Intelligent network connection inductive control platform for road traffic safety facilities

Abstract

The invention relates to the field of traffic control, in particular to an intelligent network connection inductive control platform for road traffic safety facilities. The traffic data acquisition module is used for acquiring a data flow of a traffic sensor and outputting traffic data with a traffic data source identifier through an identification technology; the traffic data processing module is used for obtaining standardized traffic parameters according to a traffic data source identifier adaptive analysis protocol; evaluating a data reliability index according to the index system; the situation fusion module is used for forming a vehicle driving track through a graph neural network, generating a global traffic situation map by using a data reliability index, and constructing a traffic network digital twinborn model; and the decision and control module is used for analyzing traffic states from the traffic network digital twin model, predicting traffic events and collision risks and generating traffic control instructions. According to the platform, through the equipment fingerprint and protocol reverse technology, an information island is broken, the comprehensiveness and high credibility of a data source are ensured, and the road traffic safety and passing efficiency are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of traffic control, and in particular to an intelligent networked sensing control platform for road traffic safety facilities. Background Art

[0002] With the rapid development of my country's transportation infrastructure, society has placed higher demands on road safety and traffic efficiency. Driven by national policies and the concerted efforts of the industry, the digitalization, networking, and intelligent transformation of the transportation system is accelerating. Among these, the "vehicle-road-cloud integrated" intelligent transportation system, centered on cellular vehicle-to-vehicle (V2X) technology, has become a recognized industry trend due to its wide coverage and low-latency communications. This system is not only a key enabler for achieving high-level autonomous driving but also an effective path to addressing current challenges such as urban traffic congestion and frequent accidents. It aims to establish a "perspective" that transcends the perception capabilities of any single traffic participant through comprehensive, real-time information exchange and collaborative computing between vehicles, roadside infrastructure, and cloud platforms. This allows for the prediction of risks, optimization of traffic control, and routing of traffic, enabling more accurate and efficient traffic decision-making and control.

[0003] However, in the transition to full intelligence, existing technologies still have some common flaws. Currently, a large number of road safety facilities deployed, such as independent video surveillance, radar speed guns, and various active warning devices, are essentially localized systems with single functions and passive responses. These systems operate independently, and the collected data cannot be effectively communicated, forming a typical "information island." Even some safety systems with more comprehensive functions are often limited to their perception and control range at specific points or in a single scene, unable to obtain information on the wider traffic situation. As a result, their decisions are localized and lack foresight, resulting in insufficient capabilities to deal with complex and dynamic traffic scenarios such as cross-regional traffic incidents, chain congestion, sudden crossing of pedestrians or non-motorized vehicles, and inclement weather, creating bottlenecks in safety assurance and efficiency improvement.

[0004] To sum up, it is now necessary to effectively network and intelligently integrate these localized road safety sensors and warning facilities with different functions and dispersed deployments.

[0005] To this end, an intelligent network sensing control platform for road traffic safety facilities is proposed. Summary of the Invention

[0006] The purpose of the present invention is to provide an intelligent networked sensing control platform for road traffic safety facilities to solve the problems raised in the above background technology.

[0007] To achieve the above objectives, the present invention provides the following technical solutions: an intelligent networked sensing control platform for road traffic safety facilities, comprising:

[0008] Traffic data collection module: collects data streams from multiple roadside traffic sensors, establishes a traffic data source identifier by combining physical layer radio frequency characteristics with network communication behavior identification technology, and outputs traffic data with the traffic data source identifier;

[0009] Traffic data processing module: parses the traffic data into standardized traffic parameters according to the traffic data source identification and parsing protocol; and evaluates the data reliability index of the traffic parameters based on an index system covering timeliness, spatial accuracy and integrity;

[0010] Situation Fusion Module: This module uses a graph neural network to associate traffic targets in the traffic parameters to form continuous vehicle driving trajectories, uses the data reliability index as a dynamic weight for data fusion, and generates a global traffic situation map. This global traffic situation map is used to construct and update a dynamic traffic network digital twin model in real time.

[0011] Decision-making and control module: Analyzes traffic status from the traffic network digital twin model, predicts potential traffic incidents and collision risks, and generates traffic control instructions for traffic guidance and early warning to vehicle terminals.

[0012] Preferably, the specific implementation process of generating the traffic data with the traffic data source identifier includes:

[0013] The method comprises extracting physical layer features of the traffic sensor communication signal including signal impairments introduced by hardware tolerances to generate a physical layer radio frequency fingerprint; extracting network behavior features including protocol headers, data packet sequences, and communication timing patterns from the traffic sensor communication data stream to generate a network behavior fingerprint; using the physical layer radio frequency fingerprint and the network behavior fingerprint as input features, fusing them through a weighted scoring mechanism, and calculating and assigning a traffic data source identifier for the traffic sensor.

[0014] Preferably, the specific implementation process of parsing into standardized traffic parameters includes:

[0015] The communication protocol type is determined by querying the device archive based on the traffic data source identifier. If the communication protocol type is known, a standard parser is called. If the communication protocol type is unknown or private, protocol reverse engineering is initiated. The data payload including the message format, field boundaries and data types is inferred through sequence alignment, statistical analysis and heuristic algorithms; the data payload is analyzed by a data pattern inference engine, and key business fields including vehicle identity, speed, location and timestamp are identified and marked, and the traffic data is converted into the standardized traffic parameters.

[0016] Preferably, the specific implementation process of evaluating the reliability index of the traffic parameter data includes:

[0017] Calculate the delay between the timestamp of the traffic parameter and the current processing time to quantify the timeliness; evaluate the spatial positioning error and quantify the spatial accuracy based on the inherent technical specifications of the traffic sensor of the traffic parameter and the current working environment factors; check whether the traffic parameter contains all predefined key business fields to quantify the completeness; and perform a weighted summation on the quantified results of the timeliness, spatial accuracy and completeness to generate the data reliability index.

[0018] Preferably, the specific implementation process of generating the global traffic situation map includes:

[0019] Traffic targets in the traffic parameters that originate from different times and different traffic sensors are used as nodes of the graph; based on the spatiotemporal proximity and appearance feature similarity of the traffic targets, edges representing potential associations are established between the nodes to construct a multi-target association graph; the graph neural network is applied to reason on the multi-target association graph, the association confidence represented by each edge is calculated, and based on the association path with high confidence, the nodes belonging to the same traffic target are connected to form the continuous vehicle driving trajectory; the continuous vehicle driving trajectory and traffic parameters are weighted averaged using the corresponding data reliability index to generate the global traffic situation graph.

[0020] Preferably, the specific generation process of constructing the digital twin model of the traffic network includes:

[0021] Construct a digital twin model of the traffic network that includes the road network topology, static properties of traffic facilities, and traffic rules; use the vehicle driving trajectories and traffic parameters in the global traffic situation map as real-time data sources, and instantiate them as dynamic traffic participant objects with real-time positions, speeds, and properties in the traffic network digital twin model; continuously receive updated global traffic situation maps, and refresh the status and relationships of the dynamic traffic participant objects at a high frequency.

[0022] Preferably, the specific implementation process of generating the traffic control instruction includes:

[0023] The traffic network digital twin model is abstracted into a dynamic spatiotemporal graph, with intersections and road sections as nodes of the graph, connections between nodes as edges, and node characteristics including real-time traffic flow parameters; a spatiotemporal graph neural network is used to analyze the evolution of the dynamic spatiotemporal graph to obtain prediction results including the probability, type and spatiotemporal impact range of traffic events occurring within a specific time window in the future; a hierarchical reinforcement learning framework is applied to make control decisions, with high-level meta-agents setting macro-traffic optimization strategic goals based on the prediction results, and low-level local agents learning and outputting optimal tactical actions based on the traffic optimization strategic goals and the traffic conditions in their respective areas; the tactical actions are converted into specific instructions for traffic guidance and signal timing optimization and the generation of vehicle terminal warning messages containing predictive risk attributes.

[0024] Compared with the prior art, the present invention has the following beneficial effects:

[0025] 1. It improves the breadth and efficiency of information fusion. By combining hybrid identification technology using physical-layer RF fingerprints and network behavioral fingerprints, it establishes unique identity profiles for various heterogeneous facilities. It also adaptively triggers protocol reverse engineering, converting non-standard data into unified, standardized traffic parameters. This improves the seamless integration and data interoperability of regional transportation facilities. This system breaks through the limitations of traditional systems that rely on standard interfaces or manual configuration, enabling automated, low-cost access to existing and new facilities. This significantly alleviates the problem of "information silos" and lays a solid foundation for building a networked, collaborative perception and control system.

[0026] 2. Ensures the accuracy and reliability of situational awareness. By establishing a real-time, multi-dimensional data quality assessment system, a dynamic data reliability index is assessed for each traffic parameter. This index is used as a dynamic weight in the subsequent fusion process, intelligently trusting high-reliability data sources and suppressing interference from low-quality data sources, thereby improving the accuracy and reliability of global traffic situational awareness. This quality-driven fusion mechanism ensures that the resulting situation map is a high-fidelity result of information refinement, rather than a simple aggregation of data, providing high-quality data input for subsequent precise prediction and control.

[0027] 3. Proactive and forward-looking traffic management is achieved. By leveraging a high-fidelity road network model for forward-looking risk prediction, applying a hierarchical reinforcement learning framework for networked collaborative control, and using digital twins for closed-loop verification of control commands, this platform achieves a paradigm shift in traffic management from "passive response" to "active prediction," improving traffic safety and efficiency. The platform can anticipate and mitigate potential risks, implement cross-regional, long-chain collaborative optimization strategies, and ensure the reliable execution of control commands, thereby preventing and mitigating traffic risks at the source and maximizing road network operational efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 This is a structural diagram of an intelligent networked sensing control platform for road traffic safety facilities proposed in an embodiment of the present invention;

[0029] Figure 2 A collaborative perception flow chart proposed in an embodiment of the present invention;

[0030] Figure 3 A prediction and collaborative control closed-loop diagram is proposed for the embodiment of the present invention application. DETAILED DESCRIPTION

[0031] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0032] See also Figure 1-Figure 3 The present invention provides an intelligent network sensing control platform for road traffic safety facilities, and the specific modules are as follows:

[0033] Traffic data collection module: collects data streams from multiple roadside traffic sensors, establishes a traffic data source identifier by combining physical layer radio frequency characteristics with network communication behavior identification technology, and outputs traffic data with the traffic data source identifier;

[0034] Traffic data processing module: parses the traffic data into standardized traffic parameters according to the traffic data source identification and parsing protocol; and evaluates the data reliability index of the traffic parameters based on an index system covering timeliness, spatial accuracy and integrity;

[0035] Situation Fusion Module: This module uses a graph neural network to associate traffic targets in the traffic parameters to form continuous vehicle driving trajectories, uses the data reliability index as a dynamic weight for data fusion, and generates a global traffic situation map. This global traffic situation map is used to construct and update a dynamic traffic network digital twin model in real time.

[0036] Decision-making and control module: Analyzes traffic status from the traffic network digital twin model, predicts potential traffic incidents and collision risks, and generates traffic control instructions for traffic guidance and early warning to vehicle terminals.

[0037] The technical solution of the present invention is further described in detail below with reference to specific embodiments.

[0038] Example 1

[0039] The embodiment of the present application discloses an intelligent network-connected sensing control platform for road traffic safety facilities. At intersection A, the platform is used to sense the current traffic conditions. Figure 1 The specific modules proposed in the present invention include: a traffic data acquisition module: collecting data streams from traffic sensors, and outputting traffic data with the traffic data source identifier through identification technology; a traffic data processing module: parsing the traffic data into standardized traffic parameters according to the traffic data source identifier; and evaluating the data reliability index of the traffic parameters based on a multi-dimensional indicator system; a situation fusion module: forming continuous vehicle driving trajectories through a graph neural network, and generating a global traffic situation map using the data reliability index; using the global traffic situation map to construct and update a dynamic traffic network digital twin model in real time; a decision and control module: analyzing traffic status from the traffic network digital twin model, predicting potential traffic incidents and collision risks, and generating traffic control instructions for traffic guidance and warning vehicle terminals.

[0040] Furthermore, the data streams of multiple roadside traffic sensors are collected, and a traffic data source identifier is established by combining the physical layer radio frequency characteristics with the identification technology of network communication behavior, and the traffic data with the traffic data source identifier is output; corresponding to the above-mentioned traffic data collection module, refer to Figure 2 ; The specific implementation process includes:

[0041] Using a multimodal RF front-end and a network sniffer, the module scans and monitors communication channels in the area surrounding Intersection A. For example, it continuously monitors the 24 GHz frequency band to capture millimeter-wave radar signals, simultaneously monitoring wired Ethernet traffic through switch port mirroring and scanning the LoRaWAN gateway frequency band to discover IoT devices. Upon detecting any regular signal or packet activity, the module locks onto the signal source and captures its raw I / Q sampled data, electrical signal waveform, or network packet. The captured raw signal is processed using a signal analysis algorithm. For wireless devices like millimeter-wave radar, a fast Fourier transform (FFT) is performed on the signal to calculate the stable offset of the carrier frequency from the nominal value. The power spectral density is analyzed to quantify the phase noise at specific offset frequencies. For wired devices like IP cameras, identification is achieved by combining network traffic analysis with device handshake characteristics, primarily including MAC address patterns, network protocol stack fingerprints, and packet timing characteristics. These physical characteristics are quantified to form a multi-dimensional physical fingerprint vector.

[0042] During the observation window of approximately 5 minutes, a statistical analysis of all captured data packets from the same signal source revealed that the millimeter-wave radar would send UDP data packets with a length of consistently 128 bytes at a 50-millisecond period. During RTSP streaming, the IP camera maintained a stable image group length of 50 in the video stream, and the average size ratio of I-frames to P-frames remained within a relatively stable range of 7.5:1. The geomagnetic detector sent 32-byte packets with a single content format at a fixed interval of 2 seconds. These stable patterns in protocol headers, packet timing, payload size, and entropy were ultimately combined into network behavior fingerprint features. As device hardware ages and malfunctions, both the physical layer RF fingerprint and the network behavior fingerprint will drift to a certain extent. In practice, many variable environmental factors can also cause short-term fluctuations in fingerprint characteristics. Therefore, when tracking the time series of each device fingerprint, an adaptive filter is used for preprocessing operations to distinguish between long-term trend drift caused by hardware aging and short-term noise caused by environmental factors. A long-term health drift model of the device fingerprint is constructed through a long-short-term memory network. Abnormal drift conditions that exceed the normal environmental fluctuation range are monitored, and early warning information and predictive maintenance work orders are generated. In this way, the platform is upgraded from a simple traffic sensing system to an intelligent infrastructure management platform with both asset health management and predictive operation and maintenance capabilities, thereby improving its predictive capabilities.

[0043] The N extracted physical layer RF fingerprint features (such as carrier frequency offset (CFO) and I / Q imbalance) and M network behavioral fingerprint features (such as mean inter-packet interval (I / Q) and load entropy) are normalized using the Z-score method to construct a mixed feature vector V of dimension N+M. This vector is then input into a pre-trained support vector machine classification model. This model is trained on fingerprint samples from 5,000 devices of 100 different models. Using radial basis functions as the kernel function, a grid search method is used to determine the optimal hyperparameters (such as C = 10 and gamma = 0.1). This model then calculates the most likely identity of the device and outputs a corresponding confidence score. For example, after the model determines with high confidence that the current device is a "Company D Millimeter Wave Radar - Model A" device in the current RF environment, the platform immediately assigns it a universally unique identification code (UUID: 1a2b3c4d-5e6f-7a8b-9c0d-1e2f3a4b5c6d, etc.) as a traffic data source identifier, and stores it in the device archive together with the complete fingerprint feature data.

[0044] By combining physical-layer RF fingerprints with network behavioral fingerprints, digital identities are established for traffic sensors, solving the costly and time-consuming manual configuration and vendor lock-in issues inherent in traditional system integration. This approach enables reliable identification and documentation of heterogeneous facilities across the entire region, eliminating the need to rely on device-reported information. This provides a solid and reliable authentication foundation for breaking down information silos and enabling subsequent data analysis and collaborative control.

[0045] Furthermore, according to the traffic data source identification adaptation parsing protocol, the traffic data is parsed into standardized traffic parameters; corresponding to the traffic data processing module, refer to Figure 2 ; The specific implementation process includes:

[0046] When receiving a data stream with a traffic data source identifier, the platform uses this identifier to query the device archive to determine the communication protocol type associated with the data source. For example, if the device is listed as "Company H IP Camera" in the archive, the query result will indicate that its protocol is "ONVIF Profile T." The subsequent data stream will then be dispatched to a pre-configured standard ONVIF parser for processing. However, if the device is "Company D Millimeter Wave Radar," its archive will likely be initially marked as "Protocol Unknown." In this case, the platform will route the data stream to the protocol reverse engineering unit for processing.

[0047] For unencrypted unknown protocol data, protocol reverse engineering will be performed. For example, for the UDP data packets sent by the radar of Company D, the module will capture hundreds of consecutive data packets (for example, 500 packets with a total length of 25 seconds) to form an analysis corpus. Using sequence alignment based on the Needleman-Wunsch algorithm, it can be found that the beginning of each 128-byte data packet is a fixed 4-byte 0xDEADBEEF, which leads to the inference that this is likely the location of the message header. The remaining payload is then analyzed using a heuristic rule base. This heuristic rule base covers, but is not limited to, the following aspects: First, field entropy rules. By calculating byte entropy within a 4-byte sliding window, regions with constant but low entropy values ​​are identified as fixed identifiers or status words. Second, field value range and variation patterns are considered. If the value distribution of a 4-byte floating-point field consistently falls within the intersection geofence coordinate range and its variation conforms to the vehicle dynamics model, the field is highly suspected to be a location coordinate field. If a 2-byte integer field exhibits a small linear increase, it is inferred to be the packet sequence number. By applying these rules, the following 2-byte field exhibits a linear increase and is therefore labeled as the "packet sequence number." Furthermore, by performing statistical analysis of the remaining payload based on byte frequency distribution and sliding window entropy, as well as inference based on the heuristic rule base, a variable-length region consisting of multiple repetitive 20-byte structures can be identified. Further analysis of the various fields within these 20 bytes reveals that the value distribution of some 4-byte floating-point fields closely matches the vehicle's coordinates and speed at the intersection. This allows us to infer the complete format of the private protocol: [Message Header]-[Serial Number]-[Number of Targets]-[Target Object 1]-[Target Object 2]..., where each target object contains fields such as ID, 3D coordinates, and speed.

[0048] Whether output from the standard parser or the protocol reverse engineering module, at this point, it's simply structured raw fields (e.g., field_1, field_2, etc.). To assign business meaning to these fields, the data pattern inference engine takes over. Based on a model pre-trained on extensive traffic datasets, this engine analyzes the numerical characteristics, changing patterns, and interrelationships of these fields. For example, in the radar data mentioned above, the engine discovered that one field is a small integer that remains constant over a short period of time, matching the characteristics of an "object ID." Two other fields' values ​​fall within the geofence of Intersection A and are labeled "location coordinates." Two more fields' values ​​match typical vehicle speed ranges and are labeled "speed components." This automatic mapping of field_1 to object_id and field_2 and field_3 to position_x and position_y completes the transformation from raw data into information with clear business semantics.

[0049] Finally, all semantically annotated data is uniformly encapsulated into the platform's predefined JSON format. For example, the vehicle target detected by the radar at a certain moment, with information such as object_id: 101, position_x: 12.5, position_y: -30.2, etc., will be converted into a standardized traffic parameter object: {source_id: UUID: 1a2b..., timestamp: 2025-07-18T20:31:00.050Z, objects: [{object_id: 101, object_type: vehicle, position: {lat: 34.0522, lon: -118.2437}, velocity: {speed_kph: 45.5, heading: 182.3}}]}.

[0050] Through protocol reverse engineering and data pattern inference, the platform understands and utilizes previously inaccessible devices that use unencrypted, proprietary, or unknown protocols. This unlocks the value of a large number of older or non-standard road infrastructure, avoiding costly equipment replacement. By converting raw bitstreams from any source into standardized traffic parameters with clear business semantics, it eliminates protocol barriers between devices and provides a rich source of data for building a comprehensive perception network.

[0051] Furthermore, the data reliability index of the traffic parameters is evaluated based on the index system covering timeliness, spatial accuracy and completeness, see Figure 2 ; The specific implementation process includes:

[0052] Timeliness is quantified by extracting the timestamp field from traffic parameter objects and comparing it with the platform's internal, precisely synchronized system clock using the Network Time Protocol (NTP). The difference between the two represents the end-to-end data latency. For example, a traffic parameter from Company D's millimeter-wave radar has a timestamp of 2025-07-18T20:31:00.050Z, while the platform processes this data using a system time of 2025-07-18T20:31:00.100Z. This latency is calculated to be 50 milliseconds. The platform then converts this latency into a timeliness score (for example, a latency of less than 100 milliseconds is scored as 0.95) based on a pre-defined latency-to-score mapping curve. For geomagnetic sensor data transmitted over a low-power LoRaWAN network, the latency may reach 1.5 seconds, resulting in a correspondingly lower timeliness score (for example, 0.60).

[0053] The module quantitatively assesses the spatial accuracy of parameters. This assessment goes beyond simply reading the nominal accuracy stated in the device manual. Instead, it's a dynamic, context-aware process. The module first retrieves the sensor's inherent technical specifications (e.g., the pixel resolution of Company H's IP camera, the angular resolution of Company D's radar) from the device's archive and combines them with real-time environmental information from other data sources. For example, when a weather sensor reports heavy rain, the system invokes an environmental impact assessment model pre-trained using historical data. This model uses real-time environmental parameters (such as rainfall, light intensity, and temperature) as input, combined with the sensor's inherent specifications, to output a dynamic spatial degradation factor. Based on this factor, the camera's spatial accuracy rating is downgraded. For example, Company D's millimeter-wave radar might maintain a stable positioning error of ±0.1 meters in clear weather, with a spatial accuracy score of 0.99. However, in heavy rain, the camera's positioning error might increase to ±2 meters, and its spatial accuracy score might drop to 0.70.

[0054] Secondly, the module conducts a quantitative assessment of the integrity of the parameters. The platform will check whether the currently received traffic parameter object contains all the required key business fields based on the predefined complete data model of this type of data. For example, a complete vehicle target object should contain all fields such as object_id, position, velocity, etc. If the platform finds that the velocity field is missing in a piece of data due to network packet loss or overload of the sensor's internal processing unit, then the integrity of this data will be judged as flawed. Its integrity score will be calculated based on the importance and number of missing fields. For example, the score for missing minor fields is 0.90, while the score for missing core fields such as position may drop directly to 0.30.

[0055] Finally, the module performs weighted generation of a Data Quality Score (DQS). It sums the quantitative scores for the three aforementioned dimensions (timeliness, spatial accuracy, and completeness) according to a configurable weighting formula to generate a final, single Data Quality Score (DQS). The weights of each dimension are adjusted based on the application scenario. For example, in collision warning applications, where spatial accuracy and timeliness are extremely demanding, their weights might be set to 0.5 and 0.4 respectively. In macro-traffic flow analysis, where completeness is a higher requirement, the weights might be increased. Ultimately, each standardized traffic parameter is assigned a DQS value between 0 and 1 (for example, 0.92).

[0056] A comprehensive data quality score (DQS) is dynamically assessed in real time for every standardized traffic parameter entering the platform, establishing a unified, quantitative credibility standard for previously uneven data. By conducting contextually aware assessments of the data's timeliness, spatial accuracy, and completeness, the platform clearly understands the reliability of each piece of information. This provides crucial quality assurance for subsequent high-quality data fusion and high-risk, high-value traffic control decisions.

[0057] Furthermore, the traffic targets in the traffic parameters are associated with each other through the graph neural network to form a continuous vehicle driving trajectory, and the data reliability index is used as a dynamic weight to perform data fusion to generate a global traffic situation map; For the corresponding situation fusion module, refer to Figure 2 ; The specific implementation process includes:

[0058] The platform uses all traffic object detection results generated by all sensors within a brief time window (e.g., 5 seconds) (e.g., point cloud objects from radar, detection boxes from cameras, and vehicle presence signals from geomagnetic coils) as nodes in a graph. Based on physical likelihood and feature similarity, it establishes edges between these nodes, representing potential connections. For example, if radar A detects an object at location P1 at time t=0, and camera B detects an object at location P2 at time t=0.5 seconds, an edge is established between these two nodes if the distance between P1 and P2 (e.g., 10 meters) falls within the normal speed range for that road segment. If camera B also provides color features (e.g., red) and radar A provides radar cross-section (RCS) data, the similarity of these features is used as an initial basis for calculating edge weights. This process constructs a complex spatiotemporal graph encompassing all possible connections. The complex spatiotemporal graph is fed into a pre-trained graph neural network (GNN) model consisting of a stack of two graph convolutional networks (GCNs) and one gated recurrent unit (GRU). Unlike traditional methods, GNNs can learn global, high-order association patterns, rather than just pairwise comparisons. For example, when a vehicle is completely obscured between two cameras by a large truck for three seconds, the GNN analyzes the flow and context of other vehicles in the area and infers with high confidence that the object that reappears after the occlusion is the same entity as the one that disappeared before the occlusion. The GNN infers the association confidence of each edge in the graph and then retains only those edges with a confidence score above a certain threshold (e.g., 0.90). Along these high-confidence connecting paths, the platform "stitches" discrete nodes belonging to the same traffic object into a smooth, complete, and uniquely identified long-term trajectory.

[0059] At this point, the platform has continuous trajectories for all targets. However, the same target may still have data from multiple sensors at the same moment (for example, radar and camera simultaneously provide its position and velocity). The module then uses the DQS values ​​generated in step S4 as dynamic weights to perform a weighted fusion of these redundant data. For example, at a given moment, Company D's radar (with a current DQS of 0.99) reports a vehicle's position as (12.50, -30.20), while Company H's camera, affected by heavy rain (with a current DQS of only 0.70), reports its position as (12.65, -30.45). The platform then calculates the final fused position as (Position_Radar * 0.99 + Position_Camera * 0.70) / (0.99 + 0.70), which closely approximates the highly reliable radar data. This weighted fusion process applies to all fusible kinematic parameters, including position, velocity, acceleration, and orientation.

[0060] All high-reliability, long-term trajectories fused through quality perception and other traffic parameters that cannot be associated with trajectories (such as average speed on a road section and traffic incident alerts) are superimposed on a high-precision geographic information layer. This ultimately forms a unified, dynamic, and high-fidelity global traffic situation map.

[0061] By stitching together target trajectories across visual domains through a graph neural network and using DQS as a dynamic weight for data fusion, the platform generates a global traffic situation map that is far more complete, accurate, and reliable than any single sensor's perception. This process not only addresses the issue of target loss in complex scenarios (such as occlusion and congestion), but also intelligently favors the most reliable data source at the time. This enables the platform to obtain a "perspective" that is closest to "ground truth," providing an unprecedented high-quality, high-fidelity data foundation for subsequent digital twin modeling and risk prediction.

[0062] Furthermore, the global traffic situation map is used to construct and update the dynamic traffic network digital twin model in real time. Figure 3 ; The specific implementation process includes:

[0063] The platform loads three key types of static data from its backend database: first, high-precision map data, which contains the precise geometry, topological connectivity, stop line locations, and slope curvature of all lanes in Intersection A and the surrounding area; second, traffic infrastructure asset data, which records the precise 3D coordinates, model, and inherent properties of each traffic light, variable message board, and camera; and third, traffic rule data, which defines the legal speed limit, turn restrictions (such as "no left turn during peak hours"), and standard signal phasing schemes at intersections. Together, these three types of data form a static yet semantically rich digital twin model of the traffic network. Using the global traffic situation map as a real-time data source, its dynamic elements are mapped onto the traffic network digital twin model. For example, a high-reliability, long-term trajectory with the ID Vehicle-123 in the situation map, after quality-aware fusion, is instantiated in the baseline network model as a vehicle object with a corresponding 3D appearance. The vehicle's real-time position, velocity, acceleration, heading, and other kinematic parameters strictly follow the trajectory data in the situation map. At the same time, other traffic parameters, such as the average speed of a road section, will be mapped into dynamic properties of the road section object. Through this process, the static map is "activated" into a scene containing vivid traffic participants.

[0064] To ensure strict synchronization between the digital twin and the physical world, the platform receives updated global traffic maps at a configurable frequency (e.g., 0.5-1 Hz for routine monitoring, 1-2 Hz for emergency response) based on the needs of different application scenarios, computing resources, and network bandwidth. The specific update frequency is dynamically adjusted based on the current processing load. Each update triggers a refresh of the status of all dynamic traffic participant objects in the road network model. More importantly, this process also serves as a closed-loop verification process. For example, if a control command requires that traffic light 2 at intersection A turn green, this module will continuously monitor the actual traffic flow through traffic light 2, as reflected by the fused data in the digital twin. If traffic flow does not start as expected within a preset time (e.g., 2 seconds), the system will determine that the command execution failed or that the physical device has failed, and feedback this abnormal status to the decision module. Through the framework of a generative adversarial network, the generator continuously creates extremely dangerous traffic scenarios, and the discriminator responds to or resolves these extreme scenarios. An adversarial relationship forms between the two: the generator strives to create scenarios that will cause the discriminator to "fail," while the discriminator continuously learns from this adversarial process, striving to improve its ability to cope with various unknown risks. Through this self-adversarial and self-competitive training process, the platform can automatically and systematically identify weak links in its control algorithms and strengthen them accordingly. This ensures that the control system undergoes a far more rigorous and comprehensive "stress test" than any historical dataset before being deployed in the real world, thereby improving its safety and robustness.

[0065] Building a digital twin model of the transportation network elevates discrete sensory data into a structured, interactive, and deducible digital twin environment. This model not only accurately mirrors the physical world but also serves as a virtual testing ground, enabling analysis of the causal relationships of complex traffic phenomena and simulation and verification of the effectiveness of control strategies. Furthermore, its closed-loop verification capabilities ensure the reliable operation of the entire system.

[0066] Furthermore, the traffic status is analyzed from the traffic network digital twin model to predict potential traffic incidents and collision risks, and traffic control instructions for traffic guidance and warnings to vehicle terminals are generated; corresponding to the decision-making and control module; the specific implementation process includes:

[0067] The platform takes the dynamic traffic network digital twin model as input and abstracts it into a dynamic space-time graph. In the graph, each intersection and key section is defined as a node of the graph, and the roads connecting these nodes are defined as edges. The module extracts rich features from the road network model in real time to fill the graph: the node features include the real-time traffic flow, average speed, lane occupancy and vehicle type distribution at that location (for example, heavy trucks account for 15%); the edge weight dynamically represents the real-time travel time or association strength between nodes.

[0068] The dynamic spatiotemporal graph is fed into a spatiotemporal graph neural network (STGNN) model trained with massive amounts of historical and simulation data. This model can predict the probability, type, and spatiotemporal impact of traffic events within a specific future time window (e.g., the next 15 minutes). For example, by analyzing the merging rate of traffic on the upstream ramp of Intersection A and the traffic density on the main road, combined with historical statistical data and current traffic patterns, the model predicts: "The risk of a queue overflow in the next 2-5 minutes is high (based on pre-set risk assessment criteria)."

[0069] Based on the regional risk prediction map output by the STGNN, decisions are made within a hierarchical reinforcement learning framework constrained by safety rules. This framework consists of two layers: a high-level meta-agent is responsible for observing the macro-risk landscape and setting strategic goals. For example, for the aforementioned overflow prediction, the meta-agent might set the strategic goal of "reducing ramp congestion by 50% within 10 minutes without significantly impacting main road traffic efficiency." Multiple local agents at the bottom level, such as those controlling the ramp signal and the signals at the upstream intersection, receive these goals and, in combination with more detailed real-time data, learn and output optimal tactical actions within their local decision space (such as adjusting signal timings and phase differences) that most efficiently achieve the macro-goal while passing deterministic safety checks and not violating any safety rules. The state space S of the local agents at the bottom level is defined as a feature vector containing [the real-time queue length vector of each entrance lane, the average waiting time vector of each entrance lane, and the upstream flow rate of the main road]. The action space A is designed as a discrete set containing five preset signal timing schemes tailored to different traffic flow patterns. The hierarchical reinforcement learning framework also introduces a more resilient decision-making mechanism. In the framework, the relationship between multiple local agents at the bottom layer is modeled as a "cooperative game." Each agent has its own independent reward function, as well as a shared reward function based on the macro-goal of the entire region. This mechanism allows for "altruistic" collaboration between agents. For example, the agent at intersection A may discover through calculation that if it reduces the green light time by 5 seconds, it can create a complete "green wave belt" for intersection B, thereby bringing shared positive rewards to the entire system. The resulting regional traffic control strategy is much more flexible and efficient than traditional centralized commands, especially when dealing with emergencies, with significantly stronger adaptability.

[0070] Finally, the framework's output of the optimal tactical action (e.g., reducing the green signal ratio of ramp lights by 15% and extending the green signal of the upstream main road by 10 seconds) is converted into a series of specific, standardized control instructions, which are then issued to the appropriate traffic control devices. Simultaneously, the platform dynamically generates content-enhanced, predictive V2X messages. For example, an e-DENM message containing fields such as predictedRiskProbability: 0.9, riskType: QUEUE_SPILLBACK, timeToImpact: 300s, and recommendedAction: MERGE_LANE_EARLY is generated and broadcasted via roadside units to connected vehicles about to enter the area, guiding drivers to take evasive action in advance.

[0071] Proactive risk propagation predictions based on digital twins and collaborative control decision-making using a hierarchical reinforcement learning framework have transformed the traditional traffic management model of passively responding to single issues. By predicting and proactively intervening in potential congestion and accident risks, and developing collaborative optimization strategies from a macro perspective across the entire road network, traffic risks are prevented and resolved at the source. This shifts the focus of traffic management from "how to handle an incident after it occurs" to "how to prevent it from happening," fundamentally improving road safety and efficiency.

[0072] The platform disclosed in the present invention accesses and parses old and new, heterogeneous roadside facilities through hybrid device fingerprint and protocol reverse technology, and establishes a real-time data quality assessment system, breaking the "information island" and ensuring the comprehensiveness and high credibility of the data source. Secondly, by using quality-driven fusion algorithms and graph neural networks, multi-source data are refined into a high-fidelity global traffic situation map, and an interactive and deducible dynamic digital twin model is constructed, realizing a qualitative change from discrete perception to global situation understanding. Based on digital twins, forward-looking risk propagation predictions are carried out, and a hierarchical reinforcement learning framework is applied to formulate networked collaborative control strategies, transforming traffic management from a "post-event response" to a new paradigm of "pre-risk intervention", avoiding congestion and accidents at the source, and significantly improving road traffic safety and traffic efficiency.

[0073] Example 2

[0074] At Intersection B, a complex, multi-phase intersection connecting a main road and an urban expressway ramp, it is currently evening rush hour, with saturated west-to-eastbound traffic and queues forming at the ramp entrance. The specific implementation of the traffic data collection module is as follows:

[0075] The platform's sensing and control process was activated at Intersection B. First, the data acquisition module monitored the intersection's communication channels in real time, capturing signals from three primary data sources: a LiDAR sensor from Company L located in the northwest corner of the intersection, a panoramic camera from Company C located in the southeast corner, and a wireless magnetometer from Company W at the ramp entrance. The module analyzed the RF signal from the wireless magnetometer from Company W, which communicates via a proprietary protocol in the 433 MHz frequency band, extracting its physical layer RF fingerprint, which showed a carrier frequency offset of +4.1 kHz and a signal envelope rise time of 15 microseconds. A fast Fourier transform (FFT) was applied to the captured I / Q sampled data to determine the peak frequency, which was then compared with the nominal frequency to determine the offset. Simultaneously, the network analysis unit performed differential statistics on the timestamps of consecutive data packets and analyzed the packet length distribution histogram. The module found that the vehicle transmitted 18-byte packets with an average period of approximately 1.5 seconds, forming its network behavior fingerprint. The quantized feature vectors of these two fingerprints are normalized using the Z-score to eliminate dimensionality. These features are then input into a support vector machine (SVM) classification model using a radial basis function (RBF) kernel. The platform determines the vehicle's identity and assigns it a unique traffic data source identifier (UUID): fabc4567-89de-f012-3456-789abcdef0. For lidar and cameras, the platform also analyzes features such as the PHY chip clock frequency deviation (+48ppm) and the I-frame interval jitter of RTSP streaming (standard deviation 12ms) to complete the identification process.

[0076] Example 3

[0077] At intersection B, the specific implementation of the traffic data processing module is as follows:

[0078] Next, the platform received an 18-byte raw data stream from a device with the UUID: fabc4567-89de-f012-3456-789abcdef0 (a magnetometer from Company W). Because the device archive indicated an unknown protocol, the data stream was routed to the protocol reverse engineering unit. Using sequence alignment technology based on the Needleman-Wunsch algorithm, the continuous data packet byte sequence was treated like a DNA sequence in biology. Alignment operations were performed to identify highly conserved regions, revealing that bytes 1 to 4 were a fixed device serial number. By calculating the entropy and conditional probabilities of subsequent bytes, it was discovered that the fifth byte alternated between 0x00 (no vehicle) and 0x01 (vehicle present), consistent with the characteristics of a status field and thus inferred to be the status field. Bytes 6 and 7 were a 16-bit integer representing the duration of the vehicle's presence (in milliseconds). The data pattern inference engine then intervened, labeling the fifth byte as occupancy_status and the sixth and seventh bytes as occupancy_duration_ms. Finally, a raw binary data stream is converted into a standardized traffic parameter object: {source_id: UUID:fabc4567-89de-f012-3456-789abcdef0, timestamp:2025-07-20T18:44:30.150Z, events:[{event_type:occupancy, status:1, duration_ms:2350}]}.

[0079] After data standardization, the platform conducts real-time, multi-dimensional data quality assessments of traffic parameters from all sources. Data from Company L's lidar has a timestamp latency of only 20 milliseconds, and its 3D point cloud has a positioning error of less than ±5 cm in clear weather conditions. Data fields are intact, and its data reliability score (DQS) is 0.98, calculated using a weighted summation formula (where timeliness is weighted as 0.4, spatial accuracy is weighted as 0.5, and completeness is weighted as 0.1). Data from Company C's panoramic camera has a reduced confidence level for object detection in the west area of ​​the intersection due to localized glare from the sun. The analysis module dynamically downgrades its spatial accuracy score by querying a built-in lookup table that maps meteorological data (e.g., solar altitude 22°, azimuth 255°) and sensor orientation (270°) to a DOP (0.88), resulting in a final DQS of 0.85. The data from Company W's magnetometer has a data delay of 1.5 seconds and can only provide existence information but not precise location. Its DQS is rated as 0.55.

[0080] Example 4

[0081] At intersection B, the specific implementation of the situation fusion module is as follows:

[0082] The platform's situational fusion module then began processing a white SUV crossing the intersection from west to east. The module's Graph Attention Network (GAT) model had already stitched together the SUV's discrete detections from the past few seconds (each detection represented as a node containing a 2D bounding box, a 128-dimensional Re-ID feature vector, a timestamp, and the sensor's DQS) into a continuous, long-term trajectory. The module performed a quality-aware weighted fusion of the position and velocity of the two data sources. Company L's lidar (DQS = 0.98) reported a position of (x: 15.22, y: -5.10) and a speed of 45.1 km / h; however, Company C's camera (DQS = 0.85), affected by perspective and glare, reported a position of (x: 15.45, y: -5.32) and a speed of 44.5 km / h. The final fused position is calculated by multiplying the lidar position coordinate vector by its quality score of 0.98, adding the camera position coordinate vector by its quality score of 0.85, and then dividing each component of the resulting vector by the sum of the two quality scores (1.83). The X coordinate is (15.22 * 0.98 + 15.45 * 0.85) / 1.83 ≈ 15.33; the Y coordinate is (-5.10 * 0.98 + -5.32 * 0.85) / 1.83 ≈ -5.2. The final fused position is (x: 15.33, y: -5.20), and the speed is 44.9 km / h, which clearly favors the highly reliable lidar data.

[0083] Based on the fused, highly reliable trajectories, the platform updates the dynamic traffic network digital twin model of Intersection B in real time. Each fused trajectory is instantiated as a renderable 3D vehicle object, and its position and rotation properties are bound to the trajectory data. This binding ensures that the digital twin maintains synchronization with the physical world at a frequency of 10 Hz. The model clearly shows that in the west-to-east through lane, 22 vehicles have been instantiated as dynamic objects, and their average speed has dropped to 15 km / h. In contrast, the east-to-west direction is much smoother, with an average speed of 55 km / h. More importantly, the model reveals a potential risk: the right-turn lane leading to the expressway ramp is already long, with the tail of the queue extending into the through lane of the main road, intersecting with the through traffic flow.

[0084] Example 5

[0085] At intersection B, the specific implementation of the decision and control module is as follows:

[0086] Finally, the platform's decision-making and control module receives the dynamic traffic network digital twin model and abstracts it into a dynamic spatiotemporal graph. A spatiotemporal graph neural network (STGNN) model, composed of a graph convolutional network (GCN) and a gated recurrent unit (GRU), analyzes the evolution of this graph. The GCN layer captures the spatial dependencies between Intersection B and its adjacent intersections, while the GRU layer learns how traffic conditions (such as queue length and lane occupancy) evolve over time. The model's softmax output layer generates a high-probability risk prediction: "In the next three minutes, the probability of a rear-end or lateral collision due to ramp overflow on the west side of Intersection B is 85%." Based on this prediction, a high-level meta-agent using the "soft actor-critic" (SAC) algorithm sets strategic goals, such as {target ramp queue length: <20 meters, target main road delay increment: <10 seconds}." After receiving this target, the underlying local agent controlling the traffic lights at Intersection B, using the Proximal Policy Optimization (PPO) algorithm, rapidly learned and optimized within its action space, which included five preset emergency signal scenarios and two fine-tuning actions, to output a tactical action. This tactical action was then translated into a specific control instruction: the traffic light controller at Intersection B reduced the green time for left turns from east to south by 5 seconds and increased the green time for straight-through traffic from west to east by 5 seconds in the current cycle. The controller also generated an e-DENM message containing fields such as predictedRiskProbability: 0.85, riskType: TRAFFIC_JAM_MERGING_CONFLICT, and recommendedAction: REDUCE_SPEED_AND_WATCH_FOR_MERGING_VEHICLES. This message was broadcast via the roadside unit to all connected vehicles within a 500-meter radius west of Intersection B. After issuing the instruction, the platform continuously monitored actual traffic flow at Intersection B through the digital twin to verify the effectiveness of the control instruction.

[0087] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent networked sensing control platform for road traffic safety facilities, characterized in that: include: Traffic data collection module: collects data streams from multiple roadside traffic sensors, establishes a traffic data source identifier by combining physical layer radio frequency characteristics with network communication behavior identification technology, and outputs traffic data with the traffic data source identifier; Traffic data processing module: parses the traffic data into standardized traffic parameters according to the traffic data source identification and parsing protocol; and evaluates the data reliability index of the traffic parameters based on an index system covering timeliness, spatial accuracy and integrity; Situation Fusion Module: This module uses a graph neural network to associate traffic targets in the traffic parameters to form continuous vehicle driving trajectories, uses the data reliability index as a dynamic weight for data fusion, and generates a global traffic situation map. This global traffic situation map is used to construct and update a dynamic traffic network digital twin model in real time. Decision-making and control module: Analyzes traffic status from the traffic network digital twin model, predicts potential traffic incidents and collision risks, and generates traffic control instructions for traffic guidance and early warning to vehicle terminals.

2. The intelligent networked sensing control platform for road traffic safety facilities according to claim 1 is characterized in that: The specific implementation process of generating traffic data with the traffic data source identifier includes: The method comprises extracting physical layer features of the traffic sensor communication signal including signal impairments introduced by hardware tolerances to generate a physical layer radio frequency fingerprint; extracting network behavior features including protocol headers, data packet sequences, and communication timing patterns from the traffic sensor communication data stream to generate a network behavior fingerprint; using the physical layer radio frequency fingerprint and the network behavior fingerprint as input features, fusing them through a weighted scoring mechanism, and calculating and assigning a traffic data source identifier for the traffic sensor.

3. The intelligent networked sensing control platform for road traffic safety facilities according to claim 1 is characterized in that: The specific implementation process of parsing into standardized traffic parameters includes: The communication protocol type is determined by querying the device archive based on the traffic data source identifier. If the communication protocol type is known, a standard parser is called. If the protocol type is unknown or private, protocol reverse engineering is initiated for the unknown protocol that meets the regular characteristics. The data payload including the message format, field boundaries and data types is inferred through sequence alignment, statistical analysis and heuristic algorithms; the data payload is analyzed by a data pattern inference engine to identify and mark the key business fields including vehicle identity, speed, location and timestamp, and the traffic data is converted into the standardized traffic parameters.

4. The intelligent networked sensing control platform for road traffic safety facilities according to claim 1, characterized in that: The specific implementation process of evaluating the reliability index of the traffic parameter data includes: Calculate the delay between the timestamp of the traffic parameter and the current processing time to quantify the timeliness; evaluate the spatial positioning error and quantify the spatial accuracy based on the inherent technical specifications of the traffic sensor of the traffic parameter and the current working environment factors; check whether the traffic parameter contains all predefined key business fields to quantify the completeness; and perform a weighted summation on the quantified results of the timeliness, spatial accuracy and completeness to generate the data reliability index.

5. The intelligent networked sensing control platform for road traffic safety facilities according to claim 1 is characterized in that: The specific implementation process of generating the global traffic situation map includes: Traffic targets in the traffic parameters that originate from different times and different traffic sensors are used as nodes of the graph; based on the spatiotemporal proximity and appearance feature similarity of the traffic targets, edges representing potential associations are established between the nodes to construct a multi-target association graph; the graph neural network is applied to reason on the multi-target association graph, the association confidence represented by each edge is calculated, and based on the association path with high confidence, the nodes belonging to the same traffic target are connected to form the continuous vehicle driving trajectory; the continuous vehicle driving trajectory and traffic parameters are weighted averaged using the corresponding data reliability index to generate the global traffic situation graph.

6. The intelligent networked sensing control platform for road traffic safety facilities according to claim 1, characterized in that: The specific generation process of constructing the digital twin model of the traffic network includes: Construct a digital twin model of the traffic network that includes the road network topology, static properties of traffic facilities, and traffic rules; use the vehicle driving trajectories and traffic parameters in the global traffic situation map as real-time data sources, and instantiate them as dynamic traffic participant objects with real-time positions, speeds, and properties in the traffic network digital twin model; continuously receive updated global traffic situation maps, and refresh the status and relationships of the dynamic traffic participant objects at a high frequency.

7. The intelligent networked sensing control platform for road traffic safety facilities according to claim 1, characterized in that: The specific implementation process of generating the traffic control instruction includes: The traffic network digital twin model is abstracted into a dynamic spatiotemporal graph, with intersections and road sections as nodes of the graph, connections between nodes as edges, and node characteristics including real-time traffic flow parameters; a spatiotemporal graph neural network is used to analyze the evolution of the dynamic spatiotemporal graph to obtain prediction results including the probability, type and spatiotemporal impact range of traffic events occurring within a specific time window in the future; a hierarchical reinforcement learning framework is applied to make control decisions, with high-level meta-agents setting macro-traffic optimization strategic goals based on the prediction results, and low-level local agents learning and outputting optimal tactical actions based on the traffic optimization strategic goals and the traffic conditions in their respective areas; the tactical actions are converted into specific instructions for traffic guidance and signal timing optimization and the generation of vehicle terminal warning messages containing predictive risk attributes.

Images