density changes from 15 vehicles per minute to 30 vehicles per minute within a specific time period, the magnitude of the change is 15 vehicles per minute. Environmental variables such as temperature and weather also influence traffic characteristic fluctuations. Therefore, external factors need to be weighted during data analysis. For example, for every 5°C change in temperature, traffic density increases by 5%. By defining matching thresholds, a reasonable standard is established to determine whether traffic fluctuations are normal. For example, if the traffic density changes by more than 20 vehicles per minute, it is considered abnormal. Matching sequence numbers are screened and generated into image adjustment warning number groups. Finally, a qualified warning message is output, which can be used to make immediate adjustments to traffic conditions.

[0100] The appearance adjustment submodule adjusts the warning number group according to the image, selects the traffic characteristics and event urgency information of the corresponding time period, identifies the overlapping time segment of the two types of data, and generates an appearance adjustment plan;

[0101] Based on the warning number group, the time period of the emergency event is identified. Traffic characteristics within the time period and the urgency of the event need to be synchronized through time series processing. For example, if traffic density suddenly increases and an emergency occurs within a time period, it is caused by the deterioration of traffic conditions. In this case, the data needs to be overlaid and analyzed to find the corresponding intersecting time segments. The overlapping time segments are then calibrated to generate an appearance adjustment plan. This plan includes optimization measures for traffic control and road condition adjustments. For example, the urgency of the incident determines whether to temporarily close a road or increase traffic diversion measures.

[0102] The voice feedback derivation submodule calls the appearance adjustment plan, compares the traffic characteristics of the section with the incident urgency data, extracts the rhythm overlap locations, groups and summarizes the voice feedback differences of each combination, classifies them into response levels, summarizes the level change trajectory, and generates a dynamic image configuration table;

[0103] Based on the time period set in the appearance adjustment plan, traffic characteristic data and event urgency data for the corresponding section are extracted. These two types of data are then synchronously compared to identify locations where their rhythms overlap. For example, if changes in traffic density overlap with the time of an emergency during a certain period, the two data points have the same rhythm. The overlapping locations are then grouped, and the differences in voice feedback for each combination are summarized. These feedback differences are then classified into response levels based on preset criteria. For example, if traffic density reaches 30 vehicles per minute and the emergency is at high risk, the response level for that period is Level 1 (the highest level). Based on this information, the level change trajectory is summarized, ultimately generating a dynamic image configuration table for subsequent adjustments and optimization of traffic control measures.

[0104] Specifically, if Figure 2 、 6 As shown, the event response module includes:

[0105] The trend classification submodule calls the dynamic image configuration table to extract the traffic change direction and amplitude values ​​in the event block map, classifies the trend type according to the continuity and gradient change judgment, eliminates the block numbers that overlap with the non-emergency period, and generates a set of responsive trend blocks;

[0106] To extract the direction and magnitude of traffic changes in the event block map, we first need to obtain the direction and magnitude of each traffic event. These values ​​represent the direction and magnitude of changes in traffic density, flow rate, and speed within a specific time period. For example, if traffic density increases from 10 vehicles / minute to 50 vehicles / minute during a certain period, with a magnitude of 40 vehicles / minute, the direction is increasing. The trend of traffic characteristics is analyzed based on the continuity and gradient of traffic changes. Gradient change refers to the rate of change of traffic density within a continuous time period. A faster rate of change indicates an emergency. By comparing the magnitude of change in the current time period with the previous one, for example, if the traffic density increased by 5 vehicles / minute in the previous period and by 15 vehicles / minute in the current period, we can determine that the current rate of change is higher, thus identifying a sharp change in the trend. This judgment also needs to be based on the emergency response standards in the actual scenario, eliminating some block numbers whose time periods overlap with non-emergency periods. For example, if the traffic changes are small or the event risk is low during the period, the block number should be filtered. After screening, a responsive trend block set is finally generated. The block set contains time period blocks with obvious trend changes and meeting emergency response conditions.

[0107] The priority comparison submodule calls the set of responsive trend blocks, collects the current event severity and response level standard range of the block, compares the event severity and response lower limit value by block, sorts the block numbers, and obtains the priority response block sequence;

[0108] The current event severity and response level standard range for each block are collected. For example, if a traffic accident in a block is classified as high risk (event severity level 5), and the response lower limit specified by the standard range is level 3, it is necessary to determine whether the block meets the response conditions. In practical applications, severity can be assessed by analyzing parameters such as the impact range of the traffic accident, the number of casualties, and the duration of traffic disruption. The event severity of each block is compared with the standard lower limit. This process can be completed by comparing the event severity value of each block with the set response lower limit (for example, if the event severity of block 1 is 5 and the response lower limit is 3, then the block meets the conditions). After the comparison is completed, all eligible blocks are sorted by priority. For example, if multiple blocks meet the response criteria, they are sorted according to the event severity, with higher severity blocks being responded to first. This results in a priority response block sequence, which is used in subsequent decision-making to determine which blocks should be responded to and handled first.

[0109] The intervention interval identification submodule extracts the corresponding response time interval according to the block number in the priority response block sequence, calculates the event intervention interval configuration value, and sets the response level and processing time interval configuration according to the continuous deviation of the event, and generates the event intervention interval configuration table;

[0110] Extract the response period data of each block. The data includes information such as the duration of the event, traffic density, and flow rate. For example, the response period of a certain block is from 12:00 to 14:00. During this period, the traffic density gradually increases from 10 vehicles / minute to 60 vehicles / minute, and the flow rate fluctuates greatly during this period. Calculate the event intervention interval configuration value. The setting of the intervention interval needs to take into account the continuous deviation of the event. If the traffic accident or congestion lasts too long, it is necessary to increase the intervention intensity and issue an early warning. The calculation of the intervention interval configuration value includes analyzing the trend of traffic data changes in the time period and adjusting it in combination with the traffic model. Assuming that the traffic density of the event fluctuates by more than 50% between 12:00 and 14:00, the intervention configuration value can be set to a higher value. Based on this, the response level and processing period configuration of the intervention interval can be set according to the severity of the event. Higher-level events require longer response periods and higher intervention intensities. Finally, an event intervention interval configuration table is generated to guide subsequent traffic control and intervention measures.

[0111] The event intervention interval configuration value uses the formula:

[0112]

[0113] Calculation is performed, where I represents the event intervention interval configuration value, T1 represents the end time corresponding to the block number, T0 represents the start time corresponding to the block number, n represents the number of intervals in the response period, and δ j represents the continuous offset of the jth response period, P represents the basic response offset, α j represents the weight of the jth response period;

[0114] The event intervention interval configuration value indicates that during the event response process, appropriate adjustments need to be made based on the duration, offset, and weight of the response period. This value comprehensively considers the duration of the event, the offset of each response period, and the predetermined basic response time, reflecting the intervention intensity that should be applied during the entire event handling process. A larger value indicates a higher adjustment requirement for the intervention interval, requiring more resources or time to respond to ensure that the event is properly handled.

[0115] In this formula, I is the configuration value of the event intervention interval, which involves the calculation and acquisition of multiple parameters. The following describes in detail the acquisition process, quantification method, and dimensional unification of each parameter:

[0116] Get the start time and end time: T1 and T0 represent the end time and start time corresponding to the block number, respectively. The timestamp of each event block is obtained through the real-time data of the event monitoring system. When obtaining the timestamp, ensure that the data source of the timestamp is accurate and synchronized to avoid the impact of system clock deviation. The timestamp unit is seconds or milliseconds, but to ensure consistency and simplify calculations, all time data will be converted to a unified unit (for example, minutes) before processing;

[0117] Dimensional unification process: Assuming that T1 and T0 are recorded in minutes, the duration calculated by T1-T0 is minutes. If the time unit is seconds, it needs to be divided by 60 to convert to minutes;

[0118] Calculate the number of response time intervals: n represents the number of response time intervals. Based on the actual event type and response strategy, the response time intervals are divided according to the importance or urgency of the event. The number of response time intervals is determined by historical data or event characteristics.

[0119] Quantification process: This value is obtained by counting the response time intervals of historical events or by manually setting the number of response time periods. If it is set to 3 time periods, then n = 3;

[0120] Calculation of the continuous offset: δ i represents the continuous offset of the i-th response period, and the continuous offset of each response period δ iIt is calculated based on the difference between the actual response time and the expected response time. The sign and size of the offset reflect the advance or delay of the time.

[0121] Quantification process: For each response period, the actual time difference is calculated by comparing the real-time data with the preset value. For example, if the preset period is 10 minutes, but the actual response time is 12 minutes, then δ i = 2 minutes, indicating that the time period has timed out; if the actual response time is 8 minutes, then δ i = -2 minutes, indicating early end;

[0122] Dimensional unification process: The unit of the continuous offset must be the same as T1-T0, which is minutes. All offset data should be converted to a unified unit (for example, minutes) to avoid unit inconsistency issues.

[0123] Determination of basic response offset: P is the basic response offset. This parameter reflects the basic response time and is derived from empirical data or based on the processing speed and emergency response settings of different systems.

[0124] Quantification process: This value can be calculated by analyzing historical data to determine the response time deviation under standard conditions. A typical value is 5 minutes, representing the standard deviation when there are no special events. To ensure consistent dimensionality, P is expressed in minutes.

[0125] Allocation of response period weights: α i is the weight of the i-th time period, which is determined by factors such as the importance and urgency of the event. Important time periods have larger weights, reflecting their higher priority and should be given more response resources. The weight value of each time period is determined through experience or historical data analysis;

[0126] Quantification process: For each response period, a weight is assigned based on the nature of the incident. For example, the most urgent period is given a larger weight (such as 1.5), while the lower priority period is given a smaller weight (such as 1.0). All weights need to be normalized to ensure that the total weight is within a reasonable range.

[0127] Dimensional unification and normalization: In the formula, all time-related parameters (T1-T0 and δ i ) all use minutes as the unit. To ensure that all parameters can be calculated collaboratively, it is first necessary to unify the data from different sources into minutes.

[0128] Normalization: In order to avoid the difference in the magnitude of each parameter from having too much impact on the final result, δ i and α i Normalize the data, for example, δ iNormalize by dividing by the maximum value to ensure that the range of all offset data is consistent, and the weight α i You can also normalize the weights of each period to ensure that the sum of the weights is 1;

[0129] After obtaining and quantifying each parameter, you can substitute it into the formula for calculation. Assuming that in an actual event, the following parameter values ​​have been obtained:

[0130] T1 = 10:30, T0 = 10:00, so T1 - T0 = 30 minutes;

[0131] n = 3 response periods;

[0132] The duration offset of each period is δ1=2, δ2=-1, δ3=0 minutes;

[0133] Basic response offset P = 5 minutes;

[0134] The weights of each time period are α1=1.2, α2=1.0, and α3=1.5;

[0135] Substituting into the formula:

[0136] The result shows that the event intervention interval configuration value is 6.75, which means that during the response process of this event, appropriate intervention adjustments are needed to ensure that the handling of the event complies with the predetermined response strategy.

[0137] Specifically, if Figure 2 、 7 As shown, the correction marking module includes:

[0138] The traffic fluctuation screening submodule extracts the block event processing end time according to the event intervention interval configuration table, identifies the traffic status data after the corresponding time, analyzes the difference between the recovery amplitude and the processing target value, filters out the blocks that exceed the recovery deviation threshold, and generates a traffic status deviation block set;

[0139] Extract the end time of the event processing for each block and identify traffic status data after that time based on the time point. For example, if the processing of a traffic incident ends at 3:00 PM, extract traffic status data between 3:00 PM and 3:30 PM. This data includes information such as traffic density, vehicle speed, and volume. Analyze the difference between the recovery amplitude and the target value. The recovery amplitude refers to the degree of improvement in traffic conditions after the incident processing. For example, if the traffic density decreases from 50 vehicles per minute to 30 vehicles per minute after the incident processing, the recovery amplitude is 20 vehicles per minute. Calculate the difference between this recovery amplitude and the set target value. For example, if the target value is set to 25 vehicles per minute, the recovery deviation is 5 vehicles per minute. By comparing the recovery amplitude with the target value, filter out blocks that exceed the recovery deviation threshold. For example, if the deviation threshold is set to 3 vehicles per minute, any blocks with a recovery amplitude deviation of more than 3 vehicles will be filtered out, generating a set of traffic status deviation blocks. These blocks represent areas where traffic recovery significantly deviates from the target value.

[0140] The performance difference measurement submodule extracts the performance feedback change readings of the virtual digital human in the corresponding block based on the traffic state deviation block set, analyzes the performance change amplitude, and classifies them according to the ratio of the performance change value to the traffic state recovery deviation value. It then filters out the blocks with inconsistent responses and generates a traffic-performance imbalance distribution group.

[0141] For each block, the corresponding virtual human performance feedback change readings are extracted. These feedback change readings are related to traffic flow, vehicle behavior, or the efficiency of traffic incident handling. For example, if the virtual human's feedback in a certain block is "delayed response" or "inconsistent behavior," the performance change needs to be quantified by analyzing the relationship between various traffic characteristic data and the virtual human's feedback data. The magnitude of the performance change is analyzed. The performance change magnitude refers to the virtual human's response to traffic issues changing from "low" to "high," and this change directly affects subsequent incident responses. Classification is performed by calculating the ratio of the performance change value to the traffic state recovery deviation value. A threshold is set. When the ratio exceeds a certain standard, it indicates a significant imbalance between traffic conditions and the virtual human's performance. For example, if the traffic state recovery deviation is 5 vehicles / minute and the virtual human's performance change is level 3, the ratio is 5 / 3 = 1.67. If the imbalance threshold is set to 1.5, the block is classified as unbalanced. This method can filter out blocks with inconsistent responses and generate a traffic-performance imbalance distribution group. This data set will help identify areas of poor performance in traffic management.

[0142] The processing offset identification submodule extracts the corresponding processing parameters based on the traffic-performance imbalance distribution group and compares them with the processing targets in the configuration table, marks the blocks with set offsets and the set parameters, and generates a toll station event processing offset label;

[0143] Extract the processing parameters of each block. The processing parameters involve event processing time, intervention intensity, traffic control measures, etc. For example, in a certain block, the processing parameters include a delay time of 10 minutes, a high intensity of traffic diversion measures, etc. Extract the corresponding processing targets in the configuration table, compare them with the processing parameters of the actual block, and determine whether there is an offset. For example, the target processing time set in the configuration table is 5 minutes, while the actual processing time is 10 minutes, indicating an offset of 5 minutes. Mark the block with the set offset and its set parameters. For example, if the processing time offset of a block exceeds 5 minutes, the block will be marked as "processing offset". By calibrating the processing offset of each block, a toll station event processing offset label is finally generated. The label will help to optimize the subsequent processing process of traffic events and provide valuable data support for management decisions.

[0144] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. The highway toll station special event processing system based on virtual digital human is characterized by: The system comprises: The image annotation training module obtains road image data from highway toll booths, extracts traffic object categories, locations, and time information, and combines weather and time period environment descriptions to determine the event type interval within the current sample set. It then selects event blocks distributed in risky time periods and regions, and generates an annotated event recognition map. The contextual reasoning module collects traffic density, vehicle type, and behavior characteristics of the corresponding block based on the annotated event recognition map, compares the change direction and correlation differences of the multi-dimensional data according to the sampling time series, identifies the dynamic characteristics of events in the block, and generates a contextual reasoning state; The dynamic image generation module dynamically adjusts the facial expressions, body movements, and voice intonation of the virtual digital human based on the contextual reasoning state and the real-time traffic flow and urgency of the incident in the perception area to optimize the interactive experience and response effect, and generates a dynamic image configuration table; The event response module calls the dynamic image configuration table, classifies the event block map, eliminates areas overlapping with non-emergency time periods, compares the severity of the event with the response level, marks the time periods and areas that require priority processing, and generates an event intervention interval configuration table.

2. The highway toll station special event processing system based on virtual digital human according to claim 1 is characterized by: The annotated event recognition map includes event type classification labels, block time series indexes, risk period and area identification information; the situational reasoning state includes traffic density change trends, vehicle behavior pattern characteristics, and event correlation analysis results; the dynamic image configuration table includes digital human appearance adjustment parameters, voice tone change range, and interactive behavior descriptions; the event intervention interval configuration table includes intervention block numbers, target response levels, and processing time windows.

3. The highway toll station special event processing system based on virtual digital human according to claim 1 is characterized by: The image annotation training module includes: The object annotation submodule obtains highway toll booth road image data, identifies traffic object change sequences through the same block time series, combines category and location information, extracts event control blocks that meet dynamic change characteristics, and generates a set of annotation feature response block numbers; The situation description submodule calls the annotated feature response block number set, extracts weather, time period and traffic flow data, identifies the correlation between environmental variables and event occurrence, filters the block numbers and time periods that are both within the risk period and area, and generates a risk section identification set; The event block determination submodule compares the traffic density and the changing trend of the time series nodes in the block according to the risk section identification set, matches the abnormal event benchmark range, extracts the valid blocks and spatial positions, and generates a marked event recognition map.

4. The highway toll station special event processing system based on virtual digital human according to claim 3 is characterized by: The situational reasoning module includes: The traffic feature acquisition submodule collects traffic density and vehicle behavior feature readings based on the annotated event recognition map, unifies the sampling time series, removes anomalies and repairs missing segments of the block data, extracts valid feature samples, and generates a basic set of block feature monitoring; The behavior pattern comparison submodule calls the block feature monitoring basic set, extracts the traffic density and vehicle behavior change values ​​in adjacent time periods, compares the multi-dimensional feature change directions and correlation differences, identifies coupling features, and generates a coupling feature coefficient set; The situational reasoning modeling submodule calls the coupling characteristic coefficient set, selects the stable trend block, extracts the traffic density change amplitude and vehicle behavior change rate, combines the sampling interval, sequence position and number of sample points, calculates the traffic density change amplitude, and maps it with the spatial range and time node of the event block to generate the situational reasoning state.

5. The highway toll station special event processing system based on virtual digital human according to claim 4 is characterized by: The traffic density variation range is calculated using the formula: Calculation is performed, where ΔD represents the change in traffic density, ρ i represents the traffic density of the data point at time i, ρ i-1 represents the traffic density of the data point at the i-1th moment, Q represents the sampling interval, and N represents the number of consecutive data points at each moment.

6. The highway toll station special event processing system based on virtual digital human according to claim 4 is characterized by: The dynamic image generation module includes: The environmental perception submodule extracts traffic feature data at each layer based on the contextual reasoning state, combines real-time traffic flow with the urgency of the incident, analyzes the changing rhythm of traffic features and the fluctuation amplitude of environmental variables, determines whether they are within the matching threshold range, selects the number sequence that meets the conditions, and generates an image adjustment warning number group; The appearance adjustment submodule selects the traffic characteristics and event urgency information of the corresponding time period based on the image adjustment warning number group, identifies the overlapping time period of the two types of data, and generates an appearance adjustment plan; The voice feedback derivation submodule calls the appearance adjustment scheme, compares the traffic characteristics of the section with the event urgency data, extracts the rhythm overlap positions, groups and summarizes the voice feedback differences of each combination, and classifies them into response levels, summarizes the level change trajectory, and generates a dynamic image configuration table.

7. The highway toll station special event processing system based on virtual digital human according to claim 6 is characterized by: The event response module includes: The trend classification submodule calls the dynamic image configuration table, extracts the traffic change direction and amplitude value of the area in the event block map, classifies the trend type according to the continuity and gradient change judgment, eliminates the block numbers that overlap with the non-emergency period, and generates a responsive trend block set; The priority comparison submodule calls the respondable trend block set, collects the current event severity and response level standard range of the block, compares the event severity and response lower limit value by block, sorts the block numbers, and obtains the priority response block sequence; The intervention interval identification submodule extracts the corresponding response time interval according to the block number in the priority response block sequence, sets the response level and processing time interval configuration according to the continuous deviation of the event, and generates an event intervention interval configuration table.

8. The highway toll station special event processing system based on virtual digital human according to claim 7 is characterized by: The event intervention interval configuration value adopts the formula: Calculation is performed, where I represents the event intervention interval configuration value, T1 represents the end time corresponding to the block number, T0 represents the start time corresponding to the block number, n represents the number of intervals in the response period, and δ j represents the continuous offset of the jth response period, P represents the basic response offset, α j Represents the weight of the j-th response period.

9. The highway toll station special event processing system based on virtual digital human according to claim 1 is characterized by: The system also includes a correction marking module: The correction marking module detects the traffic state recovery extent and the change in the performance feedback of the virtual digital human after the event processing according to the event intervention interval configuration table, screens and groups the event processing results that deviate from the expected block, marks the corresponding processing settings of the deviated block, and generates a toll station event processing offset label; The toll booth event processing offset tag includes the traffic state recovery deviation level, the virtual digital human performance change range, and the processing setting offset mark.

10. The highway toll station special event processing system based on virtual digital human according to claim 9 is characterized by: The correction marking module includes: The traffic fluctuation screening submodule extracts the block event processing end time according to the event intervention interval configuration table, identifies the traffic status data after the corresponding time, analyzes the difference between the recovery amplitude and the processing target value, screens the blocks that exceed the recovery deviation threshold, and generates a traffic status deviation block set; The performance difference determination submodule extracts the performance feedback change readings of the virtual digital human in the corresponding blocks based on the traffic state deviation block set, analyzes the performance change amplitude, and classifies them according to the ratio of the performance change value to the traffic state recovery deviation value, filters out the response inconsistent blocks, and generates a traffic-performance imbalance distribution group; The processing offset identification submodule extracts corresponding processing parameters according to the traffic-performance imbalance distribution group and compares them with the processing targets in the configuration table, marks the blocks and setting parameters for setting offsets, and generates a toll booth event processing offset label.

Images