Road abnormal event real-time monitoring and warning system

Abstract

The invention discloses a road abnormal event real-time monitoring and warning system, and the system comprises a data processing module which is used for collecting and preprocessing multi-source monitoring data; the region segmentation module is used for performing multi-scale region segmentation on the multi-source monitoring data by utilizing a pyramid scene analysis network to form a region segmentation map; the weight calculation module is used for constructing a weighted relation graph of road scene areas and abnormal event types and calculating the weight of the abnormal event in each area; the state evaluation module is used for updating node states in the graph and evaluating the occurrence probability of road abnormal events in different areas; the classification labeling module is used for carrying out risk classification on the abnormal events according to the state updating result and generating real-time alarm information; and the feedback adjustment module is used for dynamically adjusting network parameters according to the real-time monitoring result and alarm information feedback. According to the method, the pyramid scene analysis network, the graph attention neural network and the Bellman equation are combined, so that efficient and accurate road abnormal event real-time monitoring and warning are realized.

Description

Technical Field

[0001] The present invention relates to the technical field of road traffic monitoring and intelligent early warning, and in particular to a real-time monitoring and warning system for abnormal road events. Background Art

[0002] With rapid socioeconomic development and the continuous advancement of urbanization, the complexity of road traffic systems is increasing. Factors such as the dramatic increase in traffic volume, aging road infrastructure, and extreme weather have led to the frequent occurrence of abnormal road events. These events not only pose a traffic safety hazard but also place significant pressure on road management and maintenance. Abnormal road events, such as traffic accidents, road obstructions, and sudden natural disasters, directly impact traffic flow and driving safety, and can even trigger chain reactions leading to more serious social problems. Therefore, how to effectively monitor, warn, and respond to these abnormal events is a key issue in modern intelligent transportation systems.

[0003] Traditional roadside incident monitoring systems typically rely on a single data source, such as CCTV, radar sensors, or GPS data. These independent monitoring methods lack synergy, resulting in poor real-time and accurate event detection. Particularly on high-traffic urban roads or inclement weather conditions, traditional systems experience long response times and are prone to missed detections or false alarms. In this context, existing traffic monitoring systems often face challenges such as limited coverage, complex data processing, and poor response times, making them difficult to meet the demands of modern traffic management.

[0004] In recent years, with the rapid development of artificial intelligence and big data technologies, traffic monitoring and management have entered the era of intelligence. Existing intelligent transportation systems integrate a variety of sensors and monitoring devices to collect various road information and use data analysis techniques to identify and respond to events. These intelligent systems can capture large amounts of data in real time, such as traffic flow, vehicle speed, and weather conditions, thus supporting early warning of abnormal events. However, despite the continuous improvement of the technology level of intelligent transportation systems, existing technologies still have certain shortcomings in practical application. First, the multi-source and heterogeneous nature of data complicates data integration and analysis, and how to effectively process data from different devices remains an urgent problem. Second, existing abnormal event monitoring methods generally rely on rules and threshold settings and lack intelligent learning and adaptive capabilities. This results in the system being unable to respond promptly to new and unknown abnormal events.

[0005] While existing deep learning-based anomaly detection methods have achieved remarkable results in certain areas, their application to roadway anomaly monitoring still faces numerous challenges. For example, existing deep learning methods rely primarily on large amounts of labeled data for training, making it difficult to obtain high-quality labeled data in practical applications in the transportation sector. Furthermore, most existing methods rely on a single monitoring approach and lack the deep integration of multi-source information, making it difficult to accurately and comprehensively detect anomalies in complex road environments.

[0006] In recent years, road monitoring methods based on pyramid scene parsing networks and graph neural networks have gained increasing attention. Pyramid scene parsing networks have demonstrated strong capabilities in image processing through multi-scale feature extraction, effectively capturing spatial information at different scales within an image. Graph neural networks, by constructing a relationship graph between regions, can effectively process structured information within an image, fully accounting for inter-regional dependencies and improving the accuracy of event prediction. However, despite their broad application potential in road monitoring, these methods still have several shortcomings in real-time monitoring and alerting of abnormal road events. First, the feature extraction of pyramid scene parsing networks is still limited to a single data source, lacking the ability to process multi-source heterogeneous data and fully exploring potential correlations between data. Second, while graph neural networks can model inter-regional relationships, their high computational complexity makes real-time processing difficult in large-scale road scenarios.

[0007] Furthermore, dynamic programming methods based on the Bellman equation have been widely used in the field of optimization control. However, their use in monitoring and early warning systems for road abnormalities remains relatively limited. The Bellman equation dynamically updates states through optimal control strategies, thereby achieving optimal predictions of events. However, existing applications of the Bellman equation have primarily focused on static models or simple traffic flow models. These models have not effectively integrated actual road scenarios with the complex characteristics of abnormal events, and lack dynamic adjustment and adaptive capabilities, limiting the accuracy and real-time performance of predictions in complex traffic environments.

[0008] Therefore, how to provide a real-time monitoring and warning system for abnormal road events is a problem that those skilled in the art urgently need to solve. Summary of the Invention

[0009] One objective of this invention is to provide a real-time monitoring and warning system for abnormal road events. This system leverages a pyramid scene parsing network for multi-scale feature extraction, a graph attention neural network for constructing a weighted relationship graph between regions, and the Bellman equation for dynamic state updates and abnormal event probability assessment. Through these technical approaches, the system achieves accurate identification and timely warning of abnormal road events.

[0010] A real-time monitoring and warning system for abnormal road events according to an embodiment of the present invention includes:

[0011] Data processing module, used to collect multi-source monitoring data in the road area and perform pre-processing;

[0012] A region segmentation module is used to perform multi-scale region segmentation on the pre-processed multi-source monitoring data using a pyramid scene parsing network to form a region segmentation map of the road scene, wherein the region segmentation map includes regions of different scales and road abnormal event information of the corresponding regions;

[0013] The weight calculation module is used to construct a weighted relationship graph between road scene areas and abnormal event types, and calculate the weight of abnormal events in each area through the attention mechanism between nodes in the graph;

[0014] The state assessment module is used to combine the weighted relationship graph and the abnormal event weights to update the state of the nodes in the graph and evaluate the probability of abnormal road events in different areas;

[0015] The classification and labeling module is used to classify abnormal road events into low risk, medium risk, and high risk based on status update results, automatically label the risk level and time of occurrence of abnormal events, generate real-time alarm information, and notify relevant departments or emergency response systems;

[0016] The feedback adjustment module is used to dynamically adjust the parameters of the pyramid scene parsing network and the graph attention neural network based on real-time monitoring results and alarm information feedback to ensure timely response to abnormal road events.

[0017] Optionally, modules can be connected using the following methods:

[0018] S1. Collect multi-source monitoring data in the road area and perform pre-processing;

[0019] S2. Based on the pre-processed multi-source monitoring data, a pyramid scene parsing network is used to perform multi-scale region segmentation to form a region segmentation map of the road scene, wherein the region segmentation map includes regions of different scales and road abnormal event information of the corresponding regions;

[0020] S3. Based on the region segmentation graph, a graph attention neural network is used to construct a weighted relationship graph between road scene regions and abnormal event types, and the weight of abnormal events in each region is calculated through the attention mechanism between nodes in the graph;

[0021] S4. Combining the weighted relationship graph and the abnormal event weights, a dynamic programming method based on the Bellman equation is used to update the states of the nodes in the graph and evaluate the probability of abnormal road events occurring in different areas.

[0022] S5. Based on the status update results, road abnormal events are classified into low risk, medium risk, and high risk, and the risk level and time of occurrence of abnormal events are automatically marked. Real-time alarm information is generated and notified to relevant departments or emergency response systems;

[0023] S6. Based on real-time monitoring results and warning information feedback, dynamically adjust the parameters of the pyramid scene parsing network and graph attention neural network to ensure timely response to abnormal road events.

[0024] Optionally, the multi-source monitoring data includes video data captured by a vehicle driving recorder, image data captured by a traffic monitoring camera, distance data collected by a radar sensor, temperature data collected by a temperature sensor, and traffic flow monitoring data.

[0025] Optionally, the preprocessing includes removing noise, filling missing data, correcting sensor errors and data normalization.

[0026] Optionally, the S2 specifically includes:

[0027] S21. Perform multi-channel fusion encoding on the pre-processed multi-source monitoring data to generate a data fusion matrix, wherein the data fusion matrix includes red, green, and blue image channels, a radar range channel, a traffic flow density channel, and a road surface temperature channel. All channels are spatially aligned at a uniform sampling rate and normalized to the interval [0, 1].

[0028] S22. Input the data fusion matrix into the backbone feature extraction structure of the pyramid scene parsing network, and perform seven convolution operations in sequence. The convolution kernel sizes are 7×7, 3×3, 3×3, 1×1, 3×3, 3×3, and 1×1, respectively. Batch normalization layers and ReLU activation functions are nested between each convolution layer to generate a backbone feature map. The backbone feature map is reduced to one-fourth of the data fusion matrix in spatial dimension, and the number of channels is fixed at 512.

[0029] S23, copying the trunk feature map into four branches, respectively inputting the four average pooling channels of the pyramid scene parsing network pooling module, wherein the kernel sizes of the pooling channels are 1×1, 2×2, 3×3 and 6×6, respectively. Each channel extracts contextual semantic features of different scales, wherein 1×1 extracts global semantic information of the image, and 6×6 extracts structural features of local boundary areas. Each pooling output is followed by a 1×1 convolution for channel dimensionality reduction, and the number of output channels is 128;

[0030] S24, upsampling the four-way pooling results to the spatial size of the main feature map, and splicing them with the main feature map in the channel dimension to form a fused feature map. The number of channels of the fused feature map is 1024, and the splicing order is the main channel, 6×6 pooling, 3×3 pooling, 2×2 pooling and 1×1 pooling channel;

[0031] S25. Input the fused feature map into a feature compression module. The feature compression module consists of two consecutive 1×1 convolutions. After each convolution layer, batch normalization and ReLU activation operations are performed to output a fused feature map. The number of channels of the fused feature map after the feature compression module is 256.

[0032] S26. Input the fused feature map into the bilinear interpolation upsampling module for size reduction to restore it to the same spatial resolution as the data fusion matrix, and then input it into the pixel-by-pixel classification convolution layer. The number of output channels of the convolution layer is equal to the total number of abnormal event types, and each channel corresponds to a response confidence map of one abnormal type.

[0033] S27, performing maximum response value judgment on all confidence maps according to pixel positions, extracting the abnormal type label of each pixel point, and forming a regional label distribution map, wherein the regional label distribution map is composed of pixel-level semantic segmentation results of the road scene;

[0034] S28. Perform regional connectivity analysis based on the regional label distribution map, identify target regions with spatial consistency through a morphological dilation-erosion structure, and classify adjacent pixel blocks with the same abnormal label into the same region to form a road region structure map;

[0035] S29. Based on the boundary contours, internal consistency strength, and label credibility of each region in the road region structure map, a region redistribution optimization strategy is adopted to form a region segmentation map of the road scene. The region segmentation map consists of multiple regions with clear anomaly types, boundary closure, and scale descriptions.

[0036] Optionally, the pyramid scene parsing network extracts global semantic information, local edge features and mid-scale region relationships by inputting the backbone feature map into multiple context paths with different pooling kernel sizes, and performs channel splicing and fusion of multi-scale features on the basis of maintaining the integrity of the spatial structure. Then, a high-resolution, multi-semantic fusion feature map is constructed through feature compression and upsampling operations, thereby realizing accurate segmentation and structured expression of multiple types of abnormal event areas in complex road scenes.

[0037] Optionally, the S3 specifically includes:

[0038] S31, perform region coding processing on the region segmentation map, assign a unique number to each region with boundary closure and abnormal type label, and generate a region set Where n is the number of regions, each R i Represents a road abnormal area, including the attribute vector V i , the attribute vector V iIncluding region center coordinates, area, shape compactness, anomaly type label encoding and average response confidence;

[0039] S32, Each region R in i For nodes, construct an undirected graph G = (N, E), where N is the node set and E is the edge set. The edge set is established based on the Euclidean center distance between regions and the spatial adjacency relationship, and the initial value of the edge connection weight is defined Where d(R i ,R j ) represents region R i With R j Euclidean distance of the center point;

[0040] S33, the attribute vector V of each region i Input to the graph attention neural network to construct the node feature matrix H (0) =[V1; V2; ...; V n ], the node feature dimension is recorded as f, the graph attention neural network adopts a two-layer structure, each layer contains a multi-head attention module;

[0041] S34. In the l-th layer graph attention module, calculate the attention weight of the i-th node to the j-th adjacent node under the k-th attention head

[0042]

[0043] in, is the feature representation of node i in layer l, is the feature representation of node j in layer l, is the node j in layer l ′ The feature representation of is the feature map weight matrix of the lth layer under the kth attention head, a k is the attention weight vector of the kth attention head, for a k The transpose of is the connection weight between nodes i and j in layer l, are nodes i and j in layer l ′ The connection weight of is the set of adjacent nodes of node i, ‖ represents the vector concatenation operation, LeakyReLU is the activation function, and exp represents the natural exponential function with e as the base;

[0044] S35. Concatenate the output results of all attention heads to form the feature representation of the l+1 layer nodes And input it to the next layer of graph attention module, and finally after two layers of graph attention propagation, the updated feature representation of each node is obtained

[0045] S36, the feature representation of each node Input the fully connected weight layer and calculate the corresponding area R i The abnormal event weight ω i ,in W o is the output layer weight matrix, b o is the bias term, σ is the Sigmoid activation function, and the weight ω i Used to indicate the importance score of abnormal events in the area;

[0046] S37, with weight ω i Based on the topological structure of the undirected graph G, a weighted relationship graph between regions and event types is generated, and the weighted relationship graph is used for node status update and event probability estimation.

[0047] Optionally, the weighted relationship graph performs multi-head attention calculations on the attribute features of road area nodes by combining a graph attention neural network, dynamically adjusts edge weights based on the spatial distance and adjacency relationship between regions, and generates abnormal event weights of nodes through a Sigmoid activation function, thereby reflecting the relative importance of different regions in the occurrence of events, and generating the final weighted graph structure based on the weighted connection relationship between nodes.

[0048] Optionally, the S4 specifically includes:

[0049] S41. Take the weighted relationship graph and the abnormal event weight as input to construct the state transition model of the Bellman equation. i To make dynamic updates:

[0050]

[0051] in, represents the state of node i at time t, represents the state of node i at time t+1, The independent variable a when the function takes its maximum value i The value of S i is the instant reward of node i, corresponding to region R i The abnormal event weight ω i ,γ is the discount factor, which controls the weight of future rewards, P ij is the state transition probability from node i to node j, indicating the transition from region R i To area R j The likelihood of an event occurring, is the set of adjacent nodes of node i, representing the relationship with region R i Connected road area, a i is the action choice of node i, indicating that The action taken under is used to determine the state transition of the node;

[0052] S42, through iterative update Optimize the state of each node in the graph until the convergence condition is reached, and obtain the final state of each node under the optimal path Combined with the topological structure of the weighted relationship graph, calculate each region R i Probability of abnormal events occurring on inner roads:

[0053]

[0054] Among them, P(R i ) represents region R i The probability of an abnormal event occurring on the inner road, exp represents the natural exponential function with e as the base, n is the number of regions, and represents the total number of nodes in the graph;

[0055] S43, output probability P(R i ), and continues to iterate and update through the Bellman equation to ensure efficient assessment and real-time response of the probability of event occurrence.

[0056] Optionally, the S5 specifically includes:

[0057] S51, according to the area R i The probability of an abnormal event occurring P(R i ), set probability thresholds τ1, τ2, τ3 to classify the risks of abnormal events in the area, and the classification rules are:

[0058] When P(R i )≤τ1, that is, when the probability of an abnormal event is lower than the low-risk threshold, it is judged as a low-risk event, indicating that the area R i The possibility of abnormal events occurring within the period is low and requires continuous observation;

[0059] When τ1 <P(R i )≤τ2, that is, when the probability of an abnormal event is between low risk and medium risk, it is judged as a medium risk event, indicating that the area R i The possibility of abnormal events occurring within is high and requires priority attention;

[0060] When P(R i )>τ2, that is, when the probability of an abnormal event is higher than the medium risk threshold, it is judged as a high risk event, indicating that the area R i The possibility of abnormal events occurring within the system is extremely high, and emergency response measures must be taken immediately;

[0061] S52: Automatically generate risk level labels for abnormal events based on the classification results. The risk level labels include low risk, medium risk, and high risk, and generate risk level labels for each region R. i Mark the corresponding incident time, where the incident time is area R i The timestamp of the abnormal event, indicating the time when the event was first detected;

[0062] S53, according to each area R i The risk level label and the time of occurrence of the incident are used to generate real-time alarm information, which includes the area R i risk level, type of abnormal event, time of occurrence, and specific location of the abnormal event;

[0063] S54. Transmitting the real-time alarm information to relevant departments or emergency response systems. The transmission process ensures timely communication and response of the alarm information through an information transmission protocol.

[0064] S55. Based on the transmission feedback results, the alarm information sending mechanism is dynamically adjusted to optimize the transmission speed and accuracy of the alarm information, ensuring that all relevant departments can receive risk information and take response measures in the shortest possible time.

[0065] The beneficial effects of the present invention are:

[0066] First, this invention addresses the challenges of multi-source data fusion and complex event recognition in existing road anomaly monitoring technologies by combining a pyramid scene parsing network, a graph attention neural network, and the Bellman equation. The pyramid scene parsing network extracts deep features of road scenes at multiple scales, improving the ability to capture road anomalies. The graph attention neural network, by constructing a weighted relationship graph between regions, accurately describes the dependencies between road areas, effectively improving event recognition accuracy and system response speed.

[0067] Secondly, this invention utilizes a dynamic programming approach based on the Bellman equation to dynamically assess and update the probability of abnormal events. This innovative application enables the system to adjust the event prediction model in real time based on current status and historical data, further optimizing its ability to warn of abnormal events. This adaptive mechanism allows the system to flexibly respond to changing traffic conditions, avoiding the limitations of traditional methods that rely on fixed rules and threshold settings, and enhancing its flexibility in handling complex situations.

[0068] Finally, this invention achieves efficient road anomaly monitoring and alerting capabilities, with strong real-time and accuracy, playing a vital role in traffic management and emergency response. Through the deep integration of multi-source data and dynamic status assessment, the system not only promptly detects anomalies such as traffic accidents and road obstructions, but also predicts potential risks, issues early warnings, and helps relevant departments take effective countermeasures, thereby improving road traffic safety and emergency response efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0070] Figure 1 This is a module structure diagram of the real-time monitoring and warning system for abnormal road events proposed by the present invention;

[0071] Figure 2 This is a method flow chart of the real-time monitoring and warning system for abnormal road events proposed by the present invention;

[0072] Figure 3 This is a schematic diagram of the abnormal event classification and real-time alarm process of the real-time monitoring and alarm system for abnormal road events proposed by the present invention. DETAILED DESCRIPTION

[0073] The present invention will now be described in further detail with reference to the accompanying drawings, which are simplified schematic diagrams that illustrate the basic structure of the present invention in a schematic manner.

[0074] refer to Figure 1 , real-time monitoring and warning system for abnormal road events, including:

[0075] Data processing module, used to collect multi-source monitoring data in the road area and perform pre-processing;

[0076] A region segmentation module is used to perform multi-scale region segmentation on the pre-processed multi-source monitoring data using a pyramid scene parsing network to form a region segmentation map of the road scene, wherein the region segmentation map includes regions of different scales and road abnormal event information of the corresponding regions;

[0077] The weight calculation module is used to construct a weighted relationship graph between road scene areas and abnormal event types, and calculate the weight of abnormal events in each area through the attention mechanism between nodes in the graph;

[0078] The state assessment module is used to combine the weighted relationship graph and the abnormal event weights to update the state of the nodes in the graph and evaluate the probability of abnormal road events in different areas;

[0079] The classification and labeling module is used to classify abnormal road events into low risk, medium risk, and high risk based on status update results, automatically label the risk level and time of occurrence of abnormal events, generate real-time alarm information, and notify relevant departments or emergency response systems;

[0080] The feedback adjustment module is used to dynamically adjust the parameters of the pyramid scene parsing network and the graph attention neural network based on real-time monitoring results and alarm information feedback to ensure timely response to abnormal road events.

[0081] The present invention provides a real-time monitoring and warning system for abnormal road events. By collecting, preprocessing and fusing multi-source monitoring data, the system combines a pyramid scene parsing network to perform multi-scale regional segmentation of road scenes, then calculates the weighted relationship between regions through a graph attention neural network, and uses the Bellman equation for dynamic updating and probability evaluation, thereby achieving real-time monitoring, classification and warning of abnormal road events.

[0082] refer to Figure 2-3 In this embodiment, the modules are connected through the following methods:

[0083] S1. Collect multi-source monitoring data in the road area and perform pre-processing;

[0084] S2. Based on the pre-processed multi-source monitoring data, a pyramid scene parsing network is used to perform multi-scale region segmentation to form a region segmentation map of the road scene, wherein the region segmentation map includes regions of different scales and road abnormal event information of the corresponding regions;

[0085] S3. Based on the region segmentation graph, a graph attention neural network is used to construct a weighted relationship graph between road scene regions and abnormal event types, and the weight of abnormal events in each region is calculated through the attention mechanism between nodes in the graph;

[0086] S4. Combining the weighted relationship graph and the abnormal event weights, a dynamic programming method based on the Bellman equation is used to update the states of the nodes in the graph and evaluate the probability of abnormal road events occurring in different areas.

[0087] S5. Based on the status update results, road abnormal events are classified into low risk, medium risk, and high risk, and the risk level and time of occurrence of abnormal events are automatically marked. Real-time alarm information is generated and notified to relevant departments or emergency response systems;

[0088] S6. Based on real-time monitoring results and warning information feedback, dynamically adjust the parameters of the pyramid scene parsing network and graph attention neural network to ensure timely response to abnormal road events.

[0089] This invention provides a specific implementation method, including multi-channel data fusion, feature extraction using a pyramid scene parsing network, weight calculation using a graph attention neural network, and dynamic programming using the Bellman equation. By organically integrating these steps, the invention can fully utilize multi-source information when handling road anomaly events, achieving precise modeling across regions and improving event recognition accuracy and real-time response capabilities.

[0090] In this embodiment, the multi-source monitoring data includes video data captured by a vehicle driving recorder, image data captured by a traffic monitoring camera, distance data collected by a radar sensor, temperature data collected by a temperature sensor, and traffic flow monitoring data.

[0091] This invention utilizes multi-source monitoring data, including vehicle dashcam video data, traffic surveillance camera image data, radar sensor distance data, temperature sensor data, and traffic flow monitoring data, to achieve comprehensive monitoring of road anomalies. By integrating and processing this data, the invention effectively improves the comprehensiveness and accuracy of monitoring, ensuring that no anomalies are missed.

[0092] In this embodiment, the preprocessing includes removing noise, filling missing data, correcting sensor errors and data normalization.

[0093] This invention significantly improves the quality of raw monitoring data by optimizing data preprocessing, including denoising, filling missing data, correcting sensor errors, and normalizing data. This step ensures that subsequent data analysis can be performed based on high-quality data, thereby improving the accuracy and reliability of abnormal event detection.

[0094] In this embodiment, S2 specifically includes:

[0095] S21. Perform multi-channel fusion encoding on the pre-processed multi-source monitoring data to generate a data fusion matrix, wherein the data fusion matrix includes red, green, and blue image channels, a radar range channel, a traffic flow density channel, and a road surface temperature channel. All channels are spatially aligned at a uniform sampling rate and normalized to the interval [0, 1].

[0096] S22. Input the data fusion matrix into the backbone feature extraction structure of the pyramid scene parsing network, and perform seven convolution operations in sequence. The convolution kernel sizes are 7×7, 3×3, 3×3, 1×1, 3×3, 3×3, and 1×1, respectively. Batch normalization layers and ReLU activation functions are nested between each convolution layer to generate a backbone feature map. The backbone feature map is reduced to one-fourth of the data fusion matrix in spatial dimension, and the number of channels is fixed at 512.

[0097] S23, copying the trunk feature map into four branches, respectively inputting the four average pooling channels of the pyramid scene parsing network pooling module, wherein the kernel sizes of the pooling channels are 1×1, 2×2, 3×3 and 6×6, respectively. Each channel extracts contextual semantic features of different scales, wherein 1×1 extracts global semantic information of the image, and 6×6 extracts structural features of local boundary areas. Each pooling output is followed by a 1×1 convolution for channel dimensionality reduction, and the number of output channels is 128;

[0098] S24, upsampling the four-way pooling results to the spatial size of the main feature map, and splicing them with the main feature map in the channel dimension to form a fused feature map. The number of channels of the fused feature map is 1024, and the splicing order is the main channel, 6×6 pooling, 3×3 pooling, 2×2 pooling and 1×1 pooling channel;

[0099] S25. Input the fused feature map into a feature compression module. The feature compression module consists of two consecutive 1×1 convolutions. After each convolution layer, batch normalization and ReLU activation operations are performed to output a fused feature map. The number of channels of the fused feature map after the feature compression module is 256.

[0100] S26. Input the fused feature map into the bilinear interpolation upsampling module for size reduction to restore it to the same spatial resolution as the data fusion matrix, and then input it into the pixel-by-pixel classification convolution layer. The number of output channels of the convolution layer is equal to the total number of abnormal event types, and each channel corresponds to a response confidence map of one abnormal type.

[0101] S27, performing maximum response value judgment on all confidence maps according to pixel positions, extracting the abnormal type label of each pixel point, and forming a regional label distribution map, wherein the regional label distribution map is composed of pixel-level semantic segmentation results of the road scene;

[0102] S28. Perform regional connectivity analysis based on the regional label distribution map, identify target regions with spatial consistency through a morphological dilation-erosion structure, and classify adjacent pixel blocks with the same abnormal label into the same region to form a road region structure map;

[0103] S29. Based on the boundary contours, internal consistency strength, and label credibility of each region in the road region structure map, a region redistribution optimization strategy is adopted to form a region segmentation map of the road scene. The region segmentation map consists of multiple regions with clear anomaly types, boundary closure, and scale descriptions.

[0104] This paper optimizes the application of the pyramid scene parsing network, employing multi-scale pooling and feature fusion techniques to capture abnormal events in road scenes at different scales. Through feature compression and upsampling, it generates high-resolution feature maps with multi-semantic fusion, enabling precise segmentation and identification of abnormal event areas, effectively improving the accuracy of road anomaly detection.

[0105] In this embodiment, the pyramid scene parsing network extracts global semantic information, local edge features and mid-scale region relationships by inputting the backbone feature map into multiple context paths with different pooling kernel sizes, and performs channel splicing and fusion of multi-scale features on the basis of maintaining the integrity of the spatial structure. Then, a high-resolution, multi-semantic fusion feature map is constructed through feature compression and upsampling operations, thereby achieving accurate segmentation and structured expression of multiple types of abnormal event areas in complex road scenes.

[0106] By extracting global semantic information, local edge features and mesoscale regional relationships, and performing feature fusion while maintaining the integrity of the spatial structure, the present invention achieves accurate segmentation and structured expression of multiple types of road abnormal event areas, significantly improving the accuracy and processing efficiency of abnormal event detection, and enhancing the system's adaptability and real-time response capabilities in complex traffic environments.

[0107] In this embodiment, S3 specifically includes:

[0108] S31, perform region coding processing on the region segmentation map, assign a unique number to each region with boundary closure and abnormal type label, and generate a region set Where n is the number of regions, each R i Represents a road abnormal area, including the attribute vector V i , the attribute vector V i Including region center coordinates, area, shape compactness, anomaly type label encoding and average response confidence;

[0109] S32, Each region R in i For nodes, construct an undirected graph G = (N, E), where N is the node set and E is the edge set. The edge set is established based on the Euclidean center distance between regions and the spatial adjacency relationship, and the initial value of the edge connection weight is defined Where d(R i ,R j ) represents region R i With R j Euclidean distance of the center point;

[0110] S33, the attribute vector V of each region iInput to the graph attention neural network to construct the node feature matrix H (0) =[V1; V2; ...; V n ], the node feature dimension is recorded as f, the graph attention neural network adopts a two-layer structure, each layer contains a multi-head attention module;

[0111] S34. In the l-th layer graph attention module, calculate the attention weight of the i-th node to the j-th adjacent node under the k-th attention head

[0112]

[0113] in, is the feature representation of node i in layer l, is the feature representation of node j in layer l, is the node j in layer l ′ The feature representation of is the feature map weight matrix of the lth layer under the kth attention head, a k is the attention weight vector of the kth attention head, for a k The transpose of is the connection weight between nodes i and j in layer l, are nodes i and j in layer l ′ The connection weight of is the set of adjacent nodes of node i, ‖ represents the vector concatenation operation, LeakyReLU is the activation function, and exp represents the natural exponential function with e as the base;

[0114] S35. Concatenate the output results of all attention heads to form the feature representation of the l+1 layer nodes And input it to the next layer of graph attention module, and finally after two layers of graph attention propagation, the updated feature representation of each node is obtained

[0115] S36, the feature representation of each node Input the fully connected weight layer and calculate the corresponding area R i The abnormal event weight ω i ,in W o is the output layer weight matrix, b o is the bias term, σ is the Sigmoid activation function, and the weight ω i Used to indicate the importance score of abnormal events in the area;

[0116] S37, with weight ω iBased on the topological structure of the undirected graph G, a weighted relationship graph between regions and event types is generated, and the weighted relationship graph is used for node status update and event probability estimation.

[0117] This invention uses a graph attention neural network to model the spatial relationships between regions as a weighted graph and utilizes a multi-head attention mechanism to calculate the weights between nodes. This design effectively reflects the importance of different regions in the occurrence of an event, thereby improving the ability and accuracy of predicting abnormal road events. The generation of a weighted graph allows the system to consider the dependencies between regions, further enhancing the accuracy of monitoring and alerting.

[0118] In this embodiment, the weighted relationship graph performs multi-head attention calculation on the attribute features of road area nodes by combining the graph attention neural network. Based on the spatial distance and adjacency relationship between regions, the edge weight is dynamically adjusted and the abnormal event weight of the node is generated through the Sigmoid activation function, thereby reflecting the relative importance of different regions in the occurrence of events, and generating the final weighted graph structure according to the weighted connection relationship between nodes.

[0119] This method combines a graph attention neural network with multi-head attention calculations on the attribute features of road area nodes, dynamically adjusting the edge weights between regions to accurately reflect the relative importance of different regions in abnormal events. By using a sigmoid activation function to generate abnormal event weights for nodes and creating a weighted graph structure based on the weighted connections between nodes, the accuracy and flexibility of abnormal event detection are significantly improved, ensuring that the system can efficiently identify and promptly respond to potential risks in complex road environments.

[0120] In this embodiment, the S4 specifically includes:

[0121] S41. Take the weighted relationship graph and the abnormal event weight as input to construct the state transition model of the Bellman equation. i To make dynamic updates:

[0122]

[0123] in, represents the state of node i at time t, represents the state of node i at time t+1, The independent variable a when the function takes its maximum value i The value of S i is the instant reward of node i, corresponding to region R i The abnormal event weight ω i ,γ is the discount factor, which controls the weight of future rewards, P ij is the state transition probability from node i to node j, indicating the transition from region Ri To area R j The likelihood of an event occurring, is the set of adjacent nodes of node i, representing the relationship with region R i Connected road area, a i is the action choice of node i, indicating that The action taken under is used to determine the state transition of the node;

[0124] S42, through iterative update Optimize the state of each node in the graph until the convergence condition is reached, and obtain the final state of each node under the optimal path Combined with the topological structure of the weighted relationship graph, calculate each region R i Probability of abnormal events occurring on inner roads:

[0125]

[0126] Among them, P(R i ) represents region R i The probability of an abnormal event occurring on the inner road, exp represents the natural exponential function with e as the base, n is the number of regions, and represents the total number of nodes in the graph;

[0127] S43, output probability P(R i ), and continues to iterate and update through the Bellman equation to ensure efficient assessment and real-time response of the probability of event occurrence.

[0128] This method uses a state transition model based on the Bellman equation to dynamically assess the probability of abnormal events occurring in road areas. By iteratively updating node states and combining the topological structure of a weighted relationship graph, it accurately calculates the probability of abnormal events occurring in different areas, providing a scientific basis for subsequent event risk classification. This method improves the real-time and accuracy of abnormal event prediction.

[0129] In this embodiment, the S5 specifically includes:

[0130] S51, according to the area R i The probability of an abnormal event occurring P(R i ), set probability thresholds τ1, τ2, τ3 to classify the risks of abnormal events in the area, and the classification rules are:

[0131] When P(R i )≤τ1, that is, when the probability of an abnormal event is lower than the low-risk threshold, it is judged as a low-risk event, indicating that the area R i The possibility of abnormal events occurring within the period is low and requires continuous observation;

[0132] When τ1 <P(R i)≤τ2, that is, when the probability of an abnormal event is between low risk and medium risk, it is judged as a medium risk event, indicating that the area R i The possibility of abnormal events occurring within is high and requires priority attention;

[0133] When P(R i )>τ2, that is, when the probability of an abnormal event is higher than the medium risk threshold, it is judged as a high risk event, indicating that the area R i The possibility of abnormal events occurring within the system is extremely high, and emergency response measures must be taken immediately;

[0134] S52: Automatically generate risk level labels for abnormal events based on the classification results. The risk level labels include low risk, medium risk, and high risk, and generate risk level labels for each region R. i Mark the corresponding incident time, where the incident time is area R i The timestamp of the abnormal event, indicating the time when the event was first detected;

[0135] S53, according to each area R i The risk level label and the time of occurrence of the incident are used to generate real-time alarm information, which includes the area R i risk level, type of abnormal event, time of occurrence, and specific location of the abnormal event;

[0136] S54. Transmitting the real-time alarm information to relevant departments or emergency response systems. The transmission process ensures timely communication and response of the alarm information through an information transmission protocol.

[0137] S55. Based on the transmission feedback results, the alarm information sending mechanism is dynamically adjusted to optimize the transmission speed and accuracy of the alarm information, ensuring that all relevant departments can receive risk information and take response measures in the shortest possible time.

[0138] This invention uses an automated risk classification mechanism to assess and categorize abnormal events in road areas, generating real-time alerts and notifying relevant departments or emergency response systems. By dynamically adjusting the alert information transmission mechanism, this invention optimizes the speed and accuracy of alert transmission, ensuring that relevant personnel can respond to incidents and take effective emergency measures in the shortest possible time. This innovative design significantly improves the response efficiency and emergency response capabilities of road traffic safety management.

[0139] Example 1:

[0140] To verify the feasibility of the present invention, we applied it to a traffic monitoring system in a certain city. Monitoring abnormal traffic events in this city has always been a major challenge for the traffic management department. Traffic accidents, road obstructions, and other abnormal events frequently occur during peak hours, especially during peak hours. Traditional monitoring methods often struggle to effectively address these issues due to their limited data sources and slow response times.

[0141] In this embodiment, the system collects multi-source data from traffic surveillance cameras, road sensors, dashcams, and other monitoring devices, including road videos, vehicle speed, temperature, and traffic flow. The data processing module first performs denoising, fills in missing data, corrects sensor errors, and normalizes the data. The processed data is then passed to the region segmentation module, which uses a pyramid scene parsing network to perform multi-scale region segmentation on the road scene, generating a region segmentation map.

[0142] After the region segmentation map is generated, the system treats each region as a node and uses a graph attention neural network to calculate a weighted relationship graph between regions. The graph attention neural network performs multi-head attention calculations based on the attributes of each region (such as region size, shape, and abnormal event type), dynamically adjusts the edge weights between regions, and uses a sigmoid activation function to generate abnormal event weights for nodes. These weighted relationship graphs help the system reflect the relative importance of different regions in abnormal events, ultimately generating a weighted graph structure.

[0143] Then, using dynamic programming methods based on the Bellman equation, the system iteratively updates the probability of abnormal events occurring in each area based on current status and historical data. These probabilities are then passed to the classification and labeling module. Based on the assessment results, the system classifies abnormal events by risk level and automatically generates real-time alerts. Alert information includes event type, location, time of occurrence, and risk level, ensuring that traffic management departments receive and process the information promptly.

[0144] In practical applications, the system not only detects common anomalies such as traffic accidents and road obstructions, but also effectively identifies more subtle anomalies, such as flooded areas caused by sudden weather events and temporary traffic control areas. Once these events are identified and assessed as high-risk by the system, relevant departments receive an alert within seconds, enabling rapid response to ensure traffic safety.

[0145] To verify the effectiveness of the present invention, the system was tested on the city's main traffic arteries and in areas prone to traffic accidents. Test data showed that over the past three months, the system detected 98% of traffic anomalies, with a false alarm rate of less than 2%. All high-risk incidents were accurately alerted and notified to relevant departments within three minutes of the incident. Compared to traditional systems (which have a false alarm rate of 10% and a processing time of over 10 minutes), the present invention significantly improves accuracy and response time.

[0146] Specific data shows that on the main roads and busy traffic areas of the city, the probability distribution of different types of abnormal events monitored by the system is as follows:

[0147] Table 1 Road abnormal event monitoring results and response time statistics

[0148]

[0149]

[0150] As shown in Table 1, the system accurately identifies the risk level of each detected traffic anomaly and promptly notifies relevant departments through alerts. Alarm response times are consistently under 5 seconds, ensuring that anomalies are promptly addressed. Furthermore, the false alarm rate is effectively controlled, below 10% of traditional systems.

[0151] The application of this example demonstrates that the present invention's real-time monitoring and alerting method for abnormal road events offers significant advantages in multi-source data fusion, precise event detection and classification, and rapid response. In particular, the present invention significantly improves the efficiency and response speed of road traffic safety management by enabling timely identification and handling of high-risk events, providing a more efficient and intelligent solution for road management departments.

[0152] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. Real-time monitoring and warning system for abnormal road events, characterized by: include: Data processing module, used to collect multi-source monitoring data in the road area and perform pre-processing; A region segmentation module is used to perform multi-scale region segmentation on the pre-processed multi-source monitoring data using a pyramid scene parsing network to form a region segmentation map of the road scene, wherein the region segmentation map includes regions of different scales and road abnormal event information of the corresponding regions; The weight calculation module is used to construct a weighted relationship graph between road scene areas and abnormal event types, and calculate the weight of abnormal events in each area through the attention mechanism between nodes in the graph; The state assessment module is used to combine the weighted relationship graph and the abnormal event weights to update the state of the nodes in the graph and evaluate the probability of abnormal road events in different areas; The classification and labeling module is used to classify abnormal road events into low risk, medium risk, and high risk based on status update results, automatically label the risk level and time of occurrence of abnormal events, generate real-time alarm information, and notify relevant departments or emergency response systems; The feedback adjustment module is used to dynamically adjust the parameters of the pyramid scene parsing network and the graph attention neural network based on real-time monitoring results and alarm information feedback to ensure timely response to abnormal road events.

2. The real-time monitoring and warning system for abnormal road events according to claim 1 is characterized in that: The modules are implemented as follows: S1. Collect multi-source monitoring data in the road area and perform pre-processing; S2. Based on the pre-processed multi-source monitoring data, a pyramid scene parsing network is used to perform multi-scale region segmentation to form a region segmentation map of the road scene, wherein the region segmentation map includes regions of different scales and road abnormal event information of the corresponding regions; S3. Based on the region segmentation graph, a graph attention neural network is used to construct a weighted relationship graph between road scene regions and abnormal event types, and the weight of abnormal events in each region is calculated through the attention mechanism between nodes in the graph; S4. Combining the weighted relationship graph and the abnormal event weights, a dynamic programming method based on the Bellman equation is used to update the states of the nodes in the graph and evaluate the probability of abnormal road events occurring in different areas. S5. Based on the status update results, road abnormal events are classified into low risk, medium risk, and high risk, and the risk level and time of occurrence of abnormal events are automatically marked. Real-time alarm information is generated and notified to relevant departments or emergency response systems; S6. Based on real-time monitoring results and warning information feedback, dynamically adjust the parameters of the pyramid scene parsing network and graph attention neural network to ensure timely response to abnormal road events.

3. The real-time monitoring and warning system for abnormal road events according to claim 2 is characterized in that: The multi-source monitoring data includes video data captured by vehicle driving recorders, image data captured by traffic monitoring cameras, distance data collected by radar sensors, temperature data collected by temperature sensors, and traffic flow monitoring data.

4. The real-time monitoring and warning system for abnormal road events according to claim 2 is characterized in that: The preprocessing includes noise removal, filling missing data, correcting sensor errors and data normalization.

5. The real-time monitoring and warning system for abnormal road events according to claim 2 is characterized in that: The S2 specifically includes: S21. Perform multi-channel fusion encoding on the pre-processed multi-source monitoring data to generate a data fusion matrix, wherein the data fusion matrix includes red, green, and blue image channels, a radar range channel, a traffic flow density channel, and a road surface temperature channel. All channels are spatially aligned at a uniform sampling rate and normalized to the interval [0, 1]. S22. Input the data fusion matrix into the backbone feature extraction structure of the pyramid scene parsing network, and perform seven convolution operations in sequence. The convolution kernel sizes are 7×7, 3×3, 3×3, 1×1, 3×3, 3×3, and 1×1, respectively. Batch normalization layers and ReLU activation functions are nested between each convolution layer to generate a backbone feature map. The backbone feature map is reduced to one-fourth of the data fusion matrix in spatial dimension, and the number of channels is fixed at 512. S23, copying the trunk feature map into four branches, respectively inputting the four average pooling channels of the pyramid scene parsing network pooling module, wherein the kernel sizes of the pooling channels are 1×1, 2×2, 3×3 and 6×6, respectively. Each channel extracts contextual semantic features of different scales, wherein 1×1 extracts global semantic information of the image, and 6×6 extracts structural features of local boundary areas. Each pooling output is followed by a 1×1 convolution for channel dimensionality reduction, and the number of output channels is 128; S24, upsampling the four-way pooling results to the spatial size of the main feature map, and splicing them with the main feature map in the channel dimension to form a fused feature map. The number of channels of the fused feature map is 1024, and the splicing order is the main channel, 6×6 pooling, 3×3 pooling, 2×2 pooling and 1×1 pooling channel; S25. Input the fused feature map into a feature compression module. The feature compression module consists of two consecutive 1×1 convolutions. After each convolution layer, batch normalization and ReLU activation operations are performed to output a fused feature map. The number of channels of the fused feature map after the feature compression module is 256. S26. Input the fused feature map into the bilinear interpolation upsampling module for size reduction to restore it to the same spatial resolution as the data fusion matrix, and then input it into the pixel-by-pixel classification convolution layer. The number of output channels of the convolution layer is equal to the total number of abnormal event types, and each channel corresponds to a response confidence map of one abnormal type. S27, performing maximum response value judgment on all confidence maps according to pixel positions, extracting the abnormal type label of each pixel point, and forming a regional label distribution map, wherein the regional label distribution map is composed of pixel-level semantic segmentation results of the road scene; S28. Perform regional connectivity analysis based on the regional label distribution map, identify target regions with spatial consistency through a morphological dilation-erosion structure, and classify adjacent pixel blocks with the same abnormal label into the same region to form a road region structure map; S29. Based on the boundary contours, internal consistency strength, and label credibility of each region in the road region structure map, a region redistribution optimization strategy is adopted to form a region segmentation map of the road scene. The region segmentation map consists of multiple regions with clear anomaly types, boundary closure, and scale descriptions.

6. The real-time monitoring and warning system for abnormal road events according to claim 5 is characterized in that: The pyramid scene parsing network extracts global semantic information, local edge features, and mid-scale region relationships by inputting backbone feature maps into multiple context paths with different pooling kernel sizes. It then performs channel concatenation and fusion on multi-scale features while maintaining the integrity of the spatial structure. It then constructs a high-resolution, multi-semantic fusion feature map through feature compression and upsampling operations, achieving accurate segmentation and structured representation of multiple types of abnormal event areas in complex road scenes.

7. The real-time monitoring and warning system for abnormal road events according to claim 2 is characterized in that: The S3 specifically includes: S31, perform region coding processing on the region segmentation map, assign a unique number to each region with boundary closure and abnormal type label, and generate a region set Where n is the number of regions, each R i Represents a road abnormal area, including the attribute vector V i , the attribute vector V i Including region center coordinates, area, shape compactness, anomaly type label encoding and average response confidence; S32, Each region R in i For nodes, construct an undirected graph G = (N, E), where N is the node set and E is the edge set. The edge set is established based on the Euclidean center distance between regions and the spatial adjacency relationship, and the initial value of the edge connection weight is defined Where d(R i ,R j ) represents region R i With R j Euclidean distance of the center point; S33, the attribute vector V of each region i Input to the graph attention neural network to construct the node feature matrix H (0) =[V1; V2; ...; V n ], the node feature dimension is recorded as f, the graph attention neural network adopts a two-layer structure, each layer contains a multi-head attention module; S34. In the l-th layer graph attention module, calculate the attention weight of the i-th node to the j-th adjacent node under the k-th attention head in, is the feature representation of node i in layer l, is the feature representation of node j in layer l, is the feature representation of node j′ in layer l, is the feature map weight matrix of the lth layer under the kth attention head, a k is the attention weight vector of the kth attention head, for a k The transpose of is the connection weight between nodes i and j in layer l, is the connection weight between nodes i and j′ in layer l, is the set of adjacent nodes of node i, ‖ represents the vector concatenation operation, LeakyReLU is the activation function, and exp represents the natural exponential function with e as the base; S35. Concatenate the output results of all attention heads to form the feature representation of the l+1 layer nodes And input it to the next layer of graph attention module, and finally after two layers of graph attention propagation, the updated feature representation of each node is obtained S36, the feature representation of each node Input the fully connected weight layer and calculate the corresponding area R i The abnormal event weight ω i ,in W o is the output layer weight matrix, b o is the bias term, σ is the Sigmoid activation function, and the weight ω i Used to indicate the importance score of abnormal events in the area; S37, with weight ω i Based on the topological structure of the undirected graph G, a weighted relationship graph between regions and event types is generated, and the weighted relationship graph is used for node status update and event probability estimation.

8. The real-time monitoring and warning system for abnormal road events according to claim 7 is characterized in that: The weighted relationship graph performs multi-head attention calculation on the attribute features of road area nodes by combining a graph attention neural network. Based on the spatial distance and adjacency relationship between regions, the edge weights are dynamically adjusted and the abnormal event weights of the nodes are generated through the Sigmoid activation function, thereby reflecting the relative importance of different regions in the occurrence of events. The final weighted graph structure is generated according to the weighted connection relationship between nodes.

9. The real-time monitoring and warning system for abnormal road events according to claim 2 is characterized in that: The S4 specifically includes: S41. Take the weighted relationship graph and the abnormal event weight as input to construct the state transition model of the Bellman equation. i To make dynamic updates: in, represents the state of node i at time t, represents the state of node i at time t+1, The independent variable a when the function takes its maximum value i The value of S i is the instant reward of node i, corresponding to region R i The abnormal event weight ω i ,γ is the discount factor, which controls the weight of future rewards, P ij is the state transition probability from node i to node j, indicating the transition from region R i To area R j The likelihood of an event occurring, is the set of adjacent nodes of node i, representing the relationship with region R i Connected road area, a i is the action choice of node i, indicating that The action taken under is used to determine the state transition of the node; S42, through iterative update Optimize the state of each node in the graph until the convergence condition is reached, and obtain the final state of each node under the optimal path Combined with the topological structure of the weighted relationship graph, calculate each region R i Probability of abnormal events occurring on inner roads: Among them, P(R i ) represents region R i The probability of an abnormal event occurring on the inner road, exp represents the natural exponential function with e as the base, n is the number of regions, and represents the total number of nodes in the graph; S43, output probability P(R i ), and continues to iterate and update through the Bellman equation to ensure efficient assessment and real-time response of the probability of event occurrence.

10. The real-time monitoring and warning system for abnormal road events according to claim 2, characterized in that: The S5 specifically includes: S51, according to the area R i The probability of an abnormal event occurring P(R i ), set probability thresholds τ1, τ2, τ3 to classify the risks of abnormal events in the area, and the classification rules are: When P(R i )≤τ1, that is, when the probability of an abnormal event is lower than the low-risk threshold, it is judged as a low-risk event, indicating that the area R i The possibility of abnormal events occurring within the period is low and requires continuous observation; When τ1 <P(R i )≤τ2, that is, when the probability of an abnormal event is between low risk and medium risk, it is judged as a medium risk event, indicating that the area R i The possibility of abnormal events occurring within is high and requires priority attention; When P(R i )>τ2, that is, when the probability of an abnormal event is higher than the medium risk threshold, it is judged as a high risk event, indicating that the area R i The possibility of abnormal events occurring within the system is extremely high, and emergency response measures must be taken immediately; S52: Automatically generate risk level labels for abnormal events based on the classification results. The risk level labels include low risk, medium risk, and high risk, and generate risk level labels for each region R. i Mark the corresponding incident time, where the incident time is area R i The timestamp of the abnormal event, indicating the time when the event was first detected; S53, according to each area R i The risk level label and the time of occurrence of the incident are used to generate real-time alarm information, which includes the area R i risk level, type of abnormal event, time of occurrence, and specific location of the abnormal event; S54. Transmitting the real-time alarm information to relevant departments or emergency response systems. The transmission process ensures timely communication and response of the alarm information through an information transmission protocol. S55. Based on the transmission feedback results, the alarm information sending mechanism is dynamically adjusted to optimize the transmission speed and accuracy of the alarm information, ensuring that all relevant departments can receive risk information and take response measures in the shortest possible time.

Images