Multi-level highway event distinguishing and early warning method based on self-adaptive network topology and electronic equipment

Abstract

The invention provides a multi-level highway event distinguishing and early warning method based on self-adaptive network topology and electronic equipment, and relates to the field of traffic event early warning. According to the method, key nodes are selected through correlation analysis of the whole road network, and event discrimination is carried out by using traffic parameters of video statistics, so that algorithm complexity and computing power consumption are effectively reduced; networking video streams and ETC door frame data are fully utilized to obtain perception analysis data of road section flow and target vehicles, the perception analysis data are input into an adaptive network traffic flow abnormity discovery alarm model, deep research and judgment analysis is carried out, various traffic abnormity conditions are accurately identified, timely and effective decision reference is provided for a traffic management department, and the traffic flow abnormity discovery alarm model is established. Therefore, the safety and the passing efficiency of the road are improved.

Description

Technical Field

[0001] The present invention relates to the field of traffic incident warning, and in particular to a multi-level highway incident identification and warning method based on adaptive network topology and electronic equipment. Background Art

[0002] With economic development, highway traffic has surged, leading to frequent congestion and accidents, severely impacting travel efficiency and safety. Simultaneously, natural disasters (such as earthquakes, rockfalls, and landslides) frequently occur, wreaking havoc on highway transportation systems. These disasters not only directly damage highway infrastructure but also disrupt traffic, hindering the delivery of relief supplies and the evacuation of personnel, severely restricting economic activity, and significantly impacting the social economy and people's lives. Therefore, research on traffic parameter-based incident detection and early warning is of great significance. However, traditional highway early warning methods have limitations, such as focusing on a single event type, relying on fixed sensor networks, and lacking consideration of traffic dynamics.

[0003] In summary, current research on highway detection and early warning systems is largely based on information such as traffic flow, images, and speed collected by loop coils, video surveillance, and sensing devices, which is then further analyzed to generate event detection results. At the same time, a growing number of studies are focusing on the fusion of multi-source data for detection. However, most existing research neglects understanding the overall operational status of the highway network and fails to analyze correlations between equipment points. Furthermore, implementing video algorithm detection on all surveillance videos requires significant computing power and is costly to cover the entire road network, making it impractical. Furthermore, in addition to video surveillance, highway gantries (ETCs) can also provide high-quality highway traffic data without adding new equipment. Summary of the Invention

[0004] Purpose of the invention: To propose a multi-level highway event identification and warning method and electronic equipment based on adaptive network topology, which selects key nodes through correlation analysis of the overall road network and uses traffic parameters from video statistics to identify events, effectively reducing algorithm complexity and computing power consumption; making full use of networked video streams and ETC gantry data to obtain perception analysis data of road section traffic and target vehicles, inputting the adaptive network traffic flow anomaly detection and alarm model, conducting in-depth research and analysis, accurately identifying various traffic anomalies, and providing timely and effective decision-making references for traffic management departments to improve road safety and traffic efficiency, thereby effectively solving the above-mentioned problems existing in the existing technology.

[0005] In a first aspect of the present invention, a multi-level highway event identification and warning method based on an adaptive network topology is proposed, comprising the following steps:

[0006] Use existing highway video cameras to collect video data; connect to the ETC system, and its gantry automatically records ETC data when vehicles pass through;

[0007] Based on map resources and known fixed points, the connectivity of the intervals is analyzed according to the road directions in the video data and the vehicle travel paths in the ETC data, and whether there are direct traffic connections between the fixed points is determined, and an adaptive fusion point topology is constructed;

[0008] Count the video data and ETC data of the predetermined time series to form historical traffic flow data and historical interval travel time; establish traffic anomaly discrimination mechanisms based on traffic flow and based on interval travel time respectively;

[0009] The collected real-time video data and real-time ETC data are substituted into a traffic anomaly identification mechanism based on traffic flow and / or a traffic anomaly identification mechanism based on interval travel time. If an abnormal event is determined, the source point causing the abnormality is traced back according to the adaptive fusion point topology, the abnormal road section is determined, and the abnormal event information is released.

[0010] In a further embodiment of the first aspect, the known fixed points include each video camera point and each ETC gantry point;

[0011] The adaptive fusion point topology includes the distance correlation and flow correlation between each fixed point.

[0012] In a further embodiment of the first aspect, the actual spatial distance d between the fixed points is calculated. BB , based on the actual space distance d AB , use the inverse proportional function method to analyze the distance correlation between points k is a preset coefficient.

[0013] In a further embodiment of the first aspect, based on the traffic flow information in the video data and ETC data, a traffic flow sequence X between different points is counted as {x1, x2, ..., x n} and Y={y1,y2,…,y n};

[0014] The Pearson correlation coefficient r of the traffic flow sequence at each point is calculated to determine the traffic correlation between two video points:

[0015]

[0016] Where n is the length of the traffic flow sequence, is the mean of the sequence X, is the mean of the sequence Y;

[0017] The correlation coefficient ranges from -1 to 1. Values close to 1 indicate a high degree of positive correlation, values close to -1 indicate a high degree of negative correlation, and values close to 0 indicate a weak correlation.

[0018] In a further embodiment of the first aspect, a traffic anomaly identification mechanism based on traffic flow is established, specifically including:

[0019] Counting the video data and ETC data of a predetermined time series to form historical traffic flow data; dividing the historical traffic flow data into two time periods, with the first time period serving as a horizontal reference and the second time period serving as a vertical reference;

[0020] Constructing a neural network, inputting the horizontal benchmark into the neural network, and predicting the flow data of the next period as a prediction benchmark;

[0021] The horizontal benchmark, the longitudinal benchmark, and the prediction benchmark are combined to form a traffic anomaly determination benchmark;

[0022] The collected real-time traffic flow data is compared with the traffic anomaly judgment benchmark. If the current real-time traffic flow data exceeds the traffic anomaly judgment benchmark, the traffic flow is judged to be abnormal. The source point causing the anomaly is traced back according to the adaptive fusion point topology, the abnormal road section is determined, and the abnormal event information is released.

[0023] In a further embodiment of the first aspect, a neural network is constructed, the horizontal reference is input into the neural network, and flow data of the next time period is predicted as a prediction reference;

[0024] Calculate the percentage difference β between the predicted baseline and the actual traffic flow in the corresponding period in the historical traffic flow data:

[0025]

[0026] Where Q n is the actual traffic volume in the nth period, is the forecast benchmark for the nth period.

[0027] In a further embodiment of the first aspect, a traffic anomaly identification mechanism based on interval travel time is established, specifically including:

[0028] The average travel time series of the historical interval is calculated by detecting targets at upstream and downstream points of the interval, recording the time when the same vehicle enters and exits the interval, subtracting the time to obtain the travel time of a single vehicle, and calculating the average travel time of the vehicles according to the predetermined time step to obtain the average travel time series of the historical interval;

[0029] A multi-dimensional interval travel time prediction benchmark model is established by combining the impact of environmental data, including weather conditions and road construction conditions, on vehicle travel speed. This model is then fitted and trained using historical vehicle travel times. The trained model is then used to make real-time predictions of the next interval travel time.

[0030] Based on the upstream and downstream relationship of the adaptive fusion point topology, the vehicles passing through the starting point of the interval are tracked at a preset frequency, and the time when the target vehicle enters the interval is recorded; if there are multiple downstream points at the point where the target vehicle is extracted, the target vehicle extraction frequency is increased, and it is ensured that the target vehicle passes through all downstream points; at the same time, the appearance features of the target vehicle are identified through the target detection algorithm based on deep learning, and the target vehicle extracted upstream is identified and matched at the downstream point of the interval, and the time when the target vehicle leaves the interval is recorded.

[0031] In a further embodiment of the first aspect, several verification points are selected at the downstream points of the interval. If the target vehicle is not detected at the downstream points of the interval but is detected at the verification points, the abnormal alarm is eliminated in time.

[0032] In a further embodiment of the first aspect, establishing a traffic anomaly identification mechanism based on interval travel time further includes:

[0033] Combined with the travel time prediction value T predict The target vehicle’s theoretical departure time is calculated with the proportional factor α. If the target vehicle has not left the interval after exceeding the theoretical time, the target vehicle’s overtime percentage γ is recorded, and the number of consecutive overtime vehicles is counted, where:

[0034]

[0035] When the time between the monitoring point and the last extraction reaches the preset value, a new target vehicle extraction is carried out; each target vehicle is tracked in real time and the timeout percentage and the number of consecutive timeout vehicles of each target vehicle are calculated. If the vehicle timeout percentage or the number of consecutive timeout vehicles exceeds the threshold, it is judged as an abnormality.

[0036] In a second aspect of the present invention, an electronic device is proposed, which includes a processor and a memory storing computer program instructions; when the processor executes the computer program instructions, it implements the multi-level highway event identification and warning method based on adaptive network topology as disclosed in the first aspect.

[0037] Compared with the prior art, the present invention has the following beneficial effects:

[0038] 1. The method of the present invention makes full use of existing road data resources (monitoring video streams and high-speed ETC), fully considers the static and dynamic adjacency topological relationship between monitoring points, extracts different traffic parameter features from multi-source data, and constructs corresponding abnormality discrimination features based on data characteristics. At the same time, it uses an event discrimination algorithm based on road section traffic and target vehicles, which improves the timeliness of event recognition while maintaining accuracy and robustness.

[0039] 2. This method reflects the correlation and degree of correlation between points by calculating the static adjacency matrix, distance correlation matrix, and flow correlation matrix, and uses weighted fusion to construct an adaptive fusion point topology to assist in establishing a prediction benchmark for flow and interval travel time, as well as screening key nodes in the road network.

[0040] 3. This method combines the fusion topology weights to screen out key nodes in the road network, conducts real-time statistics and key analysis on the monitoring points at the key node locations, and simultaneously traces back to other upstream nodes when problems occur at the monitoring nodes, thus achieving comprehensive perception while reducing computing power consumption.

[0041] 4. Traffic parameter statistics from multi-source data processing fully consider spatiotemporal factors, employing different anomaly detection methods and models based on the characteristics of different traffic parameters. Fusion analysis is then performed on the results, with varying confidence levels set for different event sources based on event type. A hierarchical fusion strategy is employed to ultimately achieve accurate traffic event identification. Furthermore, independent anomaly detection results for both traffic volume and vehicles ensure proper event identification when single data collection equipment is ineffective.

[0042] 5. This approach incorporates continuous monitoring and information dissemination after incident discovery into the system, enabling precise tracking of incident developments and real-time data updates. This overcomes the static assessment limitations of traditional methods and provides a dynamic and accurate basis for decision-making. In the information dissemination process, multi-channel integration effectively avoids information lags and poor dissemination, improving the timeliness and efficiency of emergency response and forming a closed-loop process of "discovery-identification-warning-notification." BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 Flowchart of a multi-level highway event identification and warning method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0044] In the following description, numerous specific details are provided to provide a more thorough understanding of the present invention. However, it will be apparent to those skilled in the art that the present invention may be practiced without one or more of these details. In other instances, certain technical features well known in the art have not been described to avoid confusion with the present invention.

[0045] The proposed multi-level highway incident identification and warning method based on adaptive network topology leverages the advantages of multi-source highway data. It comprehensively analyzes networked video streams and ETC gantry data, and applies targeted macro- and micro-analysis models to different data types. The resulting identification results are both timely and stable. This method involves four phases: multi-source data acquisition and preprocessing, adaptive fusion point topology construction, traffic flow anomaly detection, and incident warning notification and information dissemination. The specific steps are as follows:

[0046] (1) Multi-source data collection and preprocessing

[0047] Step 1: Multi-source data collection. ① Video data collection: Utilize existing highway video cameras to comprehensively cover traffic scenarios, effectively capturing information such as vehicle trajectories, traffic flow, and road conditions. ② ETC data collection: Leverage the existing ETC system, whose gantries automatically record relevant information as vehicles pass through, forming a natural data collection network. ③ Road network topology data collection: Leveraging map resources and basic information on fixed points, initially construct the road network topology and the relationships between points.

[0048] Step 2: Data Cleaning. ① Video Data Cleaning: Remove video clips with poor image quality due to environmental factors (such as lighting and weather) and equipment issues, and calibrate timestamps. ② ETC Data Cleaning: Identify and correct erroneous data caused by equipment failures (such as card reader and network issues), and remove duplicate records.

[0049] (2) Adaptive fusion point topology construction

[0050] Step 1: Obtain the point adjacency matrix

[0051] According to the video networking naming standards proposed in the "National Comprehensive Transportation Information Platform Video Resource Coding and Naming Specifications", the interval connectivity is analyzed by combining the road directions in the video data and the vehicle driving paths in the ETC data to determine whether there are direct traffic connections between each point, and then construct an initial point adjacency matrix.

[0052] Step 2: Calculation of point distance correlation matrix

[0053] Using the pile number, GIS technology and the location information in the traffic data, the actual spatial distance between each point is calculated. Based on the distance between the points, the inverse proportional function method is used to analyze the distance correlation between the points. Let the distance correlation between the two video points be C AB , the distance between the two points is d A1 , construct the inverse proportional function as Where k is an undetermined coefficient. Points that are close to each other and have close transportation connections have a higher distance correlation.

[0054] Step 3: Calculation of point flow correlation matrix

[0055] Based on the traffic flow information in the video data and ETC data, the vehicle flow and traffic flow sequence between different points are counted. The Pearson correlation coefficient of the traffic flow sequence at each point is calculated to determine the degree of traffic correlation between two video points. The specific calculation method is as follows:

[0056] Assume that the traffic flow sequences of two points are X={x1,x2,…,x n} and Y={y1,y2,…,y n}, the Pearson correlation coefficient is recorded as r, then

[0057]

[0058] Where n is the length of the sequence, is the mean of the sequence X, is the mean of the sequence Y, that is

[0059] The correlation coefficient ranges from -1 to 1. Values close to 1 indicate a high degree of positive correlation, values close to -1 indicate a high degree of negative correlation, and values close to 0 indicate a weak correlation.

[0060] Step 4: Adaptively fusion point topology

[0061] By analyzing historical traffic data and applying machine learning algorithms, we determine the weights of the static adjacency matrix, distance correlation matrix, and flow correlation matrix to reflect the actual road network topology. Using a weighted fusion approach, we construct an adaptive fused point topology. This topology dynamically reflects the actual relationships between points in the traffic network and continuously optimizes as traffic data is updated. By combining the fused topology weights, we identify key nodes in the road network and conduct real-time statistics and analysis of monitoring points at these key node locations.

[0062] (3) Discovery of traffic flow anomalies

[0063] Step 1: Extracting time correlation between points and dividing time periods

[0064] Data analysis: Conduct statistical analysis on long time series (such as months or even years) of multivariate fusion data to observe the changing patterns of traffic flow parameters on different time scales (days, weeks, months, and years).

[0065] Time Segmentation: Based on data analysis results, cluster analysis and other methods are used to divide the day into different traffic periods. Based on the final clustering results, different time points are divided into corresponding peak, off-peak, and off-peak periods. The impact of special dates (such as holidays and major events) on traffic periods is also taken into account, and the division results are dynamically adjusted.

[0066] Step 2: Traffic anomaly identification based on flow rate

[0067] Traffic flow statistics:

[0068] ① Statistical analysis of high-speed gantry ETC vehicle passing data to obtain a fixed-step gantry vehicle flow time series.

[0069] ② Perform cross-sectional vehicle detection and traffic statistics on high-speed video streams to obtain the traffic time series of monitoring points.

[0070] Selection and establishment of traffic anomaly judgment benchmark:

[0071] Historical traffic flow data is analyzed to determine the normal flow range for different time periods and road sections, which serves as the benchmark for determining traffic anomalies. Taking into account the impact of factors such as seasonal changes and special events on traffic flow, the judgment benchmark needs to be dynamically adjusted. Furthermore, given the inherent temporal correlation of traffic flow sequences, the changing trend of traffic flow is closely related to historical flow and is also affected by periodicity at different scales. Therefore, the traffic flow of the previous period is selected as the horizontal benchmark, and the average traffic flow of the next period on the same date in the past is used as the vertical benchmark. The real-time traffic flow prediction of the next period by the neural network is used as the prediction benchmark. Together, these form the benchmark value for determining traffic anomalies.

[0072] The acquisition and calculation methods corresponding to each benchmark are:

[0073] Horizontal benchmark: directly extracted from the interval traffic sequence of video statistics;

[0074] Vertical benchmark: query and calculate the mean of traffic flow in the same interval, date and time period with the same attributes;

[0075] Prediction benchmark: Based on the results of spatial correlation analysis, combined with traffic attributes (traffic time period, fusion topology) and external environmental data, the traffic flow at each point in the next period is predicted using historical point traffic flow.

[0076] Abnormal discrimination model construction and model training:

[0077] Select appropriate machine learning algorithms, such as decision trees, deep learning, and ensemble learning, to build a traffic anomaly identification model. Calculate the percentage difference between historical traffic flow and three prediction benchmarks as the original dataset, splitting it into a training set and a test set. Use the training set to train the model, improving its accuracy by adjusting model parameters and optimizing the algorithm structure. Use the test set to validate and evaluate the trained model (using metrics such as accuracy, recall, F1 score, and ROC-AUC). Based on the evaluation results, fine-tune the model to ensure its generalization ability on unknown data.

[0078] Model input feature calculation:

[0079] The percentage difference between the actual traffic flow and the three predicted benchmarks is calculated in real time. The specific calculation method is as follows:

[0080]

[0081] Among them, Q n is the actual traffic volume in the nth period, The traffic baseline for the nth period. These features are normalized and standardized to make them suitable for the input requirements of the model.

[0082] Point anomaly identification and upstream tracing and positioning:

[0083] The real-time calculated features are fed into a trained traffic anomaly discrimination model, and the traffic discriminant features are then fed into the discriminant model to obtain anomaly identification results and event probability. When a traffic anomaly is discovered, the system uses the road network topology information to trace the source of the anomaly upstream based on the traffic flow direction and volume correlation. The system then combines the traffic status of the upstream points to determine whether an event has occurred and the interval between the points where the event occurred.

[0084] Step 3: Vehicle-based traffic anomaly identification

[0085] Establishment of interval travel time prediction benchmark:

[0086] ① Statistical analysis of the historical average travel time series: By detecting targets at upstream and downstream points in the interval, the entry and exit time of the same vehicle is recorded, and the subtraction is used to obtain the single vehicle travel time. The average vehicle travel time is calculated at a certain time step to obtain the historical average travel time series.

[0087] ② Combine traffic attributes (traffic time periods, fusion topology) with environmental data (such as weather and road construction) to understand the impact of factors on vehicle speed. A multi-dimensional baseline model for interval travel time prediction is established. This model is fitted and trained using historical data. The trained model is then used to make real-time predictions of travel times for the next interval. As new data is generated, the model is regularly updated and optimized to adapt to dynamic changes in traffic conditions and ensure its accuracy and timeliness.

[0088] Target vehicle extraction and tracking:

[0089] Based on the upstream and downstream relationships of the point topology, vehicles passing through the starting point of the interval are tracked at a preset frequency, and the time the target vehicle enters the interval is recorded. If the point where the target vehicle is extracted has multiple downstream points, the target vehicle extraction frequency needs to be increased to ensure that the target vehicle passes through all downstream points. At the same time, a deep learning-based target detection algorithm accurately identifies the appearance features of the target vehicle in complex traffic scenarios, and identifies and matches the target vehicle extracted upstream at the downstream point of the interval, recording the time when the target vehicle leaves the interval.

[0090] In order to reduce abnormal false alarms caused by vehicle identification errors, a small number of verification points are selected downstream of the downstream points in the interval. If the target vehicle is not detected at the downstream points in the interval but is detected at the verification points, the abnormal alarm is eliminated in time.

[0091] Calculation of overtime percentage and number of consecutive overtime vehicles:

[0092] Combined with the travel time prediction value T predict With the proportional factor α, the theoretical time for the target vehicle to leave the interval is calculated. If the target vehicle has not left the interval after the theoretical time, the overtime percentage γ of the target vehicle is recorded, and the number of consecutive overtime vehicles is counted, where

[0093]

[0094] Threshold setting and anomaly identification:

[0095] Based on historical data analysis, the distribution of the percentage of vehicles that exceed the time limit and the number of vehicles that exceed the time limit continuously under normal traffic conditions is statistically analyzed. For the binary classification model, the true positive rate (TPR) and false positive rate (FPR) are calculated at different thresholds. The ROC curve is drawn with FPR as the horizontal axis and TPR as the vertical axis. Each point on the ROC curve corresponds to a threshold. Ideally, the ROC curve will be close to the upper left corner (TPR=1, FPR=0). The threshold corresponding to the maximum area under the ROC curve (AUC) is selected as the preliminary classification threshold, and it is adjusted based on traffic management experience and actual traffic needs.

[0096] When the time between the monitoring point and the last extraction reaches the preset value, a new target vehicle extraction is carried out; each target vehicle is tracked in real time and the timeout percentage and the number of consecutive timeout vehicles of each target vehicle are calculated. If the vehicle timeout percentage or the number of consecutive timeout vehicles exceeds the threshold, it is judged as an abnormality.

[0097] Step 4: Active event discovery and multi-data fusion event analysis

[0098] The flow-based traffic anomaly discrimination and vehicle-based traffic anomaly discrimination are fused and analyzed, and the confidence levels of different event sources are set according to the event types. A hierarchical fusion strategy is adopted to finally obtain accurate traffic events.

[0099] (IV) Event warning notification and information release

[0100] Step 1: Event reporting and summary

[0101] After an anomaly is discovered, the points and sections will be continuously verified and monitored. If necessary, the detection frequency will be increased, and the verified alarm location, alarm type, alarm time and other information will be reported to the traffic management department.

[0102] Step 2: Multi-channel connection and information release

[0103] The traffic management department coordinates with the traffic police and connects with information release channels to summarize emergency information, event impact information, road control information, detour plan information, etc., and releases information through Internet maps, public travel service websites, roadside information boards and other channels, with a focus on instant push of navigation services to increase the speed and coverage of alarm dissemination.

[0104] Step 3: Continuous monitoring and disposal completed

[0105] After the incident is released, continue to use multi-source data to monitor traffic conditions in and around the incident area in real time. Regularly update traffic data and analyze traffic trends. Use this monitoring data to determine the handling of the incident until traffic returns to normal.

[0106] The above embodiments can be implemented in whole or in part through software, hardware, firmware, or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The specific embodiment process is as follows:

[0107] 1. Select the implementation area for the plan and determine the monitoring equipment and high-speed gantry ETC information available in the area, including point locations, point names, and vehicle passing data. Clean the data and perform statistics on the high-speed gantry ETC and video stream vehicle passing data to obtain a fixed-step historical traffic flow time series and a fixed-step interval average traffic time series.

[0108] 2. Obtain a point adjacency matrix from point information; calculate a point distance correlation matrix based on the actual spatial distance between each point; and form a point flow correlation matrix by calculating the similarity of flow curves at different points. Use weighted fusion to construct an adaptive fused point topology, reflecting the degree of correlation between points and screening key nodes.

[0109] 3. Combine the historical traffic flow time series from step 1 and the fused point topology from step 2 to extract or predict different traffic anomaly discrimination benchmarks and calculate model input features based on actual traffic flow. Select an appropriate machine learning algorithm (a random forest model is used in this specific example) and divide the integrated and annotated data into training, validation, and test sets in a 6:2:2 ratio for model training and tuning.

[0110] 4. Input the real-time calculated features into the trained traffic anomaly identification model. The traffic anomaly identification features are then fed into the model to obtain anomaly identification results and event probability. When a traffic anomaly is discovered, the network topology is used to trace the source of the anomaly upstream, and the traffic status of the upstream point is combined to determine whether an event has occurred.

[0111] 5. Combine the interval average travel time series in step 1 and the fused point topology in step 2 to predict the interval travel time prediction benchmark, extract and track target vehicles at upstream and downstream points in the interval, and calculate the vehicle overtime percentage and the number of consecutive overtime vehicles.

[0112] 6. Based on historical data analysis, the percentage of vehicles exceeding their designated time limit and the number of vehicles exceeding their designated time limit under normal traffic conditions are calculated to determine the threshold for identification. Each target vehicle is tracked in real time and its percentage of vehicles exceeding their designated time limit and the number of vehicles exceeding their designated time limit are calculated. If the percentage of vehicles exceeding their designated time limit or the number of vehicles exceeding their designated time limit exceeds the threshold, the system identifies the vehicle as abnormal.

[0113] 7. Fusion analysis is performed on the abnormality discrimination results in steps 4 and 6. The confidence levels of different event sources are set according to the event type. A hierarchical fusion strategy is used to finally fuse and obtain accurate traffic events.

[0114] 8. After an anomaly is discovered, continue to verify and monitor the point and interval conditions, and promptly report to relevant departments for multi-channel docking and information release, and continue to monitor the anomaly until the disposal is completed.

[0115] The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from a website, a computer, a server or a data center to another website, a computer, a server or a data center by a wired (e.g., infrared, wireless, microwave, etc.) method. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or a data center that includes one or more available media sets. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a magnetic tape), an optical medium (e.g., a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state drive.

[0116] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0117] As described above, although the present invention has been shown and described with reference to specific preferred embodiments, it should not be construed as limiting the present invention itself. Various changes may be made to it in form and detail without departing from the spirit and scope of the present invention as defined in the appended claims.

Claims

1. A multi-level highway incident identification and warning method based on adaptive network topology, characterized in that: The steps include: Use existing highway video cameras to collect video data; connect to the ETC system, and its gantry automatically records ETC data when vehicles pass through; Based on map resources and known fixed points, the connectivity of the intervals is analyzed according to the road directions in the video data and the vehicle travel paths in the ETC data, and whether there are direct traffic connections between the fixed points is determined, and an adaptive fusion point topology is constructed; Count the video data and ETC data of the predetermined time series to form historical traffic flow data and historical interval travel time; Establish traffic anomaly identification mechanisms based on traffic flow and interval travel time respectively; The collected real-time video data and real-time ETC data are substituted into a traffic anomaly identification mechanism based on traffic flow and / or a traffic anomaly identification mechanism based on interval travel time. If an abnormal event is determined, the source point causing the abnormality is traced back according to the adaptive fusion point topology, the abnormal road section is determined, and the abnormal event information is released.

2. The multi-level highway event identification and warning method based on adaptive network topology according to claim 1 is characterized in that: The known fixed points include the video camera points and the ETC gantry points; The adaptive fusion point topology includes the distance correlation and flow correlation between each fixed point.

3. The multi-level highway event identification and warning method based on adaptive network topology according to claim 2 is characterized in that: Calculate the actual spatial distance d between each fixed point AB , based on the actual space distance d AB , use the inverse proportional function method to analyze the distance correlation between points k is a preset coefficient.

4. The multi-level highway event identification and warning method based on adaptive network topology according to claim 2 is characterized in that: Counting traffic flow sequences between different points based on traffic flow information in the video data and ETC data; The Pearson correlation coefficient of the traffic flow sequence at each point is calculated to determine the traffic flow correlation between two video points; the value range of the correlation coefficient is between -1 and 1, and a value close to 1 indicates a high positive correlation, a value close to -1 indicates a high negative correlation, and a value close to 0 indicates a weak correlation.

5. The multi-level highway event identification and warning method based on adaptive network topology according to claim 1 is characterized in that: Establish a traffic anomaly identification mechanism based on traffic flow, specifically including: Counting the video data and ETC data of a predetermined time series to form historical traffic flow data; dividing the historical traffic flow data into two time periods, with the first time period serving as a horizontal reference and the second time period serving as a vertical reference; Constructing a neural network, inputting the horizontal benchmark into the neural network, and predicting the flow data of the next period as a prediction benchmark; The horizontal benchmark, the longitudinal benchmark, and the prediction benchmark are combined to form a traffic anomaly determination benchmark; The collected real-time traffic flow data is compared with the traffic anomaly judgment benchmark. If the current real-time traffic flow data exceeds the traffic anomaly judgment benchmark, the traffic flow is judged to be abnormal. The source point causing the anomaly is traced back according to the adaptive fusion point topology, the abnormal road section is determined, and the abnormal event information is released.

6. The multi-level highway event identification and warning method based on adaptive network topology according to claim 5 is characterized in that: Constructing a neural network, inputting the horizontal benchmark into the neural network, and predicting the flow data of the next period as a prediction benchmark; Calculate the percentage difference β between the predicted baseline and the actual traffic flow in the corresponding period in the historical traffic flow data: Where Q n is the actual traffic volume in the nth period, is the forecast benchmark for the nth period.

7. The multi-level highway event identification and warning method based on adaptive network topology according to claim 1 is characterized in that: Establish a traffic anomaly identification mechanism based on interval travel time, specifically including: The average travel time series of the historical interval is calculated by detecting targets at upstream and downstream points of the interval, recording the time when the same vehicle enters and exits the interval, subtracting the time to obtain the travel time of a single vehicle, and calculating the average travel time of the vehicles according to the predetermined time step to obtain the average travel time series of the historical interval; A multi-dimensional interval travel time prediction benchmark model is established by combining the impact of environmental data, including weather conditions and road construction conditions, on vehicle travel speed. This model is then fitted and trained using historical vehicle travel times. The trained model is then used to make real-time predictions of the next interval travel time. Based on the upstream and downstream relationship of the adaptive fusion point topology, the vehicles passing through the starting point of the interval are tracked at a preset frequency, and the time when the target vehicle enters the interval is recorded; if there are multiple downstream points at the point where the target vehicle is extracted, the target vehicle extraction frequency is increased, and it is ensured that the target vehicle passes through all downstream points; at the same time, the appearance features of the target vehicle are identified through the target detection algorithm based on deep learning, and the target vehicle extracted upstream is identified and matched at the downstream point of the interval, and the time when the target vehicle leaves the interval is recorded.

8. The multi-level highway event identification and warning method based on adaptive network topology according to claim 7 is characterized in that: Several verification points are selected at the downstream points of the section. If the target vehicle is not detected at the downstream points of the section but is detected at the verification points, the abnormal alarm is eliminated in time.

9. The multi-level highway event identification and warning method based on adaptive network topology according to claim 7 is characterized in that: Establishing a traffic anomaly identification mechanism based on interval travel time also includes: Combined with the travel time prediction value T predict The target vehicle’s theoretical departure time is calculated with the proportional factor α. If the target vehicle has not left the interval after exceeding the theoretical time, the target vehicle’s overtime percentage γ is recorded, and the number of consecutive overtime vehicles is counted, where: When the time between the monitoring point and the last extraction reaches the preset value, a new target vehicle extraction is carried out; each target vehicle is tracked in real time and the timeout percentage and the number of consecutive timeout vehicles of each target vehicle are calculated. If the vehicle timeout percentage or the number of consecutive timeout vehicles exceeds the threshold, it is judged as an abnormality.

10. An electronic device, characterized in that: include: a processor and a memory storing computer program instructions; When the processor executes the computer program instructions, the multi-level highway event identification and warning method based on adaptive network topology according to any one of claims 1 to 9 is implemented.

Images