Distributed sensor abnormal event identification method for intelligent traffic

Abstract

The invention discloses a distributed sensor abnormal event identification method for intelligent traffic, and particularly relates to the technical field of traffic information perception and identification, and the method comprises the following steps: generating an abnormal information initial confidence value through deploying a sensor node with a confidence value dynamic adjustment function; during fusion processing, independent identification priorities of single node anomalies are reserved; after sequence reconstruction of a unified time reference, judging whether a response condition is met or not in combination with trajectory evolution; if yes, an abnormal response process is triggered, a space-time compensation set is constructed based on historical data of the sensing blind area for verification, and finally an abnormal intervention instruction is output and a traffic scheduling system is linked; according to the method, the sensing sensitivity, the fusion accuracy and the response timeliness of the abnormal information in a complex traffic environment are improved, the problem of an identification blind area that single-point abnormity is covered is avoided, state reconstruction and supplementary verification of the sensing blind area are realized, and the rapid intervention capability of an intelligent traffic system on emergencies is enhanced.

Description

Technical Field

[0001] This invention relates to the field of traffic information perception and recognition technology, and more specifically, to a method for identifying abnormal events using distributed sensors in intelligent transportation. Background Technology

[0002] With the continuous improvement of intelligent transportation infrastructure, distributed deployment of multiple types of sensors has become an important means to achieve real-time perception of traffic conditions and identification of emergencies in urban road environments. Through the coordinated operation of multi-source sensing devices such as video surveillance, environmental detection, and radar sensing, current intelligent transportation systems can obtain a relatively comprehensive understanding of the road network's operating status and rely on multi-source data fusion strategies to identify and respond to abnormal events.

[0003] In practical applications, existing multi-source fusion mechanisms generally rely on data consistency judgment between sensors as the core basis for event confirmation, leading to the frequent occurrence of the problem that "a few anomalies are masked by a majority of normal data." This is especially true in scenarios of sudden, micro-level extreme events, such as electric vehicle fires, small-scale explosions, localized high-concentration smoke, and severe air pressure fluctuations. Because these events often have characteristics such as small spatial range, short duration, and weak signal features, they can only be detected by a very small number of sensor nodes individually. In such cases, if other sensors fail to detect the same event simultaneously due to factors such as obstructed viewpoints, blind spots, or inconsistent sampling periods, the fusion strategy often misjudges the abnormal signal as noise or occasional error, resulting in the event not being identified and triggering a response mechanism in the initial stages.

[0004] This problem is particularly prominent in open urban transportation environments. Areas such as non-motorized vehicle lanes, bus stops, underpass entrances, and under overpasses are often in blind spots of monitoring coverage or areas with sparse multi-sensor perception. If an early fire or abnormal behavior occurs in such areas, and the system cannot accurately identify and respond in the first instance, it can easily escalate into large-scale traffic congestion, casualties, or secondary accident risks, seriously affecting traffic safety and the city's emergency response capabilities.

[0005] Essentially, existing technologies lack the ability to independently identify "isolated high-risk anomaly signals" during the process of anomaly information fusion and judgment, and have not established a mechanism for dynamic adjustment and reinforcement of anomaly confidence. Furthermore, existing methods lack effective anomaly supplementary verification paths within perception blind spots, resulting in low identification accuracy and delayed response decisions when the entire system faces sudden micro-events. Therefore, this invention proposes a distributed sensor anomaly event identification method for intelligent transportation to address the aforementioned problems. Summary of the Invention

[0006] To achieve the above objectives, the present invention provides the following technical solution: A method for identifying abnormal events from distributed sensors in intelligent transportation includes the following steps: Distributed sensor nodes with dynamic confidence value adjustment function are deployed in the traffic monitoring area. The sensor nodes generate the initial confidence value of the corresponding abnormal information based on the continuous characteristics of historical monitoring data, the magnitude of changes in environmental background and the degree of deviation of the current sampling data. The initial confidence values ​​are transmitted to the central processing unit. The central processing unit performs normalization processing on each initial confidence value based on the spatial distribution relationship of sensor nodes, the intersection ratio of sensing areas, and the collaborative correlation of historical identifications. During the data fusion process, the independent identification priority of abnormal information from a single sensor node is retained. The normalized abnormal information is reconstructed in sequence under a unified time reference. The central processing unit reassembles the abnormal information uploaded at different times in sequence according to the set time window to ensure that the abnormal information detected by a single sensor node in the early stage of the event can fully participate in the subsequent identification process. After the sequence reconstruction is completed, the central processing unit performs a local state analysis process based on the spatial distribution characteristics of the current abnormal information, the normalized confidence value, and the historical abnormal probability of the corresponding area. It combines the changes in sensor data in the vicinity of the abnormal information, traffic flow state deviation, and the evolution trend of the target trajectory to determine whether the event response conditions are met. When the judgment result meets the set conditions, the abnormal response process is immediately initiated without relying on the data consistency judgment of multiple sensor nodes. After the anomaly response process is initiated, the central processing unit constructs a spatiotemporal compensation data set for the area within the perception blind zone related to the current anomaly information, based on historical image fragments, trajectory records, and regional behavior evolution data. It then reconstructs the state of the physically limited area and completes supplementary verification based on the reconstruction results. When the verification results meet the response judgment criteria, it outputs an anomaly intervention command and links it to the intelligent traffic dispatch system.

[0007] In a preferred embodiment, generating the initial confidence value corresponding to the anomaly information includes the following steps: The sensor nodes continuously collect monitoring data based on a set time window, and analyze the continuous characteristics of historical monitoring data to extract the changing trends and statistical stability of various preset data. The sensor node compares and analyzes the environmental background information at the current sampling location, calculates the difference between the current monitoring state and the long-term stable state, and quantifies the magnitude of change in the environmental background. The sensor node constructs a multi-factor confidence calculation model based on the continuous characteristics of historical monitoring data, the magnitude of changes in the environmental background, and the degree of deviation of the current sampled data, and generates the initial confidence value of the current anomaly information.

[0008] In a preferred embodiment, the multi-factor confidence calculation model is constructed through the following steps: Based on the continuous characteristics of historical monitoring data, the magnitude of changes in environmental background, and the degree of deviation of current sampling data, each item is compared with the preset anomaly criteria. Input factors that are irrelevant to the current anomaly or do not meet the screening conditions are eliminated, and only factors that are significantly related to the current situation are retained for subsequent processing. For the retained input factors, a linear mapping process is performed based on the actual impact of each factor on the false alarm rate and false negative rate in anomaly identification during long-term observation. The original values ​​are transformed into confidence contribution values ​​under a unified dimension, while maintaining the relative weight ratio between each factor. The mapped confidence contribution values ​​are weighted and summed to generate the initial confidence value of the current anomaly information.

[0009] In a preferred embodiment, normalizing the initial confidence value means: The central processing unit performs coordinate mapping on the spatial distribution relationship of each sensor node, constructs a two-dimensional spatial grid, and calculates the relative distance between each node to determine the degree of correlation between nodes in the spatial structure. The central processing unit combines the sensing area covered by the sensor nodes to calculate the intersection ratio of the sensing areas between adjacent nodes, which is used to adjust the influence factor of the initial confidence value in the fusion and weaken the interference of edge nodes on the judgment of multi-region data. The central processing unit retrieves historical identification data and assigns corresponding confidence values ​​to different nodes based on the frequency and consistency of their collaborative responses in similar events. It then incorporates spatial correlation, the intersection ratio of perceived areas, and the collaborative correction values ​​into the normalization calculation process to generate a set of comparable confidence values ​​under a unified scale.

[0010] In a preferred embodiment, incorporating spatial correlation, the intersection ratio of perceived regions, and the collaborative correction value into the normalization calculation process refers to: The central processing unit performs historical behavior analysis on each sensor node. First, it counts the number of times the node actually participated in the identification of the same type of anomaly events that have been labeled, and the number of times the node's results are consistent with the final identification results in these events, and calculates its historical response consistency rate. The collaborative correction value of each node is generated by multiplying its historical response consistency rate by its participation frequency, which is used to measure its historical reliability in group judgment. The central processing unit multiplies the collaborative correction value by the spatial correlation value and the intersection ratio of the sensing area corresponding to the node, and uses it as a weighted correction factor for the initial confidence value of the node. The central processing unit performs maximum-minimum interval linear normalization on all corrected confidence values, so that the confidence values ​​of all nodes fall within a preset uniform numerical range, generating a standardized set of confidence values.

[0011] In a preferred embodiment, when the central processing unit performs confidence value fusion judgment, for abnormal information generated by only a single sensor node and whose initial confidence value after normalization exceeds a set initial threshold, the abnormal information is retained to enter the judgment process through an independent channel, which is not limited by whether other nodes have the same abnormal information; this independent channel is used to ensure that the system can still respond effectively to isolated abnormal information in the event of missing responses from multiple nodes.

[0012] In a preferred embodiment, sequence reconstruction includes the following steps: Based on the upload time of each abnormal information, the central processing unit performs timestamp alignment on the normalized abnormal information to construct a unified timeline for unifying the starting benchmark for event perception. The central processing unit arranges the abnormal information with time differences in chronological order according to the set time window, forming a cross-node time sequence information sequence, which is used to reconstruct the development process of abnormal events.

[0013] In a preferred embodiment, determining whether the event response conditions are met includes the following steps: The central processing unit determines the occurrence time and node source of each abnormal information based on the cross-node time sequence of information, and extracts the sensor node that uploaded the abnormal information earliest and sets it as the dominant source node. The normalized confidence value of the dominant source node and the historical anomaly frequency of the region are compared with the spatial confidence threshold and the spatial risk threshold, respectively. If both exceed the threshold, a potential risk is identified. Based on the location information continuously reported by the dominant source node and its neighboring nodes, the cumulative amount of movement distance and the cumulative amount of direction change angle of the preset target object are counted within a set time window. If both exceed the displacement threshold and the deflection threshold respectively, it is determined that the target object has abnormal trajectory behavior. When there are potential risks in the primary source node and the target object's trajectory behavior is abnormal, the abnormal information is deemed to meet the conditions for triggering the response process.

[0014] In a preferred embodiment, after the anomaly response process is initiated, the supplementary verification includes the following steps: The central processing unit determines the perception blind zone area corresponding to the current abnormal information, and extracts the trajectory distribution density, behavioral state change frequency and abnormal event occurrence probability of the area in the historical time period corresponding to the time of the abnormal event, and constructs a structured historical feature set, i.e., a spatiotemporal compensation data set. The central processing unit extracts the real-time data change trends before and after the anomaly based on the time and location range of the current anomaly information, and quickly matches them with the structured historical feature groups in the structured historical feature set. When the similarity score between the current state and any set of structured historical feature groups meets the preset similarity conditions, the central processing unit outputs an abnormal intervention command and links it to the intelligent traffic dispatch system to execute the preset traffic control and guidance strategy.

[0015] The technical effects and advantages of this invention are as follows: This invention deploys distributed sensor nodes with dynamic confidence value adjustment capabilities. By utilizing the continuous characteristics of historical monitoring data, the magnitude of changes in the environmental background, and the offset of the current sampled data, it generates initial confidence values ​​for identifying anomalies. This solves the problem of traditional sensor nodes being insensitive to complex environmental changes under fixed threshold conditions. This mechanism enables sensor nodes to dynamically adjust their current monitoring state, providing a basis for judgment before anomaly characteristics have fully spread, thus offering more reliable input data for subsequent processing.

[0016] This invention normalizes the initial confidence value through a central processing unit and establishes a multi-dimensional fusion model by comprehensively considering the spatial distribution relationship of sensor nodes, the intersection ratio of sensing areas, and the collaborative correlation of historical identifications. During the fusion process, the independent identification priority of anomaly information from a single sensor node is specifically preserved. This ensures that even when multiple nodes fail to respond synchronously, anomaly information with high confidence values ​​will not be prematurely eliminated due to a lack of collaborative feedback, effectively avoiding the blind spot problem of "a few anomalies being masked by a majority of normalities" in existing technologies. After completing the temporal reconstruction of the normalized anomaly information, this invention performs joint analysis on the spatial distribution characteristics, normalized confidence values, and historical anomaly probabilities of the current anomaly information through the central processing unit. Furthermore, it combines changes in sensor data from neighboring areas, traffic flow state deviations, and target trajectory evolution trends to achieve accurate identification of event response conditions. By eliminating the dependence on multi-node consistency judgment, the system's response sensitivity to sudden, localized, or asymmetric traffic anomalies is improved, effectively shortening the time delay between anomaly identification and intervention initiation.

[0017] This invention addresses the potential perception blind spots in the associated areas of abnormal information within the anomaly response process by introducing a spatiotemporal compensation data set construction mechanism. Based on historical image fragments, trajectory records, and regional behavioral evolution data, this mechanism reconstructs the state of physically limited areas, enabling effective state estimation and event verification even in sensor-inaccessible regions. This significantly improves the system's response accuracy in environments with incomplete visibility or occlusion. Finally, the invention verifies the compensated state, outputs anomaly intervention commands based on response judgment criteria, and links to the intelligent traffic dispatch system to ensure rapid conversion of identification results into on-site control actions. This closed-loop mechanism enhances the overall timeliness and reliability of the intelligent traffic anomaly identification and response chain, making it particularly suitable for scenarios requiring immediate judgment and intervention, such as initial traffic accidents and the spread of localized faults. Attached Figure Description

[0018] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings; Figure 1 This is a schematic diagram of a distributed sensor abnormal event identification method for intelligent transportation according to the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0020] Reference Figure 1 The following examples were obtained: Example 1: A distributed sensor anomaly event identification method for intelligent transportation includes the following steps: Deploying distributed sensor nodes with dynamic confidence value adjustment function in the traffic monitoring area; the sensor nodes generate initial confidence values ​​for corresponding anomaly information based on the continuous characteristics of historical monitoring data, the magnitude of changes in environmental background, and the degree of deviation of current sampled data; constructing a distributed sensing infrastructure oriented towards the dynamic traffic environment; by bringing data acquisition, state assessment, and confidence modeling functions to the local sensor nodes, improving the system's front-end response capability to subtle anomalies, especially enabling proactive signal reporting in the early stages of single-point events, ensuring the integrity and independence of subsequent identification.

[0021] The initial confidence values ​​are transmitted to the central processing unit. The central processing unit performs normalization processing on each initial confidence value based on the spatial distribution relationship of sensor nodes, the intersection ratio of sensing areas, and the collaborative correlation of historical identification. During the data fusion process, the independent identification priority of abnormal information from a single sensor node is retained. The initial confidence values ​​of distributed acquisition are normalized through a multi-dimensional fusion mechanism to eliminate the perception bias caused by node layout and coverage overlap. At the same time, by retaining the independent identification priority of abnormal information, it ensures that a single node has the ability to independently warn of sudden anomalies and avoids ignoring early high-risk events due to the majority voting mechanism.

[0022] Normalized anomaly information is reconstructed sequentially under a unified time reference. The central processing unit reassembles the anomaly information uploaded at different times according to the set time window, ensuring that the anomaly information detected by a single sensor node in the early stage of the event can fully participate in the subsequent identification process. The alignment of multi-node asynchronous sensing data in the time domain and the reconstruction of the event process ensure that the anomaly information in each time slice can participate in the analysis in chronological order, while protecting the single-point high-confidence information in the early stage of the event from being covered, which helps to accurately restore the event evolution path and triggering causality.

[0023] After sequence reconstruction is completed, the central processing unit performs a local state analysis process based on the spatial distribution characteristics of the current anomaly information, the normalized confidence value, and the historical anomaly probability of the corresponding area. It combines sensor data changes in the vicinity of the anomaly information, traffic flow state shifts, and the evolution trend of the target trajectory to determine whether event response conditions are met. When the judgment result meets the set conditions, the anomaly response process immediately begins, without relying on data consistency judgments from multiple sensor nodes. By constructing a comprehensive judgment model that links spatial, temporal, and behavioral factors, dynamic assessment of local traffic conditions is achieved, with a focus on identifying whether there are any unmanifested risk states within the area. By eliminating the dependence on multi-node consistency, isolated high-confidence anomalies can be quickly triggered to respond under specific conditions, effectively improving the system's early warning capability.

[0024] After the anomaly response process is initiated, the central processing unit constructs a spatiotemporal compensation dataset for the perception blind zone area related to the current anomaly, based on historical image fragments, trajectory records, and regional behavior evolution data. This dataset reconstructs the state of the physically limited area and performs supplementary verification based on the reconstruction results. When the verification results meet the response judgment criteria, an anomaly intervention command is output and linked to the intelligent traffic dispatch system. To address perception blind zone issues caused by physical occlusion and viewpoint dead zones, a spatiotemporal compensation dataset is constructed by referencing historical multi-source information to reconstruct the dynamic state of currently unobservable areas, providing reinforced verification for response decisions. When the reconstructed behavioral trend matches the current anomaly, the linkage mechanism is triggered to ensure that the response behavior has interpretability and precise controllability.

[0025] In this invention, the central processing unit refers to a processing entity with centralized computing, data fusion, and logical decision-making capabilities, typically composed of a central server, edge processing devices, or cloud processing nodes within a traffic management platform. In existing intelligent transportation technologies, this processing unit primarily undertakes the aggregation and analysis of multi-source heterogeneous data. It processes anomaly information, trajectory data, and real-time monitoring data reported from multiple sensor nodes, and performs unified normalization, time-series reconstruction, state analysis, and response determination processes. The central processing unit possesses logical control over the event identification process; its decision-making results directly influence the initiation and execution of subsequent response actions, making it the core information processing link in this invention for achieving a distributed anomaly identification closed loop.

[0026] An intelligent traffic dispatch system refers to a comprehensive dispatch and control platform built upon urban traffic infrastructure. It typically includes components such as a traffic signal control platform, a guidance display system, road control equipment, and an emergency resource coordination mechanism. In existing technologies, this system receives instructions from a central processing unit and executes specific traffic control tasks, such as adjusting traffic light cycles, issuing traffic warnings, dispatching police resources, or activating on-site monitoring. In the implementation of this invention, the intelligent traffic dispatch system acts as the execution terminal for responding to instructions. It is responsible for transforming the abnormal event results identified by the central processing unit into on-site intervention actions, thereby completing a closed loop from identification to intervention. It is an indispensable execution entity for achieving the objectives of this invention.

[0027] The initial confidence value for generating corresponding anomaly information includes the following steps: Distributed sensor nodes deployed within the traffic monitoring area continuously collect multiple physical parameters of the target road segment or traffic intersection area. The sampling objects may include, but are not limited to, traffic environmental variables such as vehicle speed, vehicle distance, vibration, noise, temperature, and image frame density. Taking vehicle speed as an example, each sensor node collects vehicle speed data once per second within a set time window (e.g., a continuous 10 seconds), forming a monitoring data sequence consisting of 10 time points. The sensor nodes perform trend fitting (such as linear regression or moving average) on the data within this time window in their local processing module to obtain the speed variation trend, and simultaneously calculate the standard deviation or coefficient of variation as a statistical stability index of the data sequence. For noise data, such as traffic sound level or vibration amplitude, the same window is used to perform periodic stability analysis. This process constitutes the implementation steps of "analyzing the continuous characteristics of historical monitoring data and extracting the variation trends and statistical stability of various preset data."

[0028] The sensor node, based on its installation location and physical environment, retrieves pre-stored environmental background baseline data to assess the difference between the current sampling state and the long-term historical average. For example, if the long-term baseline for vibration amplitude is 0.08 mm / s, and the current measured average is 0.19 mm / s, it is considered a significant deviation. To quantify the "amplitude of change in environmental background," the system calculates the offset ratio between the current value and the historical average (e.g., 2.375). In actual deployments, such as nodes installed at the entrance of underground passages, background vibration shifts may occur due to engineering work. This difference will be promptly identified and quantified through this step, and input into the subsequent anomaly judgment process.

[0029] The sensor node combines the trend and stability indicators extracted from the "continuous characteristics of historical monitoring data" with the quantified differences from the "variation magnitude of environmental background," and further integrates these with the latest observations collected within the current time window to calculate the "deviation degree of the current sampled data." This deviation degree is obtained by comparing it with the standard value of the historical stable period. For example, if the noise level when a vehicle passes is 85 dB, while the long-term average noise level at that location under similar traffic conditions is 72 dB, then the deviation degree is 13 dB.

[0030] The sensor node inputs the continuous characteristics of historical monitoring data, the magnitude of changes in the environmental background, and the offset of the current sampled data into its internal confidence value calculation structure to generate an initial confidence value for the corresponding anomaly information. This confidence value will be subsequently uploaded to the central processing unit for data fusion, sequence reconstruction, and response process judgment, and has the ability to retain independent identification priority to ensure that a single node event can still trigger the initial response mechanism when most data is not synchronized.

[0031] The multi-factor confidence calculation model is constructed through the following steps: After the sensor nodes extract the continuous features of historical monitoring data, the magnitude of changes in environmental background, and the degree of deviation of current sampling data, multiple monitoring factors in these three dimensions need to be compared item by item with the currently preset anomaly criteria. In specific implementation, the continuous features of historical monitoring data may include multiple indicators such as vehicle speed fluctuation amplitude, acceleration change rate, and number of frequent starts and stops; the magnitude of changes in environmental background may involve dimensions such as sound pressure level difference, vibration frequency change, and light intensity shift; while the degree of deviation of current sampling data refers to the deviation from the historical average level, such as drastic fluctuations in temperature, humidity, and smoke concentration. The system compares these indicators with the standard anomaly thresholds recorded in the anomaly identification strategy library, such as vehicle speed fluctuation amplitude exceeding five meters per second, instantaneous increase in sound pressure level of fifteen decibels, and instantaneous increase in smoke concentration exceeding fifty units of concentration value. Input factors that do not meet the screening conditions are directly eliminated, and only indicators that meet the relevance to the current anomaly situation are retained as valid input factors for subsequent modeling. This screening mechanism ensures that the model focuses on the key elements under the current semantics, improving the pertinence of the judgment and the computational efficiency.

[0032] For all input factors retained from the previous screening step, confidence levels are constructed based on their anomaly detection performance over a long period of observation. Supported by a large amount of historical identification records, the false alarm rate and false negative rate of each factor in typical anomaly detection tasks can be statistically calculated. For example, in identifying tire blowout events, the false alarm rate for tire vibration frequency change was found to be 3%, and the false negative rate was 2%, while the false alarm rate for noise peak values ​​in the same event was 15%, and the false negative rate was 40%. Based on these statistics, the effective identification capability of each factor is assigned a value, and a linear mapping operation is performed to normalize each original monitoring value within its historical performance range. For example, the tire vibration frequency is normalized from the original 0.23 dB / s to a confidence contribution value of 0.96, and the sound pressure level peak value is normalized from the original 87 dB / s to a confidence contribution value of 0.51. Through this processing method, the confidence dimension of each effective factor is standardized while maintaining its relative strength in actual identification contributions.

[0033] Using all confidence contribution values ​​that have undergone linear mapping as the basis for calculation, a weighted sum is performed to generate a preliminary confidence output for the current anomaly assessment. In the weighted calculation, the weight coefficient corresponding to each confidence contribution value is derived from the inverse weight combination of the false alarm rate and the false negative rate obtained from long-term observations in the second step; that is, the lower the false alarm rate and the lower the false negative rate, the higher the weight. For example, in the aforementioned tire blowout scenario, the vibration frequency index, due to its low false alarm and false negative rates, might have a weight of 0.35 in the confidence weighting; while the peak sound pressure level, due to its high false alarm rate, might have a weight of 0.18. The overall confidence level is obtained by summing the products of all confidence contribution values ​​multiplied by their respective weights. For example, in the calculation results at a certain time point, if the confidence contribution value sequence is 0.96, 0.72, and 0.51, with corresponding weights of 0.35, 0.32, and 0.18, then the weighted summed confidence value is 0.762, representing the preliminary confidence value output for the current anomaly information.

[0034] The initial confidence value is considered the initial confidence value of the anomaly information and serves as the core judgment criterion in subsequent upload processing. In the subsequent normalization processing stage, time series reconstruction process, event response condition judgment, and anomaly response process, this initial confidence value will be used throughout the entire multi-stage judgment chain, determining whether it has the independent identification priority, whether it meets the response criteria, and whether it constitutes the basis for event judgment. Through the construction process of this multi-factor confidence calculation model, not only is the modeling accuracy of complex anomaly events improved, but the anomaly recognition capability of single sensor nodes in low-perception collaborative scenarios is also enhanced, meeting the core requirements of rapid early warning and high sensitivity in intelligent transportation environments.

[0035] The weighting coefficients of the confidence contribution value are disclosed as follows: During long-term traffic monitoring, the recognition performance of each input factor is recorded when participating in anomaly identification tasks. This mainly includes the number of incorrect and missed recognitions when the factor independently participates in identification in historical samples. In a labeled set of anomaly event samples, the frequency of correct, incorrect, and missed recognitions of each input factor in multiple events is statistically analyzed, thereby yielding its misjudgment and missed recognition ratios.

[0036] The false positive and false negative ratios are converted into inverse confidence scores using a normalization method. Specifically, the false positive ratio of the current factor is inversely proportional to the maximum false positive ratio among factors in that category, and the false negative ratio is also inversely proportional to the maximum false negative ratio among factors in that category. These two inverse scores represent the reliability of the factor's identification in the false positive and false negative dimensions, respectively. These two scores are then averaged or multiplied to generate the final identification stability weight. A higher identification stability weight indicates a more reliable factor, and it is subsequently assigned a higher weight to its confidence contribution value. In this way, each retained input factor has a confidence weighting coefficient derived from long-term sample statistics, based on false positive and false negative performance. This coefficient participates in subsequent weighted summation calculations to form the initial confidence value. This method has a clear data support source, is repeatable and verifiable, and can effectively improve the sensitivity of the confidence model in identifying low-frequency anomalies or single-point sudden events.

[0037] Normalizing the initial confidence values ​​involves the central processing unit spatially locating each sensor node based on its fixed deployment coordinates within the traffic monitoring area, and constructing a regularized two-dimensional spatial grid structure across the entire monitoring area. By calculating the Euclidean or Manhattan distance between nodes, the spatial interval between any two nodes in the grid is determined, thus obtaining the spatial structural correlation between the nodes. In practical implementation, for example, in a densely populated urban main road area, the spatial distance may be less than fifty meters, indicating a strong correlation, while in suburban or discontinuous monitoring areas, the distance may exceed two hundred meters, indicating a low spatial correlation between nodes. Through this mapping of spatial distribution relationships, a set of reference factors related to its spatial location can be assigned to each initial confidence value for subsequent weight adjustments.

[0038] The central processing unit further analyzes the sensing area coverage of each sensor node. By modeling sensing parameters such as viewing angle, detection radius, and occlusion area, it calculates the intersection ratio of sensing areas between adjacent nodes. For example, if two nodes each cover a radius of 50 meters, and there is a 20-meter intersection area between them, the intersection ratio is 20%. For node pairs with a high intersection ratio, their data has high redundancy and complementarity, making them suitable for multi-source information fusion. Edge nodes with a low intersection ratio, however, have more independent data and require appropriate adjustments during normalization to prevent excessive dilution in the fusion calculation. This processing ensures that the influence of different nodes in the fusion judgment is more differentiated, effectively preventing judgment bias caused by insufficient area overlap. During calculation, the maximum intersection ratio between a sensor node and its adjacent nodes is taken as the sensing area intersection ratio.

[0039] The central processing unit retrieves historical identification data records and performs statistical analysis on the frequency of each sensor node's appearance in similar anomaly events and the consistency of its judgment results to construct a quantitative index of collaborative correlation in historical identification. Specifically, in multiple archived anomaly event data sets, it statistically analyzes whether a particular node participated in the identification of the same type of event and whether its judgment results were consistent with the final identification conclusion, calculating its participation frequency and judgment consistency rate. For example, in the past twenty fire-related anomaly events, if a node participated in identification seventeen times, and its judgment results were consistent with the final event identification fifteen times, its consistency rate is 88%, and its participation frequency is 85%. The central processing unit combines the node's participation frequency and consistency rate to generate a collaborative correction value, which serves as an important basis for correcting its initial confidence value. The higher the collaborative correction value, the more reliable the node is in historical identification, and the higher its retention strength should be obtained in the subsequent normalization process.

[0040] The central processing unit uses the spatial correlation degree of each node, the intersection ratio of the perceived regions, and the historical identification collaborative correction values ​​as reference factors for confidence value normalization. These factors are input into the normalization calculation process to adjust the weight and expressive power of the initial confidence values ​​of each node, thereby generating a set of comparable confidence values ​​under a unified dimension. This provides a standardized data foundation for subsequent time series reconstruction and event judgment. This processing method not only preserves the spatial differences and source independence of anomalous information but also enhances the guiding value of historical data for current judgments, improving the stability and scientific rigor of the fusion judgment.

[0041] Incorporating spatial correlation, the overlap ratio of sensing areas, and collaborative correction values ​​into the normalization calculation process refers to the central processing unit analyzing the historical behavior of each sensor node and constructing an evaluation index for recognition capabilities related to event types. During actual deployment, each type of anomalous event is labeled and archived by a professional team, recording the time, location, and final anomaly type of the event. In the analysis process, the central processing unit first traverses all archived samples of the same type of anomalous event, counting whether each node participated in anomaly identification and recording whether its judgment result is consistent with the final result. For example, if thirty traffic conflict events occurred in a certain monitoring area, and a certain node participated in identification twenty-five times, with twenty judgment results consistent with the final judgment, then the node's historical response consistency rate is 80%. The response consistency rate reflects the reliability of the node's judgment in historical events and is a fundamental indicator for calculating the correction factor.

[0042] After obtaining the response consistency rate of a node, the central processing unit further combines this with the node's frequency of occurrence in similar abnormal events, i.e., its participation frequency, to generate a collaborative correction value for that node. The collaborative correction value is generated by multiplying the response consistency rate by the participation frequency to obtain a non-linear correction factor. Continuing with the above example, if its participation frequency is 83%, its collaborative correction value is 66.4%. This value reflects the historical stability of the node's performance in group collaborative judgment. If a node frequently participates in judgment but has a low consistency rate, its collaborative correction value will be lowered, reducing the impact of its subsequent confidence level on the final result. This step, by introducing a historical confidence evaluation mechanism, establishes a dynamic adjustment relationship between node behavioral characteristics and confidence output.

[0043] After obtaining the collaborative correction values ​​for each node, the central processing unit combines them with the calculated spatial correlation and the intersection ratio of the sensing regions to form a confidence correction factor. Specifically, spatial correlation reflects the relative density of node distribution in a two-dimensional coordinate system; for example, two nodes 20 meters apart have a much higher correlation than nodes 150 meters apart. The intersection ratio of sensing regions indicates the degree of sensing redundancy between nodes; for example, if the intersection ratio of the coverage areas of two nodes is higher than 30%, it indicates a high degree of overlap in their sensing targets. The central processing unit constructs the confidence correction factor for each node by multiplying the collaborative correction value, spatial correlation value, and intersection ratio value, using this as a weighted multiplication term for its initial confidence value. This process ensures that, under the same initial confidence value, nodes with higher spatial confidence, historical consistency, and redundant collaboration have higher fusion weights, thereby improving the recognition efficiency of the fusion result.

[0044] The central processing unit inputs all corrected initial confidence values ​​into a unified normalization process. Using a maximum-minimum interval linear transformation, it compresses the confidence values ​​of each node to a preset standard range, such as between zero and one, forming a standardized set of confidence values ​​with uniform dimensions. This process ensures that sensor nodes from different sources, locations, and with different collaborative performances have relatively fair and differentiated evaluation weights for their reported anomalies before entering time series reconstruction and subsequent judgment. This significantly enhances the method's sensitivity to isolated, sudden events and reduces judgment distortion caused by majority voting mechanisms.

[0045] When performing confidence value fusion judgment, the central processing unit retains an independent channel for anomalies generated by a single sensor node whose normalized initial confidence value exceeds a set initial threshold. This independent channel is not limited by the presence of identical anomalies at other nodes. This ensures that the system can still respond effectively to isolated anomalies even when multiple nodes are missing responses. After normalizing the initial confidence values ​​of each node, the central processing unit immediately performs a confidence filtering operation. The filtering criterion is: if an anomaly is reported independently by a single sensor node and its normalized confidence value exceeds the system's preset initial response threshold, it is considered a "high-confidence single-point anomaly" and must be specially marked and transferred to the independent identification process. For example, in an intersection area of ​​a city, if only one thermal imaging sensor node detects an anomaly signal where the temperature rises by more than ten degrees Celsius instantaneously and generates a confidence value of 0.8, exceeding the preset threshold of 0.75, this anomaly is still preferentially retained even if no other nodes report similar anomalies.

[0046] The central processing unit establishes an independent judgment channel for marked "high-confidence single-point anomalies." This channel does not participate in consistency verification with data from other nodes in the subsequent fusion judgment process, nor does it rely on majority voting decision-making mechanisms. Within this channel, anomaly information is processed independently as a single event and has the authority to trigger judgment conditions independently. This mechanism can effectively prevent important early warning signals from being mistakenly identified as noise and discarded due to the dilution of individual anomaly signals by the majority of normal data during the data fusion process.

[0047] In the independent judgment channel, independent anomaly information does not directly trigger a response, but it will participate in the fusion process during subsequent local state analysis and target trajectory evolution judgment, ensuring that it is not discarded prematurely. The central processing unit establishes a follow-up intervention mechanism for this independently identified priority anomaly information. If, within its subsequent time window, other neighboring nodes begin to successively produce highly correlated anomaly information, the independent channel information is automatically transferred to the main fusion judgment process and merged to participate in the comprehensive response decision. For example, if within 30 seconds of the first node reporting an anomaly signal, two surrounding nodes detect an increase in smoke particle concentration and a sudden change in traffic flow, the system will reconstruct the event timeline and locate the earliest single-point anomaly information as the dominant source of the event, serving as an important basis for overall event identification. This approach ensures that weak initial signals have the ability to evolve into complete events, preserving the leading data evidence of potentially high-risk events to the greatest extent possible.

[0048] Sequence reconstruction includes the following steps: The central processing unit performs timestamp alignment processing on all normalized anomaly information based on the original upload timestamp accompanying each anomaly report. On a per-second basis, the timestamps of all anomaly information are aligned down to the nearest whole second, serving as a reference point on a unified timeline. For example, if an anomaly was uploaded at 10:00:20:70 milliseconds, the system aligns it to 10:00:20. This processing aims to eliminate interference from micro-temporal differences in cross-node event identification, ensuring that all anomaly data is compared under a unified event perception starting benchmark.

[0049] After time alignment is completed, the central processing unit aggregates abnormal information within the same time window into an initial time period set according to the set time window parameters, such as a range of five seconds, ten seconds, or longer. Each time window is regarded as a segment of event state, and all abnormal behaviors occurring within it are temporarily treated as concurrent events. This step lays the foundation for subsequent reconstruction of the event evolution sequence of cross-node information, which is especially important for traffic emergencies (such as collisions, fires, and hit-and-runs) that develop rapidly in units of several seconds.

[0050] The central processing unit performs global temporal sorting on the abnormal information according to the chronological order within the aforementioned time period set, forming a cross-node time-series information sequence covering the entire time window. During the construction of this time-series information sequence, not only are the timestamps and node source identifiers of each abnormal message retained, but normalized confidence values ​​are also used as a sorting auxiliary factor. This is used to prioritize the salience of data with high confidence values ​​when multiple abnormal messages are close in time but have different confidence levels. For example, if six abnormal messages are received within a five-second time period, and three of them come from different nodes with confidence values ​​all higher than 0.85, then the backbone time paths of these three data points are constructed first.

[0051] Based on the constructed cross-node time-series information sequence, the central processing unit performs an event development path reconstruction process to ensure that abnormal information detected by a single sensor node in the early stages of the event can be fully preserved and used in subsequent judgment and analysis. This path reconstruction not only shows the spatial propagation trajectory of the event but also reflects its temporal progression. For example, in a traffic anomaly identification process, the highest temperature signal reported by the thermal sensor in the non-motorized vehicle lane area is included as the event's starting point. Subsequently, smoke monitoring nodes and traffic flow detection nodes at the lane boundaries successively report abnormal states. The reconstructed time series shows that the anomaly spread from the appearance of the heat source to the accumulation of dense smoke, and then to the evolution of traffic congestion, fully reconstructing the dynamic process of the event and helping to improve the timeliness and accuracy of the response.

[0052] The process of determining whether event response conditions are met includes the following steps: The central processing unit, based on the completed cross-node time-series information, sorts and labels all normalized anomaly information, extracting the sensor node that uploaded the anomaly information earliest and designating it as the dominant source node. This dominant source node is considered the monitoring unit that first sensed the anomaly event, and its location typically has strong spatiotemporal leading characteristics, serving as a key reference for judging subsequent evolution trends. For example, if at 10:23:32, the thermal sensor numbered seven first uploads a local high-temperature anomaly signal, and other nodes report air pressure and visual anomalies only in the following seconds, then sensor node numbered seven is designated as the dominant source node, and its identification result will have priority in subsequent judgment processes.

[0053] The normalized confidence value obtained by the dominant source node during the normalization process is retrieved, and the frequency of anomalous events recorded in the long-term historical data of the area where the node is located is extracted and compared with the preset spatial confidence threshold and spatial risk threshold. If the normalized confidence value is higher than the spatial confidence threshold, and the historical frequency of anomalous events in the area is higher than the spatial risk threshold, the dominant source node can be identified as having potential risks. This step effectively filters out low-intensity anomalous inputs with a high probability of false alarms by introducing a dual threshold mechanism of historical risk baseline and current identification confidence. Taking an anomaly in a bus stop area as an example, if the daily frequency of anomalous events in this area is relatively high, and the identification confidence value of the dominant source node reaches 0.95 and five similar events have occurred in the past month, then a high risk can be identified.

[0054] Based on the location information continuously reported by the dominant source node and its neighboring nodes, the movement trajectory of a preset target object is extracted and analyzed within a set time window. This includes calculating the accumulated movement distance and the accumulated angle of change in direction. Using the target object's sampled data per second within the current time window, the total path length in spatial coordinates is continuously calculated, and the deflection angle generated during each displacement is accumulated to construct the target object's trajectory behavior characteristics within that time period. If the accumulated displacement length exceeds a displacement threshold of three meters, and the total deflection angle exceeds a preset deflection threshold of thirty degrees, the target object is determined to have abnormal trajectory behavior, indicating that it is undergoing highly variable movement deviating from conventional traffic patterns. This judgment is particularly suitable for identifying sudden movement behaviors such as abnormal crossings, rapid reverse movement, and sudden changes in direction, improving the spatial perception capability of abnormal target states.

[0055] When the dominant source node is identified as having a potential hazard, and the trajectory behavior of the associated target object meets the abnormal characteristic judgment conditions, the central processing unit confirms this abnormal information as grounds for triggering a response process. This dual judgment mechanism breaks the consistency dependence of multi-sensor voting-based judgments, enabling the system to make independent and accurate response decisions based on temporal precedence, spatial risk intensity, and trajectory deviation trends, even when only some nodes perceive anomalies. It is particularly suitable for real-world urban road network environments with frequent perception blind spots, uneven data latency, or significant sudden behaviors of traffic targets.

[0056] It should be noted that, in this invention, to improve the response capability to micro-level sudden anomalies, an independent channel is proposed to retain anomaly information from a single sensor node in the judgment process, thus avoiding premature rejection of isolated high-risk signals under the traditional majority-node consensus strategy. However, to prevent erroneous triggering due to false alarms from individual sensors, noise interference, or extreme sampling bias, this invention does not directly use such independent anomaly information as the triggering basis for event response. Instead, it introduces subsequent trajectory behavior analysis and local state judgment mechanisms as a mandatory verification process. This design fully considers the motion evolution characteristics of dynamic targets in intelligent transportation environments and the collaborative correlation with surrounding perception data. In the absence of multi-node response support, by tracking the continuous changes in position information of the target object in the dominant source node and its neighboring nodes, and combining the accumulated movement distance and the accumulated angle of direction change, the possible evolution path of the anomaly event is reconstructed, thereby achieving reconfirmation of the anomaly's authenticity.

[0057] By incorporating factors such as spatial state evolution, trajectory deviation trends, and historical risk probabilities into the identification process, even when other nodes fail to provide synchronous feedback due to occlusion, sampling misalignment, or perception blind spots, the dynamic consistency of time and space dimensions can strengthen confidence in judging isolated anomalies and significantly reduce the risk of false alarms. Simultaneously, it ensures the robustness and response accuracy of this invention in complex sensing environments. Trajectory behavior analysis is not only a supplement to traditional fusion mechanisms but also a core step in achieving reliable verification of isolated high-confidence anomalies, ensuring that the constructed independent judgment channel possesses sufficient response credibility and execution basis. This mechanism fills the technical gap in existing technologies regarding the lack of response paths for single-point high-risk sensing events, demonstrating significant creativity and practical value.

[0058] Instead of identifying anomalies in independent channels, the identification process relies on the collaborative reporting of multiple sensor nodes to the same spatial location, similar time period, and similar anomaly type. When multiple sensor nodes detect anomalies with consistent characteristics within their respective sensing ranges, and their normalized confidence values ​​all exceed a set threshold, the central processing unit considers them as highly consistent event sources. A consistency fusion mechanism is then used to quickly determine the response to the anomaly. This mechanism fully leverages the complementary advantages of multiple nodes in spatial structure and temporal dimensions, enabling rapid confirmation of highly significant anomalies and improving the system's identification efficiency and decision-making accuracy in routine traffic safety incidents.

[0059] In existing technologies, consistency fusion mechanisms typically refer to a processing method used in multi-sensor network environments. This method involves verifying the consistency of information obtained from data observations of the same target or event by different sensors through spatiotemporal alignment, feature matching, and result comparison. This consistency is then used as the basis for fusion judgment. This mechanism is widely used in intelligent sensing systems such as traffic monitoring, video perception, and environmental early warning. Its core lies in improving the accuracy and robustness of the overall judgment by determining whether multiple data sources achieve similar or even identical identification results for the same abnormal event. For example, when three or more independent sensor nodes detect dense smoke, abnormal temperature, or traffic flow interruption signals at the same location within the same time window, and their respective confidence values ​​are all higher than a threshold, the system confirms the anomaly has high credibility through the consistency fusion mechanism, thereby triggering subsequent response operations. This type of anomaly information has high node collaboration support, broad confidence sources, and a low false alarm probability. Therefore, the system can directly trigger the response mechanism after determining that the response conditions are met, without needing to perform additional supplementary verification or spatiotemporal reconstruction processes. The supplementary verification mechanism is designed primarily to address the issue of insufficient credibility of abnormal information in the independent judgment channel. It is used to enhance the reliability of single-point anomaly response by means of historical feature mapping, perception blind zone state reconstruction, and other methods when other nodes fail to provide supporting data.

[0060] After the anomaly response process is initiated, the supplementary verification includes the following steps: The central processing unit determines the perception blind zone area corresponding to the current anomaly information. This perception blind zone area refers to the spatial area where other sensor nodes have failed to provide coverage or generate usable monitoring data within the current time period. Based on this spatial range, the central processing unit extracts indicators such as trajectory distribution density, frequency of behavioral state changes, and probability of anomaly occurrence within the historical time period corresponding to the time of the anomaly event. Trajectory distribution density is calculated by statistically analyzing the number of traffic target trajectories appearing in the area within the historical time period and the coverage area; the frequency of behavioral state changes is based on the number of times different traffic states (such as stationary, turning, gathering, sudden disappearance, etc.) change within the area in the historical records within similar time periods; the probability of anomaly occurrence is calculated based on the frequency of occurrence of events marked as true anomalies in the historical data of the area and the total sample size. The above three indicators together construct a structured historical feature set, which, together with historical image fragments and behavioral evolution data of the area, constitutes the spatiotemporal compensation data set of the area.

[0061] The central processing unit extracts real-time data from nearby sensors within a continuous timeframe before and after the anomaly, based on the temporal and locational ranges of the current anomaly information. This data includes parameters such as the movement trajectory of traffic targets, speed changes, image visual residuals, and signal strength fluctuations. By analyzing the continuous trends in these parameters, a spatiotemporal evolution segment of the current state is formed. This segment not only reflects the traffic conditions before and after the anomaly but also demonstrates the behavioral patterns of the target objects, possessing the timeliness and relevance to complete the current traffic situation.

[0062] The central processing unit converts the constructed spatiotemporal evolution segment of the current state into a multi-dimensional feature vector. This feature vector includes several publicly available metrics such as trajectory distribution density, frequency of behavioral state changes, and historical anomaly probability. Subsequently, the central processing unit matches the current feature vector with the feature vectors corresponding to each group of structured historical features in the structured historical feature set. The matching process uses cosine similarity for similarity scoring, evaluating the similarity of behavioral patterns between the current state vector and historical feature vectors by calculating the degree of directional consistency. The closer the similarity score is to one, the higher the degree of matching. Alternatively, the central processing unit can use Euclidean distance to measure the numerical difference between the current state and historical states, also denoted as a similarity score. When the Euclidean distance between vectors is below a set distance threshold, the states are considered highly similar. This similarity measurement process ensures quantitative matching of the current anomalous state under multiple dimensions, thus providing a more reliable basis for response judgment.

[0063] When the similarity score meets the preset similarity conditions, the central processing unit outputs an abnormal intervention command and links it to the intelligent traffic dispatch system to execute preset traffic control and guidance strategies. These strategies include, but are not limited to, route detour instructions, temporary traffic light control, and danger zone marking reminders, ensuring that traffic participants can avoid high-risk areas in a timely manner.

[0064] The above algorithms or formulas are all dimensionless and numerical calculations, and the results are obtained by software simulation based on a large amount of collected data to obtain the most recent real-world results. The preset parameters are set by those skilled in the art according to the actual situation.

[0065] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply 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 this application.

[0066] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0067] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0068] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for identifying abnormal events using distributed sensors in intelligent transportation, characterized in that, Includes the following steps: Distributed sensor nodes with dynamic confidence value adjustment function are deployed in the traffic monitoring area. The sensor nodes generate the initial confidence value of the corresponding abnormal information based on the continuous characteristics of historical monitoring data, the magnitude of changes in environmental background and the degree of deviation of the current sampling data. The initial confidence values ​​are transmitted to the central processing unit. The central processing unit performs normalization processing on each initial confidence value based on the spatial distribution relationship of sensor nodes, the intersection ratio of sensing areas, and the collaborative correlation of historical identifications. During the data fusion process, the independent identification priority of abnormal information from a single sensor node is retained. The normalized abnormal information is reconstructed in sequence under a unified time reference. The central processing unit reassembles the abnormal information uploaded at different times in sequence according to the set time window to ensure that the abnormal information detected by a single sensor node in the early stage of the event can fully participate in the subsequent identification process. After the sequence reconstruction is completed, the central processing unit performs a local state analysis process based on the spatial distribution characteristics of the current abnormal information, the normalized confidence value, and the historical abnormal probability of the corresponding area. It combines the changes in sensor data in the vicinity of the abnormal information, traffic flow state deviation, and the evolution trend of the target trajectory to determine whether the event response conditions are met. When the judgment result meets the set conditions, the abnormal response process is immediately initiated without relying on the data consistency judgment of multiple sensor nodes. After the anomaly response process is initiated, the central processing unit constructs a spatiotemporal compensation data set for the area within the perception blind zone related to the current anomaly information, based on historical image fragments, trajectory records, and regional behavior evolution data. It then reconstructs the state of the physically limited area and completes supplementary verification based on the reconstruction results. When the verification results meet the response judgment criteria, it outputs an anomaly intervention command and links it to the intelligent traffic dispatch system.

2. The method for identifying distributed sensor anomaly events in intelligent transportation according to claim 1, characterized in that, Generating the initial confidence value for the corresponding anomaly information includes the following steps: The sensor nodes continuously collect monitoring data based on a set time window, and analyze the continuous characteristics of historical monitoring data to extract the changing trends and statistical stability of various preset data. The sensor node compares and analyzes the environmental background information at the current sampling location, calculates the difference between the current monitoring state and the long-term stable state, and quantifies the magnitude of change in the environmental background. The sensor node constructs a multi-factor confidence calculation model based on the continuous characteristics of historical monitoring data, the magnitude of changes in the environmental background, and the degree of deviation of the current sampled data, and generates the initial confidence value of the current anomaly information.

3. The method for identifying distributed sensor anomaly events in intelligent transportation according to claim 2, characterized in that, The multi-factor confidence calculation model is constructed through the following steps: Based on the continuous characteristics of historical monitoring data, the magnitude of changes in environmental background, and the degree of deviation of current sampling data, the data is compared with the preset anomaly criteria item by item to eliminate input factors that are irrelevant to the current anomaly or do not meet the screening conditions. For the retained input factors, a linear mapping process is performed based on the actual impact of each factor on the false alarm rate and false negative rate in anomaly identification during long-term observation. The original values ​​are transformed into confidence contribution values ​​under a unified dimension, while maintaining the relative weight ratio between each factor. The mapped confidence contribution values ​​are weighted and summed to generate the initial confidence value of the current anomaly information.

4. The method for identifying distributed sensor anomaly events in intelligent transportation according to claim 3, characterized in that, Normalizing the initial confidence values ​​refers to: The central processing unit performs coordinate mapping on the spatial distribution relationship of each sensor node, constructs a two-dimensional spatial grid, and calculates the relative distance between each node to determine the degree of correlation between nodes in the spatial structure. The central processing unit combines the sensing area covered by the sensor nodes to calculate the intersection ratio of the sensing areas between adjacent nodes. The central processing unit retrieves historical identification data and assigns corresponding confidence values ​​to different nodes based on the frequency and consistency of their collaborative responses in similar events. It then incorporates spatial correlation, the intersection ratio of perceived areas, and the collaborative correction values ​​into the normalization calculation process to generate a set of comparable confidence values ​​under a unified scale.

5. A method for identifying distributed sensor anomaly events in intelligent transportation according to claim 4, characterized in that, Incorporating spatial correlation, the intersection ratio of perceived regions, and collaborative correction values ​​into the normalization calculation process refers to: The central processing unit performs historical behavior analysis on each sensor node. First, it counts the number of times the node actually participated in the identification of the same type of anomaly events that have been labeled, and the number of times the node's results are consistent with the final identification results in these events, and calculates its historical response consistency rate. The collaborative correction value for each node is generated by multiplying its historical response consistency rate by its participation frequency. The central processing unit multiplies the collaborative correction value by the spatial correlation value and the intersection ratio of the sensing area corresponding to the node, and uses it as a weighted correction factor for the initial confidence value of the node. The central processing unit performs maximum-minimum interval linear normalization on all corrected confidence values, so that the confidence values ​​of all nodes fall within a preset uniform numerical range, generating a standardized set of confidence values.

6. The method for identifying distributed sensor anomaly events in intelligent transportation according to claim 5, characterized in that, When performing confidence value fusion judgment, the central processing unit retains an independent channel for abnormal information that is generated by a single sensor node and whose initial confidence value after normalization exceeds the set initial threshold. This channel is not restricted by whether other nodes have the same abnormal information.

7. The method for identifying distributed sensor anomaly events in intelligent transportation according to claim 6, characterized in that, Sequence reconstruction includes the following steps: Based on the upload time of each abnormal information, the central processing unit performs timestamp alignment on the normalized abnormal information to construct a unified timeline for unifying the starting benchmark for event perception. The central processing unit arranges the abnormal information with time differences in chronological order according to the set time window, forming a cross-node time sequence information sequence, which is used to reconstruct the development process of abnormal events.

8. A method for identifying distributed sensor anomaly events in intelligent transportation according to claim 7, characterized in that, Determining whether the conditions for an event response are met includes the following steps: The central processing unit determines the occurrence time and node source of each abnormal information based on the cross-node time sequence of information, and extracts the sensor node that uploaded the abnormal information earliest and sets it as the dominant source node. The normalized confidence value of the dominant source node and the historical anomaly frequency of the region are compared with the spatial confidence threshold and the spatial risk threshold, respectively. If both exceed the threshold, a potential risk is identified. Based on the location information continuously reported by the dominant source node and its neighboring nodes, the cumulative amount of movement distance and the cumulative amount of direction change angle of the preset target object are counted within a set time window. If both exceed the displacement threshold and the deflection threshold respectively, it is determined that the target object has abnormal trajectory behavior. When there are potential risks in the primary source node and the target object's trajectory behavior is abnormal, the abnormal information is deemed to meet the conditions for triggering the response process.

9. A method for identifying distributed sensor anomaly events in intelligent transportation according to claim 8, characterized in that, After the exception response process is initiated, the supplementary verification includes the following steps: The central processing unit determines the perception blind zone area corresponding to the current abnormal information, and extracts the trajectory distribution density, behavioral state change frequency and abnormal event occurrence probability of the area in the historical time period corresponding to the time of the abnormal event, and constructs a structured historical feature set, i.e., a spatiotemporal compensation data set. The central processing unit extracts the real-time data change trends before and after the anomaly based on the time and location range of the current anomaly information, and quickly matches them with the structured historical feature groups in the structured historical feature set. When the similarity score between the current state and any set of structured historical feature groups meets the preset similarity conditions, the central processing unit outputs an abnormal intervention command and links it to the intelligent traffic dispatch system to execute the preset traffic control and guidance strategy.

Images