Expressway traffic incident detection method and system based on portal data

Abstract

The invention discloses a portal data-based highway traffic incident detection method and system, and relates to the technical field of traffic intelligent management, and the incident detection method comprises the steps: obtaining the historical traffic data of an ETC portal, constructing a road network topology and traffic flow reference model, and carrying out the dynamic flow prediction through combining the real-time portal data flow, the system can identify traffic events and carry out accurate positioning. According to the method, a flow deviation accumulation and space-time aggregation mechanism is adopted, and video monitoring linkage and multi-channel information release are combined, so that real-time confirmation, check and release decision of events can be realized; according to the method, the timeliness and accuracy of event response are remarkably improved, the information issuing process is optimized, the robustness of the system is enhanced, and the false alarm rate is reduced. Sufficient technical support is provided for safe and efficient operation of the expressway, and the emergency management and traffic control level is improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent traffic management technology, and more specifically to a method and system for detecting highway traffic incidents based on gantry data. Background Technology

[0002] As vital transportation arteries, highways are crucial for both operational efficiency and safety. Rapid detection and efficient handling of traffic incidents (such as accidents, congestion, and vehicle breakdowns) are key to ensuring highway traffic flow and reducing secondary accidents. Traffic incident monitoring methods can be broadly categorized into two types: direct detection methods and indirect detection methods.

[0003] Direct detection methods use computer vision models to directly analyze target behavior and spatiotemporal features in videos, achieving pixel-level recognition of events such as accidents and violations. Their advantages lie in millisecond-level response and high-precision event classification, and they support multi-target trajectory tracking. However, the model is heavily dependent on lighting conditions; the false detection rate spikes in heavy rain or strong light, and it requires high-performance computing power.

[0004] Indirect detection methods rely on parameters such as traffic flow, speed, and occupancy collected by sensor networks. They then use traditional models or machine learning to analyze anomaly patterns in the data and indirectly infer traffic events. Their advantages include low cost, wide coverage, and insensitivity to weather interference. However, they suffer from significant latency, difficulty in distinguishing event types, and require road construction for installation and maintenance.

[0005] Overall, traditional event detection methods rely on video surveillance, loop detectors, and manual inspections, which have inherent drawbacks such as wide monitoring blind spots, high construction and maintenance costs, long detection delays, and susceptibility to severe weather, thus requiring technological improvements.

[0006] Therefore, it is necessary to propose a highway traffic incident detection method and system based on gantry data to solve the above problems. Summary of the Invention

[0007] To address the shortcomings of existing technologies, this invention provides a method and system for detecting highway traffic incidents based on gantry data.

[0008] In recent years, with the comprehensive coverage of the ETC gantry system on national highways, an unprecedented massive and high-precision spatiotemporal data source has been provided for traffic condition perception. Each gantry can record the identity and timestamp information of passing vehicles in real time, forming vehicle trajectory data covering the entire road network. Gantry data provides a new approach for indirect detection methods, and reconstructing vehicle travel trajectories using existing ETC infrastructure will achieve zero additional hardware costs. Combining multi-timescale traffic flow analysis with adaptive benchmark models for traffic flow prediction enables efficient and robust detection and information dissemination of traffic incidents.

[0009] To achieve the above objectives, the present invention specifically adopts the following technical solution:

[0010] A method for detecting highway traffic incidents based on gantry data includes the following steps:

[0011] c1. Obtain historical traffic data of all ETC gantries within the target area. Construct a directed gantry topology network based on device ID, real-time traffic data of spatial location, and driving direction. Generate a gantry adjacency table containing upstream and downstream relationships. Based on vehicle trajectory matching results, use iterative filtering to clean travel time samples, establish a traffic flow benchmark parameter library under free flow conditions, and simultaneously quantify the feature detection rate and false detection rate of each gantry device to form a static decision-making basis dataset.

[0012] c2. Integrate real-time gantry data streams from multiple time scale windows, capture instantaneous flow fluctuations through a base window, filter out random noise and extract short-term trends through a smoothing window, and calculate dynamic flow ratios using a reference window; fuse equipment health weights and historical benchmark parameters to predict downstream gantry expected flow ranges in real time, and dynamically adjust confidence interval boundaries based on upstream smoothed flow and travel time delays.

[0013] c3. Design a decision function based on the accumulation of flow deviation. When the real-time smoothed flow continuously deviates from the expected range, trigger the accumulation counting mechanism of loss amount or continuous zero flow duration. Perform spatiotemporal proximity clustering on abnormal points that exceed the threshold to generate discrete event clusters and assign them unique IDs. At the same time, start the cooling-off period clock to suppress repeated alarms for automatic tracking and location of abnormal events.

[0014] c4. Automatically link the location event to the road network video surveillance system, retrieve the optimal viewpoint image and overlay traffic parameter curves onto the human-machine interface; after the operation and maintenance personnel confirm the event type and severity, drive the decision tree strategy to generate hierarchical release instructions, push structured event information through multiple channels such as gantry information boards, vehicle-mounted ECT terminals, navigation platforms, and traffic management platforms, and record handling feedback to optimize detection thresholds and cooling-off period parameters.

[0015] Further, step c1 includes:

[0016] c11. Adjacency Graph Assembly: Construct a directed topology graph using ETC data, sort devices by driving direction and mileage marker to ensure upstream and downstream relationships, and introduce redundancy strategies to form an adjacency table;

[0017] c12. Free Flow Parameter Calibration: After constructing the adjacency list, a normal traffic spatiotemporal parameter benchmark is established for each pair of adjacent devices based on historical data. By extracting travel time samples and performing filtering analysis, the expected travel time and its fluctuation range are calculated.

[0018] c13. Equipment Quality Assessment: The health status of gantry equipment is assessed by analyzing the false negative rate and false positive rate, the detection rate and false positive rate are calculated to quantify the equipment performance, and a trust score is assigned to the equipment based on these indicators.

[0019] Further, step c2 includes:

[0020] c21. Time parameter settings: In order to capture traffic phenomena at different time scales, three time windows are set, including the base window, the smoothing window and the reference window;

[0021] c22. Interval State Calculation: Based on real-time data and reference time windows, the system calculates three types of traffic characteristics: real-time traffic, smooth traffic, and long-term traffic; through vehicle trajectory matching, the average travel speed of each path segment is calculated.

[0022] c23. Upstream and downstream relationship calculation: Calculate the flow ratio between each pair of adjacent devices, integrate real-time flow, historical flow ratio and estimated flow ratio to build a robust flow ratio model, and predict the upstream and downstream flow ratio at the current moment.

[0023] c24. Downstream flow range: After calculating the upstream and downstream flow ratio, predict the reasonable flow range of downstream equipment, and estimate the upper and lower limits of downstream flow by combining historical data and flow ratio model.

[0024] Further, step c3 includes:

[0025] c31. Abnormal state judgment: When the downstream smoothed flow deviates from the prediction range, the system will accumulate "cumulative loss" or "continuous zero flow" indicators based on the duration and the situation of zero flow. Once the threshold is exceeded, the current time window is marked as an abnormal state.

[0026] c32. Event Spatiotemporal Aggregation: Aggregates discrete anomalies through temporal and spatial proximity to form continuous anomaly events; if a new anomaly is adjacent to an existing event, the event state is updated; if not adjacent, it is marked as a new event and initialized.

[0027] c33. Event Cooling-Off Period: To avoid repeated alarms, a cooling-off period is set after an event occurs. If an anomaly is detected again during the cooling-off period, the event is considered to continue. If no anomaly is detected, the event ends, the cooling-off period is lifted, and the system reactivates the alarm mechanism.

[0028] Further, step c4 includes:

[0029] c41. Video surveillance linkage: After confirming an abnormal traffic event, the system automatically retrieves relevant video surveillance resources based on the event location, and retrieves video streams or historical clips.

[0030] c42. Human-computer interaction verification: The system pushes event-related data to the verification interface of the traffic monitoring center. The operation and maintenance personnel cross-compare the video, traffic curve and alarm parameters to complete the event authenticity determination and hierarchical management.

[0031] c43. Information Release Decision: After an event is confirmed, the system determines the scope and channels for information release based on the rules in the policy library; important events will be released through a wide range of channels, while minor events will only be released within a local scope.

[0032] c44. Closed-loop information feedback: The system pushes event information to multiple platforms through standardized interfaces, records the release status, and waits for maintenance personnel to update the event handling status.

[0033] A highway traffic incident detection system based on gantry data includes:

[0034] Historical benchmark construction module: Establishes a static benchmark model of the highway network based on historical ETC gantry data;

[0035] Real-time status perception and prediction module: responsible for processing real-time data streams from ETC gantry, dynamically monitoring and predicting traffic conditions;

[0036] Intelligent event detection and localization module: used to identify and locate abnormal traffic events on highways;

[0037] Information closed-loop processing module: used for the automated closed loop of event confirmation, information release and processing feedback.

[0038] Furthermore, the historical benchmark construction module is used to construct a topological adjacency graph describing the spatial relationships of gantry structures, calibrate the segment travel time benchmark parameters under normal free flow conditions, and evaluate the detection reliability of each gantry device.

[0039] The real-time status perception and prediction module calculates gantry flow and section speed in real time through multi-level time windows, and dynamically analyzes the upstream and downstream flow ratio based on traffic flow theory; predicts the reasonable expected flow range of the downstream gantry.

[0040] Furthermore, the event intelligent detection and location module continuously compares the downstream measured traffic flow with the predicted expected traffic flow range to trigger and accumulate dynamic anomaly indicators; when the anomaly reaches a preset threshold, the system performs spatiotemporal aggregation analysis and, combined with the event status management mechanism, locks down the road segment interval where the event occurred.

[0041] The information closed-loop processing module automatically links with video surveillance for rapid manual verification; after confirmation, it intelligently matches the release strategy according to the event attributes and pushes structured event information to multiple platforms through standardized interfaces; the entire process records the status and receives processing feedback.

[0042] (III) Beneficial Effects

[0043] The beneficial effects of this invention are as follows:

[0044] 1. This invention deeply integrates historical benchmark modeling with real-time dynamic perception to construct a robust highway traffic incident detection system. By accurately establishing road network topology, free flow parameter benchmarks, and equipment health models, it provides a solid basis for real-time analysis and decision-making. At the same time, by utilizing multi-timescale analysis and dynamic traffic prediction models, it achieves accurate perception of traffic conditions and intelligent identification of abnormal traffic flow.

[0045] 2. This invention employs a dynamic cumulative judgment mechanism based on traffic prediction deviation and an intelligent spatiotemporal aggregation strategy to achieve efficient event location and deduplication. By setting thresholds to trigger abnormal indicators and combining spatiotemporal proximity analysis for event aggregation and status management, the road segment intervals where events occur can be quickly and accurately located, effectively avoiding false alarms and duplicate alarms, and significantly improving detection efficiency and location accuracy.

[0046] 3. This invention establishes a complete closed-loop handling process of "detection-verification-release-feedback," significantly improving the timeliness of event response and the accuracy of information release. The system automatically links with video surveillance for rapid manual verification, intelligently decides the scope and channels of release based on a preset strategy library, and accurately pushes structured event information through multiple channels. Simultaneously, the system records the release status and receives handling feedback, realizing fully automated closed-loop management of the entire process from event discovery to handling, greatly improving the operational safety and traffic efficiency of highways. Attached Figure Description

[0047] Figure 1 This is a schematic diagram of the overall method of the present invention. Detailed Implementation

[0048] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0049] ETC gantry system data provides an unprecedented massive and high-precision spatiotemporal data source for traffic condition perception, requiring only the reuse of existing equipment. This invention's method, based on ETC gantry data, constructs a refined road network topology model, establishes adaptive traffic flow parameter benchmarks, and combines multi-timescale real-time data analysis with a closed-loop manual verification process to achieve efficient and robust detection and information dissemination of traffic events.

[0050] Example 1

[0051] Please see Figure 1 A method for detecting highway traffic incidents based on gantry data includes the following steps:

[0052] c1. Obtain historical traffic data for all ETC gantries within the target area. Construct a directed gantry topology network based on device ID, spatial location, and driving direction, generating a gantry adjacency table containing upstream and downstream relationships. Based on vehicle trajectory matching results, use iterative filtering to clean travel time samples, establish a traffic flow benchmark parameter library under free-flow conditions, and simultaneously quantify the feature detection rate and false detection rate of each gantry device to form a static decision-making dataset.

[0053] c2. Integrate real-time gantry data streams from multiple time scales. Capture instantaneous flow fluctuations through a base window, filter out random noise and extract short-term trends through a smoothing window, and calculate dynamic flow ratios using a reference window. Fuse equipment health weights with historical benchmark parameters to predict the expected flow range of downstream gantry in real time, and dynamically adjust the confidence interval boundaries based on upstream smoothed flow and travel time delay.

[0054] c3. Design a decision function based on accumulated flow deviation. When the real-time smoothed flow continuously deviates from the expected range, trigger an accumulation counting mechanism for the amount of loss or the duration of continuous zero flow. Perform spatiotemporal proximity clustering on anomalies exceeding the threshold to generate discrete event clusters and assign them unique IDs. At the same time, activate the cool-down period clock to suppress repeated alarms, thereby achieving automatic tracking and precise location of abnormal events.

[0055] c4. Automatically link the location event to the road network video surveillance system, retrieve the optimal viewing angle image, and overlay traffic parameter curves onto the human-machine interface. After the maintenance personnel confirm the event type and severity, the decision tree strategy is driven to generate hierarchical release instructions, pushing structured event information through multiple channels such as gantry information boards, vehicle-mounted ECT terminals, navigation platforms, and traffic management platforms. At the same time, the handling feedback is recorded to optimize the detection threshold and cool-down period parameters.

[0056] Example 2

[0057] Please see Figure 1 This embodiment is a further optimization based on embodiment 1. Specifically, step c1 includes:

[0058] c11. Adjacency Graph Construction: A directed topology graph is constructed using ETC data. Devices are sorted by driving direction and mileage marker to ensure upstream and downstream relationships. Redundancy strategies are introduced to enhance robustness, ultimately forming an adjacency list to support subsequent traffic flow analysis. First, define N gantry detection devices deployed along the road, denoted as the node set, specifically represented as:

[0059] V = {G1,G2,…,G}N}

[0060] Among them, G i This represents the i-th ETC detection device, whose attributes include device ID, device name, driving direction, and location (in terms of mileage marker p). i express).

[0061] Based on the driving direction attribute of each device, all devices are grouped by direction, and within each direction, they are grouped according to the mileage marker p. i The equipment is sorted from smallest to largest to ensure that the equipment sequence corresponds to the upstream and downstream relationships in the actual driving direction of the vehicle; the sorted equipment sequence is specifically represented as:

[0062] G1→G2→…→G N

[0063] To improve the robustness of the topology and avoid misjudgments caused by gantry signal failures, not only are the connection relationships between adjacent devices considered, but a two-level upstream and downstream redundancy strategy is also introduced. For each gantry device G... i Establish its connection with the upstream equipment of the first two levels G i-1 G i-2 and the subsequent two downstream equipment G i+1 G i+2 The directed adjacency relationship; specifically defined as:

[0064] Adj(G i )={(G j ,d ij )∣j∈{i-2,i-1,i+1,i+2},1≤j≤N}

[0065] Where, d ij Indicates device G i With G j The physical distance between them is defined as:

[0066] d ij =|p j -p i |

[0067] This adjacency relationship is directed, considering only connections in the direction of travel. A road detection equipment graph is constructed based on the adjacency relationships of the gantry equipment. Its graph structure is a directed graph, specifically defined as follows:

[0068] G = (V, E)

[0069] Where V is the set of device nodes, and E is the set of edges, consisting of all directed edges that satisfy the above adjacency rules, specifically defined as:

[0070] E = {(G i G j)∣G j ∈Adj(G i )}

[0071] Simultaneously record the distance weight d of each edge. ij This provides a data foundation for subsequent calculations of traffic flow baseline parameters. Based on the constructed adjacency graph structure, the adjacency relationships of all nodes are uniformly recorded in an adjacency table, specifically represented as follows:

[0072]

[0073] In this adjacency list structure, each element contains the device number, location, and distance information of the current gantry device and its upstream and downstream devices.

[0074] c12. Free Flow Parameter Calibration: After constructing the adjacency list, a normal traffic spatiotemporal parameter benchmark is established for each pair of adjacent devices based on historical data. By extracting travel time samples and performing filtering analysis, the expected travel time and its fluctuation range are calculated.

[0075] It is necessary to establish a traffic spatiotemporal parameter benchmark for each pair of adjacent devices in the road network under normal (free-flow) conditions, mainly including the expected travel time and its fluctuation range estimation. This is achieved by pairing vehicle information from upstream and downstream detection records in long-term historical data, reconstructing vehicle trajectories, and extracting travel time samples. For upstream device G... i Downstream equipment G j For each vehicle v, calculate its corresponding travel time sample based on its trajectory:

[0076]

[0077] in, and Vehicle v is detected by the equipment G. i and G j The detection time at the location.

[0078] For each pair of adjacent devices, histogram analysis and 3-Sigma iterative filtering are performed on their travel time samples to remove outliers caused by entering service areas, temporary stops, etc. The filtered normal samples are denoted as T. ij Based on the filtered set of normal samples, the expected travel time and standard deviation are calculated respectively, as follows:

[0079]

[0080] To address data quality fluctuations caused by differences in sample size among adjacent devices, a piecewise modeling strategy is employed for standard parameter estimation. For gantry pairs with sufficient data, a linear model is used to fit the relationship between the expected travel time and its standard deviation, specifically expressed as follows:

[0081] tt std =k·tt exp +b

[0082] For gantry equipment pairs with scarce data, the standard parameters are predicted and completed using the globally linear model fitted above, specifically as follows:

[0083]

[0084] c13. Equipment Quality Assessment: To quantify the health status of each gantry inspection device, it is necessary to assess the health status of the gantry equipment by analyzing the missed detection rate and false detection rate. The detection rate and false detection rate are calculated to quantify equipment performance, and trust scores are assigned to the equipment based on these indicators. Equipment with low trust scores has lower weight. First, based on the previously constructed adjacency graph structure, the integrity analysis of vehicle trajectories is performed using historical detection data. For missing trajectory segments (e.g., trajectory interruption caused by a missed detection by a certain device), interpolation and path repair algorithms are used to complete the trajectory, generating a continuous and complete set of tracks, specifically represented as follows:

[0085] T full ={τ1,τ2,…,τ M}

[0086] Within a specified time window, the actual detection flow rate N of each detection device is counted. actual This refers to the actual number of vehicles detected and recorded. Simultaneously, based on the complete trajectory, the theoretical number of vehicles (the required inspection flow rate) N that the equipment should detect is calculated. expected The detection rate of this device is defined accordingly, and is specifically expressed as:

[0087]

[0088] This indicator can be used to quantify the rate of missed detections by gantry equipment. The higher the value, the fewer missed detections the equipment has and the stronger its detection capability.

[0089] Furthermore, identify typical abnormal events such as "false detection of opposing gantry" and count the number N vehicles that experienced erroneous detections. oncome Statistical analysis was conducted, and the false detection rate of the equipment was calculated, specifically expressed as follows:

[0090]

[0091] The smaller the value of this indicator, the fewer false alarms the detection equipment will produce, and the higher its stability.

[0092] Based on the detection rate and false positive rate metrics, a trust score is constructed for each device, which can be used in the weighted rules for subsequent data fusion and judgment. For gantry devices with low detection rates or high false positive rates, their data will be assigned lower weights or marked as low-trust devices.

[0093] Example 3

[0094] Please see Figure 1 This embodiment is a further optimization based on embodiment 2. Specifically, step c2 includes:

[0095] c21. Time Parameter Settings: To effectively capture traffic phenomena at different scales, from second-level abrupt changes to minute-level trends, three core time windows are set: a base window, a smoothing window, and a reference window. The base window is 30 seconds, which is the smallest time unit for real-time updates of traffic indicators. It is used for high-frequency data processing and rolling calculation of indicators, ensuring that the system has a second-level response capability. The smoothing window is 3 minutes, used to eliminate high-frequency noise and identify short-term trends. It performs smoothing processing such as moving average on the base traffic data to generate trend indicators. The reference window is 15 minutes, used to construct macro-statistics and equipment performance evaluation, supporting the modeling of key feature indicators that require long-term data support.

[0096] c22. Interval State Calculation: The interval state calculation algorithm runs periodically at fixed time intervals. Within each calculation cycle, the system loads data from the most recent reference time window from the real-time data stream, extracts data from the reference window that falls within the latest basic window, and calculates real-time traffic; it also extracts data from the reference window that falls within the latest smoothing window and calculates smoothed traffic; and within the complete reference window range, it calculates long-term traffic. All three types of traffic characteristics are uniformly converted to standard units (vehicles / hour) to participate in model fusion and discrimination calculations under a unified dimension.

[0097] For each pair of adjacent gantry devices (G) in the road network i G j Based on real-time data, vehicle trajectory matching is performed, upstream and downstream detection records are numbered and compared, and the set V of vehicles that have completed the path segment is identified. ij For each vehicle, first calculate its travel time, then, combining this with the actual distance between the two devices, calculate the average travel speed for that route segment, specifically expressed as:

[0098]

[0099] This speed value is updated in real time on a periodic basis to reflect the local road operating status.

[0100] c23. Upstream and Downstream Relationship Calculation: This function calculates the flow ratio between each pair of adjacent devices, integrating real-time flow, historical flow ratios, and estimated flow ratios to construct a robust flow ratio model and predict the upstream and downstream flow ratio at the current moment. This ratio is affected by traffic structure factors such as ramp divergence and merging, exhibiting some volatility; therefore, real-time calculation of the upstream and downstream flow ratio is necessary. The system calculates the flow ratio for each pair of upstream and downstream adjacent gantry detection devices in real time based on a sliding reference time window, specifically expressed as follows:

[0101]

[0102] Where, q i (t), q j (t) represents the detection equipment G within the reference window. i G j Traffic flow.

[0103] To overcome the random fluctuations in flow ratios, this method integrates three types of information to construct a robust predictive flow ratio model: the real-time flow ratio r'(t) at the current time, the estimated flow ratio r(t-1) from the previous time window, and the statistical flow ratio r at the same historical moment. h (t). Taking into account the above factors, the predicted upstream and downstream flow ratio at the current moment is calculated, specifically expressed as:

[0104] r(t)=k1·r'(t)+k2·r(t-1)+k3·r h (t)c24. Downstream Flow Range: After calculating the upstream and downstream flow ratio in real time, the reasonable flow range of the downstream gantry at the current time and in the short future time is predicted, and the smoothed flow rate V of the upstream detection equipment (gantry) at the delay time t-τ is selected. smooth,u (t-τ) serves as the prediction benchmark for the downstream gantry flow rate at the current moment, where τ represents the feasible travel time between the upstream and downstream detection devices. exp -3·tt std ,tt exp +3·tt std The process iterates through the range interval. Combining the upstream and downstream flow ratio r(t) obtained through historical fusion estimation at the current moment, a downstream expected flow estimation model is constructed, specifically expressed as:

[0105] V exp,min (t)=V smooth,u (t-τ)·r(t)·k min -C min

[0106] V exp,max (t)=V smooth,u (t-τ)·r(t)·k max +Cmax

[0107] By iterating through multiple τ values ​​within the feasible travel time interval using the above formula, multiple candidate upper and lower bounds for prediction are obtained. The final prediction interval is expressed as:

[0108]

[0109] Example 4

[0110] Please see Figure 1 This embodiment is a further optimization based on embodiment 3. Specifically, step c3 includes:

[0111] c31. Abnormal State Judgment: When the downstream smoothed flow rate V calculated in real time... smooth (t) deviates from its predicted range [V] exp,min (t),V exp,max When V (t)], the system activates the accumulation mechanism for abnormal state indicators. smooth (t) remains below V exp,min (t), then the cumulative "cumulative loss" indicator; if V smooth If (t) remains zero, the "continuous zero flow" metric is accumulated. When any of the above metrics exceeds a preset threshold, the system marks the current time window as abnormal and sets the abnormality flag variable to 1, thus completing the initial judgment of the abnormal state.

[0112] c32. Event Spatiotemporal Aggregation: Discrete anomalies are aggregated based on temporal and spatial proximity to form continuous anomalous events. If a new anomaly is adjacent to an existing event, its status is updated; otherwise, it is marked as a new event and initialized. The anomaly markers generated in the initial judgment phase are discretely distributed and require further aggregation in the temporal and spatial dimensions to identify potential persistent anomalous events. Therefore, an empty list of active events is constructed during system initialization. This is used to store currently active exception events. Let p be the newly generated exception point. i =(t i ,s i ), where t i Indicates its timestamp, s i This indicates its location information. When a new anomaly point p... i When detected, the system sequentially checks the list of active events. Each event E in j Proximity determination is performed, specifically including temporal proximity and spatial proximity. Temporal proximity determination is specifically expressed as follows:

[0113]

[0114] Spatial proximity determination can be specifically expressed as:

[0115] dist(s i ,S j )≤ΔS

[0116] If both proximity conditions are met simultaneously, then point p is considered an outlier. i For event E j The event continues, updating the event information status. The time range of the event is updated to... The gantry equipment set has been updated to S j ←S j ∪s i .

[0117] If any proximity condition is not met, then point p is considered an outlier. i The system records the new event E as the starting point for the new abnormal event. k Assign a unique event ID and initialize it. The event start time is... End time is The gantry equipment involved is S k =s i ,s u .

[0118] This spatiotemporal aggregation mechanism for anomalies ensures that the system can group discrete anomaly markers into continuous risk events, supporting subsequent alarm determination and response mechanisms.

[0119] c33. Event Cooling-Off Period: To prevent the same abnormal event from triggering alarms repeatedly, an event cooling-off period mechanism is implemented. Specifically, after an event alarm occurs, a short cooling-off period is set within the spatiotemporal range affected by the event. The purpose of the cooling-off period is to prevent frequent repeated alarms within the same event's influence range. During the cooling-off period, if the system detects an abnormal signal again, it will not trigger a new independent event alarm; instead, the signal will be treated as a continuation of the original event, and the location and time of the cooling-off period will be refreshed. If no abnormal signal occurs during the cooling-off period, the system will determine that the event has terminated, the cooling-off period will end, and the event alarm mechanism will be reactivated. At this time, the system can re-detect and trigger new independent event alarms, ensuring the effectiveness and accuracy of event identification and the alarm mechanism.

[0120] Example 5

[0121] Please see Figure 1 This embodiment is a further optimization based on embodiment 4. Specifically, step c4 includes:

[0122] c41. Video Surveillance Linkage: To improve the timeliness and accuracy of traffic incident response, after an incident is confirmed, the system automatically retrieves video surveillance resources for the corresponding road segment based on the incident's spatial location information, achieving real-time linkage between the incident and the video. Once a traffic anomaly is confirmed, the system automatically retrieves relevant video surveillance resources based on the incident location, connects with the traffic management platform via API, queries and matches suitable cameras, and retrieves video streams or historical clips in real time. When a traffic anomaly is confirmed, the system generates the incident's spatial location information, including the highway name, direction of travel, and specific mileage marker range, represented as:

[0123] Event = {road,direction,[K]} min ,K max ]}

[0124] Once the location information of a traffic anomaly is generated, the system automatically triggers the video linkage module. This module interfaces with the video surveillance system of the traffic management platform through a preset API interface. This module will define the spatial range of the event [K]. min ,K max As input, query the set of camera locations {K} in the surveillance system. i} Calculate the distance between each camera and the event interval, specifically expressed as:

[0125] d i =min(|K i -K min |,|K i -K max |)

[0126] Select condition d i Matching 1 to 3 cameras with a value of ≤δ, the system will connect to the video surveillance platform based on the matching results and automatically retrieve real-time video streams or historical video clips.

[0127] c42. Human-Computer Interaction Verification: The system pushes event-related data to the verification interface of the traffic monitoring center. Maintenance personnel cross-reference video, traffic flow curves, and alarm parameters to quickly determine the authenticity of the event and implement tiered management, reducing false alarm rates and improving response efficiency. After identifying a traffic anomaly and generating corresponding spatiotemporal location information, the system pushes multi-source data associated with the event to the verification interface of the traffic monitoring center. This data includes anomaly indicators related to the event trigger, traffic flow change curves for the corresponding time period, precise event location information, and real-time monitoring video streams or retrospective video clips retrieved through the video linkage module.

[0128] This verification interface is designed as a human-computer interactive platform to assist operations and maintenance personnel in efficiently determining the authenticity and classifying events. Within this interface, personnel can cross-reference pushed video feeds with traffic curves and alarm parameters to quickly verify the type, severity, and priority of events. This visual verification mechanism further reduces false alarm rates and improves response efficiency.

[0129] c43. Information Release Decision: After event confirmation, the system determines the scope and channels for information release based on rules in the strategy library. Important events are released through a wide range of channels, while minor events are released only within a limited scope to ensure accurate information delivery. The system's built-in information release module makes information release decisions based on a preset strategy library after the event has been manually verified or automatically confirmed by the system in specific high-confidence scenarios. The decision-making strategies in the strategy library consider multiple factors, including event type, event severity, time of event occurrence, and importance of the road segment.

[0130] Based on the information dissemination strategy, the system will automatically determine the scope and channels for information dissemination. For example, in the event of a serious accident, the system will release the event information through a wider range of channels to ensure timely communication to relevant departments, surrounding vehicles, and the public. For minor incidents such as brief traffic congestion, the system will only issue alerts on specific road sections or within designated areas to avoid excessive information dissemination. This dissemination decision-making mechanism ensures the accuracy and timeliness of event information dissemination and effectively reduces the risk of information overload.

[0131] c44. Information Closed-Loop Feedback: After traffic anomaly events are processed by the decision-making module, the system pushes the formatted event information to multiple terminals and platforms through standardized interfaces, specifically including:

[0132] Gantry Information Display (VMS): The system dynamically generates event-related prompts, such as "Accident ahead at K106, please slow down," and displays them on the gantry information display to provide real-time reminders to passing vehicles.

[0133] On-board unit (OBU): By linking with the ETC service platform, the system can push short messages or issue warnings via buzzer to vehicles about to enter the road section, reminding drivers to pay attention to road conditions.

[0134] Navigation platform: The system connects with mainstream map apps to push event information to public applications in real time, providing accurate traffic guidance to help drivers plan routes and avoid passing through event areas;

[0135] Traffic management platform: The system will report incident information to the superior command and dispatch system to ensure that traffic management departments are aware of the incident in a timely manner and take corresponding emergency measures.

[0136] After information is released, the system automatically records the release status and awaits updates from operations and maintenance personnel. Personnel can update the event handling status within the system. Through a closed-loop recording and feedback mechanism, the system ensures effective tracking and management of the entire event process, from detection and verification to release and handling. This closed-loop feedback mechanism ensures transparency and efficiency in event handling, while also providing data support for subsequent analysis and optimization.

[0137] Based on the same concept as the above method, this application also provides a highway traffic incident detection system based on gantry data, comprising:

[0138] Historical Benchmark Construction Module: This module establishes a static benchmark model of the highway network based on historical ETC gantry data. It constructs a topological adjacency graph describing the spatial relationships between gantry structures, calibrates the segment travel time benchmark parameters under normal free-flow conditions, and evaluates the detection reliability of each gantry device, providing crucial decision-making support and data quality assurance for real-time detection.

[0139] Real-time Status Awareness and Prediction Module: This module processes real-time data streams from ETC gantries, dynamically monitoring and predicting traffic conditions. It calculates gantry flow and section speed in real-time using multi-level time windows and dynamically analyzes the upstream and downstream flow ratios based on traffic flow theory. Its core function is to predict the reasonable expected flow range for downstream gantries, laying the foundation for real-time comparison in identifying abnormal flow.

[0140] Intelligent Event Detection and Location Module: Focused on identifying and accurately locating abnormal traffic events on highways. This module continuously compares downstream measured traffic flow with the predicted expected traffic flow range, triggering and accumulating dynamic anomaly indicators. When the anomaly reaches a preset threshold, the system performs spatiotemporal aggregation analysis, combined with an event status management mechanism, to accurately pinpoint the road segment where the event occurred.

[0141] The information closed-loop processing module automates the entire process of event confirmation, information dissemination, and processing feedback. This module automatically links with video surveillance for rapid manual verification. Upon confirmation, it intelligently matches dissemination strategies based on event attributes and precisely pushes structured event information to gantry information boards, vehicle-mounted ETC terminals, navigation platforms, and traffic management platforms via standardized interfaces. The system records the entire process and receives processing feedback, completing fully automated closed-loop management.

[0142] In summary, this invention deeply integrates historical benchmark modeling with real-time dynamic perception to construct a robust highway traffic incident detection system. By accurately establishing road network topology, free-flow parameter benchmarks, and equipment health models, it provides a solid basis for real-time analysis and decision-making. Simultaneously, by utilizing multi-timescale analysis and dynamic traffic prediction models, it achieves accurate perception of traffic conditions and intelligent identification of abnormal traffic flow.

[0143] This invention employs a dynamic cumulative judgment mechanism based on traffic prediction deviation and an intelligent spatiotemporal aggregation strategy to achieve efficient event location and deduplication. By setting thresholds to trigger abnormal indicators and combining spatiotemporal proximity analysis for event aggregation and status management, it can quickly and accurately pinpoint the road segment interval where the event occurred, effectively avoiding false alarms and duplicate alarms, and significantly improving detection efficiency and location accuracy.

[0144] This invention establishes a complete closed-loop handling process of "detection-verification-release-feedback," significantly improving the timeliness of event response and the accuracy of information release. The system automatically links with video surveillance for rapid manual verification, intelligently decides the scope and channels of release based on a preset strategy library, and accurately pushes structured event information through multiple channels. Simultaneously, the system records the release status and receives handling feedback, achieving fully automated closed-loop management of the entire event process from discovery to handling, greatly improving the operational safety and traffic efficiency of highways.

[0145] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. The scope of patent protection of the present invention shall be determined by the claims. Similarly, any equivalent structural changes made based on the description and drawings of the present invention shall also be included within the scope of protection of the present invention.

Claims

1. A method for detecting highway traffic incidents based on gantry data, characterized in that, Includes the following steps: c1. Obtain historical traffic data of all ETC gantries within the target area. Construct a directed gantry topology network based on device ID, real-time traffic data of spatial location, and driving direction. Generate a gantry adjacency table containing upstream and downstream relationships. Based on vehicle trajectory matching results, use iterative filtering to clean travel time samples, establish a traffic flow benchmark parameter library under free flow conditions, and simultaneously quantify the feature detection rate and false detection rate of each gantry device to form a static decision-making basis dataset. c2. Integrate real-time gantry data streams across multiple time scales, capture instantaneous flow fluctuations through a base window, filter out random noise and extract short-term trends through a smoothing window, and calculate dynamic flow ratios using a reference window. By integrating equipment health weights and historical benchmark parameters, the expected flow range of downstream gantry can be predicted in real time, and the confidence interval boundary can be dynamically adjusted based on upstream smoothed flow and travel time delay. c3. Design a decision function based on the accumulation of flow deviation. When the real-time smoothed flow continuously deviates from the expected range, trigger the accumulation counting mechanism of loss amount or continuous zero flow duration. Perform spatiotemporal proximity clustering on abnormal points that exceed the threshold to generate discrete event clusters and assign them unique IDs. At the same time, start the cooling-off period clock to suppress repeated alarms for automatic tracking and location of abnormal events. c4. Automatically link the location event to the road network video surveillance system, retrieve the optimal viewpoint image and overlay traffic parameter curves onto the human-machine interface; after the operation and maintenance personnel confirm the event type and severity, drive the decision tree strategy to generate hierarchical release instructions, push structured event information through multiple channels such as gantry information boards, vehicle-mounted ECT terminals, navigation platforms, and traffic management platforms, and record handling feedback to optimize detection thresholds and cooling-off period parameters.

2. The highway traffic incident detection method based on gantry data according to claim 1, characterized in that, Step c1 includes: c11. Adjacency Graph Assembly: Construct a directed topology graph using ETC data, sort devices by driving direction and mileage marker to ensure upstream and downstream relationships, and introduce redundancy strategies to form an adjacency table; c12. Free Flow Parameter Calibration: After constructing the adjacency list, a normal traffic spatiotemporal parameter benchmark is established for each pair of adjacent devices based on historical data. By extracting travel time samples and performing filtering analysis, the expected travel time and its fluctuation range are calculated. c13. Equipment Quality Assessment: The health status of gantry equipment is assessed by analyzing the false negative rate and false positive rate, the detection rate and false positive rate are calculated to quantify the equipment performance, and a trust score is assigned to the equipment based on these indicators.

3. The highway traffic incident detection method based on gantry data according to claim 1, characterized in that, Step c2 includes: c21. Time parameter settings: In order to capture traffic phenomena at different time scales, three time windows are set, including the base window, the smoothing window and the reference window; c22. Interval State Calculation: Based on real-time data and reference time windows, the system calculates three types of traffic characteristics: real-time traffic, smooth traffic, and long-term traffic; through vehicle trajectory matching, the average travel speed of each path segment is calculated. c23. Upstream and downstream relationship calculation: Calculate the flow ratio between each pair of adjacent devices, integrate real-time flow, historical flow ratio and estimated flow ratio to build a robust flow ratio model, and predict the upstream and downstream flow ratio at the current moment. c24. Downstream flow range: After calculating the upstream and downstream flow ratio, predict the reasonable flow range of downstream equipment, and estimate the upper and lower limits of downstream flow by combining historical data and flow ratio model.

4. The highway traffic incident detection method based on gantry data according to claim 1, characterized in that, Step c3 includes: c31. Abnormal state judgment: When the downstream smoothed flow deviates from the prediction range, the system will accumulate "cumulative loss" or "continuous zero flow" indicators based on the duration and the situation of zero flow. Once the threshold is exceeded, the current time window is marked as an abnormal state. c32. Event Spatiotemporal Aggregation: Aggregates discrete anomalies through temporal and spatial proximity to form continuous anomaly events; if a new anomaly is adjacent to an existing event, the event state is updated; if not adjacent, it is marked as a new event and initialized. c33. Event Cooling-Off Period: To avoid repeated alarms, a cooling-off period is set after an event occurs. If an anomaly is detected again during the cooling-off period, the event is considered to continue. If no anomaly is detected, the event ends, the cooling-off period is lifted, and the system reactivates the alarm mechanism.

5. The highway traffic incident detection method based on gantry data according to claim 1, characterized in that, Step c4 includes: c41. Video surveillance linkage: After confirming an abnormal traffic event, the system automatically retrieves relevant video surveillance resources based on the event location, and retrieves video streams or historical clips. c42. Human-computer interaction verification: The system pushes event-related data to the verification interface of the traffic monitoring center. The operation and maintenance personnel cross-compare the video, traffic curve and alarm parameters to complete the event authenticity determination and hierarchical management. c43. Information Release Decision: After an event is confirmed, the system determines the scope and channels for information release based on the rules in the policy library; important events will be released through a wide range of channels, while minor events will only be released within a local scope. c44. Closed-loop information feedback: The system pushes event information to multiple platforms through standardized interfaces, records the release status, and waits for maintenance personnel to update the event handling status.

6. A highway traffic incident detection system based on gantry data, characterized in that, include: Historical benchmark construction module: Establishes a static benchmark model of the highway network based on historical ETC gantry data; Real-time status perception and prediction module: responsible for processing real-time data streams from ETC gantry, dynamically monitoring and predicting traffic conditions; Intelligent event detection and localization module: used to identify and locate abnormal traffic events on highways; Information closed-loop processing module: used for the automated closed loop of event confirmation, information release and processing feedback.

7. A highway traffic incident detection system based on gantry data according to claim 6, characterized in that: The historical benchmark construction module is used to construct a topological adjacency graph describing the spatial relationships of gantry structures, calibrate the segment travel time benchmark parameters under normal free flow conditions, and evaluate the detection reliability of each gantry device. The real-time status perception and prediction module calculates gantry flow and section speed in real time through multi-level time windows, and dynamically analyzes the upstream and downstream flow ratio based on traffic flow theory; predicts the reasonable expected flow range of the downstream gantry.

8. A highway traffic incident detection system based on gantry data according to claim 6, characterized in that: The event intelligent detection and location module continuously compares the measured downstream traffic flow with the predicted expected traffic flow range to trigger and accumulate dynamic anomaly indicators. When the anomaly reaches a preset threshold, the system performs spatiotemporal aggregation analysis and, in conjunction with the event status management mechanism, locks down the road segment interval where the event occurred. The information closed-loop processing module automatically links with video surveillance for rapid manual verification; after confirmation, it intelligently matches the release strategy according to the event attributes and pushes structured event information to multiple platforms through standardized interfaces; the entire process records the status and receives processing feedback.

Images