An internet-based high-speed entrance and exit auxiliary early warning system and method

Abstract

This invention relates to the field of highway early warning system technology, and discloses an internet-based highway entrance / exit auxiliary early warning system and method. The system includes the following modules: a node sensing module deployed at key locations at highway entrances / exits for collecting traffic target data; a data fusion processing module connected to the node sensing module via the internet for receiving and processing data; and an early warning information dissemination module connected to the data fusion processing module for disseminating early warning information. The data fusion processing module includes: a device health assessment unit for calculating the health score of each node sensing module in real time; and a dynamic fault-tolerant compensation unit for dynamically adjusting the weight of each node sensing module in the fusion decision based on the health score. When a node sensing module malfunctions or fails due to unforeseen circumstances, this invention can perform dynamic fault-tolerant compensation to maintain the early warning function and provide accurate data support for system maintenance.

Description

Technical Field

[0001] This invention relates to the field of high-speed early warning system technology, specifically to an internet-based high-speed entrance / exit auxiliary early warning system and method. Background Technology

[0002] The highway entrance and exit auxiliary early warning system is applied to key points at highway entrances and exits, such as the divergence points between the main road and ramps, and the merging points between the main road and ramps. By deploying integrated radar and video detection units, it can perceive and monitor key points at highway entrances and exits in real time, record data such as vehicle color, license plate, and behavioral details on the road, and align and merge real-time road information with corresponding time nodes to form a three-dimensional perception and monitoring of traffic targets at key points at highway entrances and exits. Under the continuous perception and monitoring, it improves the accuracy and effectiveness of traffic information monitoring at key points at highway entrances and exits, and performs information monitoring and early warning to continuously ensure road traffic safety. Existing highway entrance and exit auxiliary early warning systems typically employ device heartbeat detection or simple threshold alarms to ensure the effective performance of sensing equipment such as radar and cameras during severe weather or minor malfunctions. This is achieved by adding redundant devices for backup. Multiple redundant sensing devices work together to monitor traffic targets and information at key points on highway entrances and exits. This approach avoids a sharp decline in data reliability and accuracy due to severe weather (rain, snow, fog) or device malfunctions (lens damage, radar drift), which could lead to system misjudgments and missed detections. However, the redundant backup significantly increases the cost of equipment use and maintenance, hindering timely problem detection and location by maintenance personnel. Therefore, in emergencies, such as severe weather or minor equipment malfunctions, the traffic target information at key points on highway entrances and exits may be distorted and interfered with, affecting the rapid generation and transmission of auxiliary early warning instructions. Furthermore, the continuity of real-time regional monitoring data transmission is difficult to guarantee, easily creating local blind spots due to severely degraded data quality or temporary failures at certain sensing nodes.

[0003] To address the aforementioned issues, there is an urgent need for innovative design based on the existing highway entrance and exit auxiliary early warning system. Summary of the Invention

[0004] The purpose of this invention is to provide an Internet-based highway entrance / exit auxiliary early warning system and method to solve the problem mentioned in the background art where existing highway entrance / exit auxiliary early warning systems create local blind spots due to severe weather or minor malfunctions in the sensing and monitoring equipment, leading to decreased data reliability and resulting in system misjudgments and missed judgments.

[0005] To achieve the above objectives, the present invention provides the following technical solution: an Internet-based high-speed entrance / exit auxiliary early warning system, comprising the following modules: The node perception module is deployed at key locations at highway entrances and exits to collect traffic target data. The data fusion processing module is connected to the node perception module via the Internet to receive and process traffic target data. The data fusion processing module includes an equipment health assessment unit and a dynamic fault tolerance compensation unit. The equipment health assessment unit is used to calculate the health score of each node perception module in real time. The dynamic fault tolerance compensation unit is used to dynamically adjust the weight of each node perception module in the fusion decision based on the health score, and to perform data compensation inference for monitoring blind spots formed by nodes with substandard data quality. The early warning information release module is connected to the data fusion and processing module and is used to release early warning information.

[0006] Preferably, the node perception module includes a radar-visual fusion unit, which includes a millimeter-wave radar and a camera; The device health assessment unit calculates a health score by analyzing the logical characteristics of monitoring parameters, including the average contrast and clarity of the camera's video stream, the density stability of radar point cloud data, and the success rate of radar target trajectory matching.

[0007] Preferably, the dynamic fault-tolerant compensation unit dynamically allocates the weights of each node perception module in data fusion based on the real-time health score of the node perception module. The dynamic allocation methods include: the higher the health score, the greater the weight; when the health score is lower than the preset fault threshold, the traffic target data of the node sensed by the node's sensing module is isolated.

[0008] Preferably, the weight The calculation formula is:

[0009] in This represents the weight of the i-th sensing node. This represents the health score of the i-th sensing node. Scaling factor The total number of nodes participating in the fusion; weight Dynamic weight allocation based on health scores is implemented, tilting data trust towards node perception modules with high health scores to maintain the overall reliability of the system's perception data.

[0010] Preferably, the dynamic fault-tolerant compensation unit performs data compensation inference for the monitoring blind zone. It uses the node traffic target data sensed by the node sensing module adjacent to the blind zone and infers and predicts the traffic state in the blind zone through a pre-trained spatiotemporal correlation model to achieve dynamic fault-tolerant compensation for traffic target data. The monitoring blind zone is caused by the damage or failure of the node sensing module in the corresponding area.

[0011] Preferably, the spatiotemporal correlation model is a graph neural network, which takes the road network topology and real-time traffic flow parameters as input to predict vehicle speed and density information in blind spots.

[0012] Preferably, the data fusion processing module also accesses vehicle GPS data and mobile phone signaling data via the Internet as an auxiliary verification source for the data compensation inference results; This method integrates vehicle-mounted GPS data and mobile phone signaling data with data acquired by node perception modules, verifies the accuracy of source data, corrects errors from single data sources, and obtains the actual traffic conditions.

[0013] Preferably, the data fusion processing module further includes a data credibility assessment unit, which is used to evaluate and verify the introduced vehicle GPS data and mobile phone signaling data. Based on data indicators, the credibility weights of vehicle GPS data and mobile phone signaling data are dynamically calculated to verify conflicts between multiple data sources and ensure data accuracy and security. The data indicators include data source reliability, time correspondence, and spatial consistency.

[0014] Preferably, the warning information release module releases differentiated warning information to drivers through roadside variable message signs, broadcasts, and vehicle-to-everything (V2X) communication; it also sends abnormal warnings of the node perception module to the warning system control center.

[0015] This invention also provides an internet-based auxiliary early warning method for highway entrances and exits, the specific steps of which are as follows: S1. Use of the node perception module: The node perception module is used to collect traffic target data from different sources, synchronously collect video data and radar point cloud data at a fixed frequency, and upload them to the data fusion processing module in real time via the Internet; before transmission, the data is timestamped and preliminarily filtered to remove invalid data caused by GPS drift or equipment failure. S2. Determine the health assessment of the data collected by the node perception module: The device health assessment unit of the data fusion processing module analyzes the data quality uploaded by each node perception module in real time. By calculating the average contrast and clarity of the video stream, the stability of the radar point cloud density, and comparing the matching rate of radar vision data for the same target trajectory recognition, a quantitative health score is generated. S3. Dynamic Weight Adjustment and Data Compensation: The dynamic fault-tolerant compensation unit dynamically allocates the weights of the nodes sensed by each node perception module in the fusion decision based on the latest health score. The weights of the node perception modules with high health scores are increased, while the data of the node perception modules with scores below the threshold are temporarily isolated. At the same time, for the monitoring blind spots caused by the failure of node perception modules in local areas, the dynamic fault-tolerant compensation unit calls the pre-trained spatiotemporal correlation model, combines the real-time data of adjacent healthy nodes with historical traffic flow patterns, and performs short-term traffic state inference to compensate for the missing information. S4. Warning Command Generation and Warning Information Release: The warning information release module uses the merged traffic data to determine the current risk level and generate differentiated warning commands for sending and issuing alerts.

[0016] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention acquires accurate traffic target data in real time through node perception modules and calculates the health score of the node perception modules simultaneously. When a node perception module malfunctions or is damaged due to a sudden situation, dynamic fault-tolerant compensation is performed to ensure information continuity. Compensation reasoning is performed using data from adjacent healthy nodes and historical patterns. This ensures that even when the performance of some devices degrades, the overall perception and early warning capabilities of the system will not be interrupted. It effectively avoids system misjudgment or failure caused by a single sensor problem and provides timely and effective early warning processing, improving the stability of use and the accuracy and efficiency of early warning data. 2. This invention can reduce the use of redundant backups for hardware devices, no longer relying on the absolute reliability of hardware. Instead, through algorithmic logic, the system can automatically shift the decision focus to healthier nodes when the performance of some node-perceived module components deteriorates. It provides stable and continuous early warning functions, and the generated device health score provides accurate data for the operation and maintenance team, transforming maintenance work from fault repair to predictive and proactive maintenance. Operation and maintenance personnel can quickly locate degraded devices and perform maintenance in advance, thereby improving device uptime and reducing overall operation and maintenance costs. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of a highway entrance / exit auxiliary early warning system module based on the Internet, according to Embodiment 1 of the present invention. Figure 2 This is a schematic diagram of the workflow of the equipment health assessment unit in Embodiment 1 of the present invention; Figure 3 This is a schematic diagram of the working process of the dynamic fault-tolerant compensation unit in Embodiment 1 of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Example 1 See Figure 1 This embodiment provides an Internet-based high-speed entrance / exit auxiliary early warning system, including the following modules: The node perception module is deployed at key locations at highway entrances and exits to collect traffic target data. The data fusion and processing module is connected to the node sensing module via the Internet and is used to receive and process traffic target data; The early warning information release module is connected to the data fusion and processing module and is used to release early warning information.

[0020] Preferably, the node perception module includes a radar-visual fusion unit, which includes a millimeter-wave radar and a camera; and a data fusion processing module, which is connected to the node perception module via the Internet, for receiving and processing traffic target data.

[0021] In the above solution, the core of the node perception module adopts a radar-vision fusion unit, which includes millimeter-wave radar and camera. However, it is not a simple stacking of the two. It is a deeply integrated intelligent perception terminal. The millimeter-wave radar can accurately measure the distance and radial speed of the target vehicle, including the speed of approaching or moving away, as well as the azimuth angle, by emitting and receiving electromagnetic waves. It works with the camera to provide information such as vehicle color, model and license plate, as well as the vehicle's dynamic behavior, such as whether it crosses the line and whether the hazard lights are on.

[0022] The radar-visual fusion unit fuses the raw data from millimeter-wave radar and cameras at the front end, such as through... ARM+NPUThe processing first aligns the collected data with the time information acquired during the acquisition process. Using a built-in coordinate mapping system, the targets (pixel coordinates) identified in the video are precisely projected onto the radar's coordinate system (real-world coordinates), ensuring spatial and temporal alignment. The system then associates each point cloud target detected by the radar with each image target identified in the video. For example, if the radar detects an object moving at 60 km / h 50 meters to its right and simultaneously a red car is seen 50 meters to its right in the video, the system will identify them as the same target and generate a unique, continuous trajectory ID. When outputting the data, the fused output is no longer the original video stream and radar point cloud, but processed target vector data. Each data point contains structured information such as target ID, type settings (e.g., vehicle), latitude and longitude coordinates, speed, and heading angle. This data output method reduces the processing pressure on the backend data center and lowers network bandwidth requirements.

[0023] Preferably, the data fusion processing module includes a device health assessment unit and a dynamic fault tolerance compensation unit; wherein the device health assessment unit is used to calculate the health score of each node perception module in real time. The device health assessment unit calculates the health score by analyzing the logical characteristics of the monitoring parameters, wherein the monitoring parameters include the average contrast and clarity of the camera video stream, the density stability of the radar point cloud data, and the success rate of radar target trajectory matching.

[0024] In the above solution, a device health assessment unit is set up to continuously assess the health of the devices used by the node sensing module, thereby continuously verifying the node sensing module and ensuring the accuracy and reliability of early warning information. This device health assessment unit differs from existing device heartbeat detection or simple threshold alarms, upgrading the node sensing module from a simple online / offline judgment to a refined assessment of its operational quality. By continuously analyzing the health status indicators of the device output data, it predicts the performance degradation or failure risk of the devices used by the node sensing module. The purpose is that when the data quality of a node sensing module deteriorates due to severe weather such as rain and fog, or its own malfunctions such as lens damage and radar offset, the system can automatically identify and reduce its reliance on that node, avoiding false alarms or missed alarms, thereby improving the reliability of the entire system's monitoring data output.

[0025] like Figure 2 As shown, the device health assessment unit mainly quantifies device health through video stream analysis from cameras, radar point cloud data analysis, and success rate analysis of radar target trajectory matching, as detailed below.

[0026] Camera video stream analysis includes average contrast analysis, which is achieved by calculating the dispersion of pixel grayscale distribution in video frames. Under normal lighting conditions, the contrast of a scene should be maintained within a reasonable range. If the average contrast is consistently low, it may indicate that there are stains, fog, or insufficient lighting on the lens, resulting in loss of image details and difficulty in identifying the target. It also includes sharpness analysis, which is usually quantified by image gradient analysis such as the Laplacian operator. This operation can highlight the edges and textures in the image. The sharpness score is the statistical value of all pixels in the image after gradient calculation, such as variance. A decrease in score means that the image has become blurry, which may be caused by lens inaccuracy, raindrops, or sensor failure. By calculating these indicators in real time, this module can effectively identify image quality degradation caused by weather or the equipment itself.

[0027] Radar point cloud data analysis: Millimeter-wave radar is not affected by light, making the stability of its point cloud data even more important. In point cloud density stability analysis, within a fixed monitoring area, the point cloud density (number of points per unit area / volume) of a normally operating radar will fluctuate within a reasonable range. The evaluation module will statistically analyze the density baseline under historical normal conditions. If the point cloud density drops sharply (possibly due to signal attenuation caused by icing or severe contamination on the radar surface) or surges abnormally (possibly due to clutter interference from fixed objects such as tunnel walls), it indicates that the radar is not operating properly. By calculating the deviation between the real-time density and the baseline density, the health status of the radar's sensing capability can be quantified.

[0028] Rayvision target trajectory matching success rate analysis: It does not directly check the sensor itself, but judges the health status by examining the results of the collaboration between the two. For example, in the matching logic analysis, under the premise that the monitoring scene data and time nodes are aligned during continuous operation, the system will attempt to associate the target trajectory detected by the radar with the target trajectory identified by the video. Ideally, the same physical target should be assigned the same ID. Success rate calculation analysis: This parameter is the proportion of the number of successfully matched targets per unit time to the total number of detected targets. In other words, a decrease in the matching rate is a strong signal of abnormal sensor status. For example, if the camera is partially blocked, the video will have a blind spot, and some targets already detected by the radar will not be seen, resulting in a decrease in the matching rate. Or, if the radar is offset, the position of the detected target will have a systematic deviation from the actual position in the video, resulting in association failure. Both of these will lead to a decrease in the matching rate.

[0029] The above analysis generates multiple raw indicators. The core of the equipment health assessment unit lies in how to integrate these indicators into an intuitive, quantifiable health score. H-scoreThe typical range is 0-1. First, indicators with different dimensions (such as contrast value and matching rate) are mapped to the range of 0-1 through an algorithm to make them comparable, and the indicators are normalized. Second, not all indicators are equally important. For example, in dense fog, radar data is more reliable than camera data. In this case, the weight of radar-related indicators should be higher. The module will use a weighted fusion model (such as the analytic hierarchy process to determine the weights) to combine all normalized indicators, calculate the final health score, and perform weight allocation and fusion. At the same time, the normal range of indicators is not static. A machine learning mechanism is introduced to dynamically adjust the judgment threshold based on historical data and real-time environmental changes, making the assessment more accurate and adaptive, and realizing dynamic thresholds and learning.

[0030] Based on the above scheme, this embodiment further refines the health score. H-score The formula system is broken down, such as by normalizing the indicators to eliminate differences in the dimensions and value ranges of different monitoring parameters, so that they can be compared and calculated under the same standard. The formula is as follows:

[0031] in, This represents the original value of the i-th monitoring parameter. For example, if the camera resolution (gradient variance) is 1500, the radar-visual matching rate may be 0.85 (85%). and This represents the minimum and maximum values ​​of the i-th parameter within the defined effective working range. These values ​​are not theoretical extremes, but reasonable range boundaries determined based on engineering experience or historical statistical data. For example, a resolution below 500 may be considered completely ineffective, while a resolution above 3000 is considered excellent. This represents the normalized index value, whose theoretical range is compressed to [0, 1]. The closer it is to 1, the better the state of that single index. For example, the reasonable range for camera resolution is [500, 3000], and the current measured value is 1500. =0.4 indicates that, from the perspective of clarity alone, the current node's perception module device status is at a lower-middle level within a reasonable range.

[0032] Another example is weight allocation and fusion, which involves comprehensive scoring. Based on the importance of different indicators and the current environment, appropriate weights are assigned to each normalized indicator, and then a total score is synthesized. The formula is as follows: Linear weighted fusion and satisfy .

[0033] in This indicates the total number of monitoring indicators, such as contrast, sharpness, point cloud density stability, and matching success rate. This represents the dynamic weight assigned to the m-th indicator; This is the preliminary calculated baseline health score; It's not fixed, but dynamically adjusted based on the environmental context and the logical relationships between indicators. For example, in dense fog (which can be determined by independent meteorological sensors or image features), the overall reliability of camera indicators decreases, and the system will automatically reduce [the threshold]. and The weight, while increasing and The weights are adjusted to achieve scene adaptation; and if the success rate of matching is high... The abnormally low values, while the basic indicators (clarity, density) of the radar and camera are all normal, suggest that there may be unforeseen complex interference (such as strong radio interference or special optical reflections). The system can temporarily adjust the weights accordingly and reduce the overall credibility of all indicators.

[0034] Dynamic thresholding and learning are performed, which is the final correction of the baseline health score, to the baseline score. Fine-tuning is performed based on historical performance and the current environment to obtain the final health score used for output. H-score, By learning functions To achieve this, the specific formula is as follows:

[0035] in This represents a vector of environmental factors, including time (day / night), weather (sunny, rainy, foggy), and traffic flow, etc. This represents the sequence data indicating the historical health score of the node; for example, the system learns how each node performs in a specific environment. Historical average performance If the current A reading significantly below the historical baseline for this environment may indicate a new degradation not included in the baseline indicators, and therefore... The learning module performs downward corrections to calibrate the baseline; however, if the health level shows a continuous downward trend (even if the absolute value is still acceptable), it applies a "trend penalty." Slightly lower than the current This is to reflect the risk that its performance may continue to deteriorate, in order to avoid Due to sharp fluctuations caused by a single data jitter, smoothing techniques such as moving average and first-order hysteresis filtering are typically used.

[0036] in As a smoothing factor, This indicates the current moment. Ultimately, through the above three steps, the system outputs a stable and effective health score, H-score.

[0037] Furthermore, in this solution, the data fusion processing module also includes a dynamic fault-tolerant compensation unit. For example... Figure 3 As shown, the dynamic fault-tolerant compensation unit is used to dynamically adjust the weights of each node's perception module in the fusion decision based on the health score, and to perform data compensation inference for monitoring blind spots caused by nodes with substandard data quality. Specifically, the dynamic fault-tolerant compensation unit dynamically allocates the weights of each node's perception module in the data fusion process based on its real-time health score. The dynamic allocation of weights includes the following methods: the higher the health score, the greater the weight; when the health score is lower than the preset fault threshold, the traffic target data of the node sensed by the node sense module is isolated.

[0038] Weight The calculation formula is:

[0039]

[0040] in This represents the weight of the i-th sensing node. This represents the health score of the i-th sensing node. Scaling factor The total number of nodes participating in the fusion; weight Dynamic weight allocation based on health scores is implemented, tilting data trust towards node perception modules with high health scores to maintain the overall reliability of the system's perception data.

[0041] In the above scheme, the dynamic weight allocation formula based on the health score H-score takes the real-time health score of each sensing node i as input. This score, derived from the H-score mentioned above, combines factors such as camera clarity, radar stability, and matching success rate. It is a continuous value between 0 and 1. It's an exponential operation. The characteristic of exponential functions is that they can non-linearly amplify the differences in input values, even between two nodes. The difference is not significant; after exponential amplification, its corresponding... The differences will become very significant, and the scaling factor k controls the magnitude of the magnification; The exponent values ​​of all participating nodes are summed, which gives the weight of each node. It is calculated as the proportion of its index value to the total index value. Ultimately, all weights The condition and satisfying 1 form a standard probability distribution, which means that the system's trust level is 100% allocated, but the allocation ratio is entirely determined by the health level; where This refers to the health score. The scaling factor k is an adjustable parameter greater than 0. When k is large (e.g., k=10), the formula is extremely sensitive to differences in health scores. A high-scoring node (H=0.95) will receive overwhelming weight, while the weight of a medium-to-low-scoring node (H=0.60) will be almost suppressed to 0. This is suitable for scenarios with extremely high security requirements where the optimal node must be firmly trusted. When k is small (e.g., k=2), the weight distribution is relatively mild, allowing nodes with slightly lower performance to retain some influence. This helps maintain the smoothness of system decisions when the performance of some nodes fluctuates generally. The summation range of the denominator in the above formula dynamically changes to represent the total number of nodes participating in the fusion, ensuring the real-time inclusiveness of the system. When a node is completely isolated due to its health level falling below the fault threshold, it will be excluded from the summation term, and the system will automatically redistribute trust weights among the remaining reliable nodes.

[0042] Based on the above scheme, the equipment health assessment unit first calculates the basic health score of each node; then, it performs learning correction, using a learning function combined with the environment (dense fog, night) and historical trends to fine-tune the basic score, resulting in a more robust and intelligent health score; next, it performs dynamic weighting, substituting the latest health score into the weight calculation formula to calculate the weight of each node in data fusion in real time; finally, it performs fusion and compensation, using weights to perform weighted fusion of target detection data (such as vehicle position and speed) of each node to make a final decision. At the same time, for blind spots formed by downweighted or isolated nodes, it initiates data compensation inference based on graph neural networks and other models, using spatiotemporal correlation to infer the state of the blind spots and maintain the continuity of panoramic perception.

[0043] The above solution differs from the traditional hard, sudden fault tolerance of primary-backup switching, where the backup node takes over completely, resulting in switching delays and risks. When the performance of a node begins to decline, its weight will continuously and smoothly decrease as its health score decreases. The responsibility it bears is seamlessly and gradually shared by other healthier nodes, preventing a precipitous drop in the overall system performance. At the same time, by using a design that tilts data trust towards high health scores, it ensures that at the most critical moment, the system's decisions are always dominated by the most reliable information at that time. This allows the overall functionality to be adaptively maintained even if some node perception modules are damaged, meeting the emergency handling needs when local monitoring blind spots suddenly appear.

[0044] Furthermore, the dynamic fault-tolerant compensation unit performs data compensation inference for the monitoring blind zone. Utilizing the traffic target data of the nodes sensed by the node sensing modules adjacent to the blind zone, it extrapolates and predicts the traffic state within the blind zone through a pre-trained spatiotemporal correlation model, achieving dynamic fault-tolerant compensation for the traffic target data. The monitoring blind zone is generated by the damage or failure of the node sensing modules in the corresponding area. The spatiotemporal correlation model is a graph neural network that takes the road network topology and real-time traffic flow parameters as input to predict vehicle speed and density information within the blind zone. The early warning information release module releases differentiated early warning information to drivers through roadside variable message signs, broadcasts, and vehicle-to-everything (V2X) communication. It also sends abnormal early warnings of the node sensing modules to the early warning system control center, which is configured as a control center server or main control computer capable of receiving and processing abnormal states and making decisions.

[0045] Based on the above scheme, in this embodiment, after the system determines that a node's sensing module has failed or its performance has severely deteriorated, forming a monitoring blind zone, the dynamic fault-tolerant compensation unit will immediately initiate the data compensation inference process. The identification and definition of the blind zone involves the system automatically marking the physical area covered by that node as a monitoring blind zone when the health score of a node calculated by the device health assessment unit is consistently lower than a preset fault threshold, or when the node's communication is completely interrupted. The system will combine a high-precision map to overlay and analyze the node's theoretical monitoring range with the effective monitoring range of adjacent healthy nodes, accurately defining the geometric boundary of the blind zone. This ensures that the target area for compensation inference is clearly defined.

[0046] Secondly, the spatiotemporal correlation model, namely the graph neural network, is the key to achieving high-precision compensation, abstracting the highway network into a graph structure. ;node That is, the position of each healthy sensing node is regarded as a node in the graph; edges That is, edges are established based on the physical connections of the road network (such as road connectivity) to represent the spatial relationships between nodes; adjacency matrix This involves quantifying the tightness of connections between nodes, for example, assigning values ​​based on distance or traffic flow correlation.

[0047] By setting node features X, each healthy node uploads its monitored traffic flow parameters in real time, such as average vehicle speed, traffic density, and time occupancy. These data constitute the feature vector of the corresponding node. The trained graph neural network model learns the propagation rules of traffic flow on the road network through a message passing mechanism. For blind spot nodes, whose feature values ​​are empty or invalid, the graph neural network model aggregates the feature information of all its neighboring healthy nodes and combines it with the road network topology to deduce the traffic state characteristics that should exist in the blind spot, such as vehicle speed and density. Using the data from healthy nodes, the missing information is filled in by the learned spatiotemporal rules. Dynamic fault tolerance compensation is achieved, reducing hardware redundancy. The system no longer relies on expensive backup hardware but uses algorithms and data for intelligent compensation. Even when some nodes fail, the system can still maintain its ability to perceive road conditions. The fault tolerance strategy based on software algorithms is more flexible than traditional hardware redundancy methods.

[0048] In the above scheme, the collaborative strategy of the early warning information release module and the results of data compensation inference ultimately need to be transformed into actual actions through the early warning information release module, such as roadside variable message signs, which are suitable for widely publicized regional early warnings, such as reminding vehicles behind to slow down in advance when there is congestion at the ramp 1 kilometer ahead; with the support of vehicle-to-everything (V2X) communication or broadcasting, differentiated early warning information is pushed; at the same time, for abnormal early warnings for system operation and maintenance, abnormal early warnings of node perception modules are sent to the early warning system control center. When the health score of the node perception module is lower than the threshold, a blind spot in the monitoring area is formed, and information such as the faulty equipment ID, location, historical curve of health score, and preliminary judgment of the fault type, such as camera dirt, radar inaccuracy, etc., are sent to facilitate rapid location and early repair by operation and maintenance, and ensure the overall online rate of the system. The V2X can specifically refer to vehicle-to-infrastructure wireless information interaction, i.e., C-V2X.

[0049] Meanwhile, in this solution, the data fusion processing module also accesses vehicle GPS data and mobile phone signaling data via the Internet as an auxiliary verification source for the data compensation inference results; it realizes the fusion of vehicle GPS data and mobile phone signaling data with the data acquired by the node perception module, verifies the accuracy of the source data, corrects the error of a single data source, and obtains the traffic status under real conditions.

[0050] Preferably, the data fusion processing module further includes a data credibility assessment unit, which is used to evaluate and verify the introduced vehicle GPS data and mobile phone signaling data. Based on data indicators, the credibility weights of vehicle GPS data and mobile phone signaling data are dynamically calculated to verify conflicts between multiple data sources and ensure data accuracy and security. The data indicators include data source reliability, time correspondence, and spatial consistency.

[0051] Specifically, the data credibility assessment unit is capable of performing the following operations: Using vehicle GPS and mobile phone signaling data as auxiliary verification sources, and dynamically merging them through credibility assessment, the vehicle GPS data is used in conjunction with the vehicle backend hardware or navigation software and includes anonymous vehicle ID, accurate latitude and longitude coordinates, instantaneous speed, direction and accurate timestamp, as well as signaling data from the traffic operator that includes anonymous user ID, timestamp and connection base station location.

[0052] Selective use of data includes data cleaning and removal of obvious outliers, such as coordinate drift points caused by signal loss.

[0053] For location matching, GPS data needs to be matched to specific road networks using algorithms to determine which road segment the vehicle is traveling on.

[0054] Ensure the reliability of data sources and establish reputation profiles for different data sources. For example, data from high-precision GPS devices of professional vehicle fleets will have a higher initial credibility weight than data reported by ordinary consumer mobile apps.

[0055] The system controls time compatibility by checking whether the timestamps of external data are synchronized with the system's internal time and whether there are significant delays. It also analyzes whether there are significant discrepancies in traffic conditions, such as average vehicle speed, reflected by data from different sources within a specific time period. For example, if most GPS data indicates a road segment is clear, but some signaling data shows congestion, the latter's time compatibility score will be lowered.

[0056] Adjusting spatial consistency involves cross-referencing the monitoring results of traffic parameters described by external data, such as vehicle speed and density, with those of the most reliable node perception modules in the system, such as the radar-visual fusion unit with a high health score, at the same time and location. The data credibility assessment unit can draw on the methods of evidence theory (such as Dempster-Shafer theory) to treat each data source as an evidence body. Based on its degree of consistency with the main sensor data, the confidence level of its support for a certain traffic state (such as smooth flow, slow flow, congestion) is calculated. The higher the spatial consistency between the two, the higher the credibility weight of the external data.

[0057] When multiple data sources conflict, a data source with high credibility is selected. For monitoring blind spots caused by the failure of node perception modules, the spatiotemporal correlation model - graph neural network is used for inference. At this time, the cleaned and verified vehicle GPS and mobile phone signaling data are used as real sampling points in the adjacent areas of the blind spot to provide key inputs for the inference model and correct the inference results, thereby filling the perception gaps and obtaining a more realistic global traffic state.

[0058] Example 2 This embodiment provides an internet-based auxiliary early warning method for highway entrances and exits, with the following specific steps: S1. Use of the node perception module: The node perception module is used to collect traffic target data from different sources, and video data and radar point cloud data are collected synchronously at a fixed frequency and uploaded to the data fusion processing module in real time via the Internet. Before transmission, the data is timestamped and initially filtered to remove invalid data caused by GPS drift or equipment failure.

[0059] S2. Determine the health assessment of the data collected by the node perception module: The device health assessment unit of the data fusion processing module analyzes the data quality uploaded by each node perception module in real time. By calculating the average contrast and clarity of the video stream, the stability of the radar point cloud density, and comparing the matching rate of radar vision data for the same target trajectory recognition, a quantitative health score is generated.

[0060] S3. Dynamic Weight Adjustment and Data Compensation: The dynamic fault-tolerant compensation unit dynamically allocates the weights of the nodes sensed by each node perception module in the fusion decision based on the latest health score. The weights of the node perception modules with high health scores are increased, while the data of the node perception modules with scores below the threshold are temporarily isolated. At the same time, for the monitoring blind spots caused by the failure of node perception modules in local areas, the dynamic fault-tolerant compensation unit calls the pre-trained spatiotemporal correlation model, combines the real-time data of adjacent healthy nodes with historical traffic flow patterns, and performs short-term traffic state inference to compensate for the missing information.

[0061] S4. Warning Command Generation and Warning Information Release: The warning information release module uses the merged traffic data to determine the current risk level and generate differentiated warning commands for sending and issuing alerts.

[0062] The above description is merely a preferred embodiment of the present invention. It should be understood that the present invention is not limited to the forms disclosed herein and should not be construed as excluding other embodiments. It can be used in various other combinations, modifications, and environments, and can be altered within the scope of the concept described herein through the above teachings or related technologies or knowledge. Modifications and variations made by those skilled in the art that do not depart from the spirit and scope of the present invention should be within the protection scope of the appended claims.

[0063] It should be noted that, for the sake of simplicity, the foregoing method embodiments are described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

Claims

1. A highway entrance / exit auxiliary early warning system based on the Internet, characterized in that, Includes the following modules: The node perception module is deployed at key locations at highway entrances and exits to collect traffic target data. The data fusion processing module is connected to the node perception module via the Internet to receive and process traffic target data. The data fusion processing module includes an equipment health assessment unit and a dynamic fault tolerance compensation unit. The equipment health assessment unit is used to calculate the health score of each node perception module in real time. The dynamic fault tolerance compensation unit is used to dynamically adjust the weight of each node perception module in the fusion decision based on the health score, and to perform data compensation inference for monitoring blind spots formed by nodes with substandard data quality. The early warning information release module is connected to the data fusion and processing module and is used to release early warning information.

2. The Internet-based highway entrance / exit auxiliary early warning system according to claim 1, characterized in that, The node perception module includes a radar-visual fusion unit, which includes a millimeter-wave radar and a camera. The device health assessment unit calculates a health score by analyzing the logical characteristics of monitoring parameters, including the average contrast and clarity of the camera's video stream, the density stability of radar point cloud data, and the success rate of radar target trajectory matching.

3. The Internet-based highway entrance / exit auxiliary early warning system according to claim 2, characterized in that, The dynamic fault-tolerant compensation unit dynamically allocates the weights of each node perception module in data fusion based on the real-time health scores of the node perception modules. ; The dynamic allocation methods include: the higher the health score, the greater the weight; when the health score is lower than the preset fault threshold, the traffic target data of the node sensed by the node's sensing module is isolated.

4. The Internet-based high-speed entrance / exit auxiliary early warning system according to claim 3, characterized in that, The weight The calculation formula is: in This represents the weight of the i-th sensing node. This represents the health score of the i-th sensing node. Scaling factor The total number of nodes participating in the fusion; weight Dynamic weight allocation based on health scores is implemented, tilting data trust towards node perception modules with high health scores to maintain the overall reliability of the system's perception data.

5. The Internet-based highway entrance / exit auxiliary early warning system according to claim 1, characterized in that, The dynamic fault-tolerant compensation unit performs data compensation inference for the monitoring blind zone. It uses the traffic target data of the nodes sensed by the node sensing modules adjacent to the blind zone and infers and predicts the traffic state in the blind zone through a pre-trained spatiotemporal correlation model to achieve dynamic fault-tolerant compensation for the traffic target data. The monitoring blind zone is caused by the damage or failure of the node sensing modules in the corresponding area.

6. The Internet-based high-speed entrance / exit auxiliary early warning system according to claim 5, characterized in that, The spatiotemporal correlation model is a graph neural network that takes the road network topology and real-time traffic flow parameters as input to predict vehicle speed and density information in blind spots.

7. The Internet-based high-speed entrance / exit auxiliary early warning system according to claim 6, characterized in that, The data fusion processing module also accesses vehicle GPS data and mobile phone signaling data via the Internet as an auxiliary verification source for the data compensation inference results; it realizes the fusion of vehicle GPS data and mobile phone signaling data with the data acquired by the node perception module, verifies the accuracy of the source data, corrects the error of a single data source, and obtains the traffic status under real conditions.

8. The Internet-based highway entrance / exit auxiliary early warning system according to claim 7, characterized in that, The data fusion processing module also includes a data credibility assessment unit, which is used to evaluate and verify the introduced vehicle GPS data and mobile phone signaling data. Based on data indicators, it dynamically calculates the credibility weights of vehicle GPS data and mobile phone signaling data, verifies conflicts between multi-source data, and ensures data accuracy and security. The data indicators include data source reliability, time correspondence, and spatial consistency.

9. The Internet-based highway entrance / exit auxiliary early warning system according to claim 1, characterized in that, The warning information dissemination module disseminates differentiated warning information to drivers through roadside variable message signs, broadcasts, and vehicle-to-everything (V2X) communication; it also sends abnormal warnings of the node perception module to the warning system control center.

10. A method for auxiliary early warning at highway entrances and exits based on the Internet, which is applied to the Internet-based auxiliary early warning system for highway entrances and exits as described in any one of claims 1-9, characterized in that, The method includes the following steps: S1. Use of the node perception module: The node perception module is used to collect traffic target data from different sources, synchronously collect video data and radar point cloud data at a fixed frequency, and upload them to the data fusion processing module in real time via the Internet; before transmission, the data is timestamped and preliminarily filtered to remove invalid data caused by GPS drift or equipment failure. S2. Determine the health assessment of the data collected by the node perception module: The device health assessment unit of the data fusion processing module analyzes the data quality uploaded by each node perception module in real time. By calculating the average contrast and clarity of the video stream, the stability of the radar point cloud density, and comparing the matching rate of radar vision data for the same target trajectory recognition, a quantitative health score is generated. S3. Dynamic Weight Adjustment and Data Compensation: The dynamic fault-tolerant compensation unit dynamically allocates the weights of the nodes sensed by each node perception module in the fusion decision based on the latest health score. The weights of the node perception modules with high health scores are increased, while the data of the node perception modules with scores below the threshold are temporarily isolated. At the same time, for the monitoring blind spots caused by the failure of node perception modules in local areas, the dynamic fault-tolerant compensation unit calls the pre-trained spatiotemporal correlation model, combines the real-time data of adjacent healthy nodes with historical traffic flow patterns, and performs short-term traffic state inference to compensate for the missing information. S4. Warning Command Generation and Warning Information Release: The warning information release module uses the merged traffic data to determine the current risk level and generate differentiated warning commands for sending and issuing alerts.

Images