Expressway sudden danger warning transmission system

Abstract

The invention discloses an expressway sudden danger warning transmission system, which relates to the technical field of expressway warning, and comprises a danger information acquisition module, a data processing center, an information transmission network, a vehicle-mounted receiving terminal and an emergency linkage unit, and the modules and the units realize data interaction and cooperative work through a preset communication protocol. According to the highway sudden danger warning transmission system, all-directional monitoring of road conditions, vehicle dynamic states and the like is realized; the data processing center adopts cloud computing and artificial intelligence technologies to accurately identify dangerous events through a multi-modal fusion model; the information transmission network is combined with edge computing and a 5G private network, so that high-efficiency and low-delay transmission of data is guaranteed; the vehicle-mounted receiving terminal provides multi-stage early warning; and the emergency linkage unit realizes co-processing with an emergency department. According to the system, the timeliness and accuracy of danger warning can be remarkably improved, the traffic accident rate is reduced, and a powerful guarantee is provided for safe operation of an expressway.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of highway warning technology, in particular to a highway sudden danger warning transmission system. BACKGROUND

[0002] With the continuous increase of highway mileage and the sustained growth of motor vehicle ownership in China, the traffic flow of highways is increasing, and the frequency of sudden danger events (such as traffic accidents, vehicle breakdowns, road collapses, and severe weather) is also rising. If these sudden danger events cannot be warned and disposed in time, they are likely to cause secondary accidents, resulting in serious casualties and property losses.

[0003] Currently, the danger warning of highways mainly relies on manual patrol, fixed monitoring cameras, and driver alarm. Manual patrol has limited coverage, slow response speed, high labor cost, and other problems, making it difficult to achieve real-time monitoring of the entire road section. Although fixed monitoring cameras can achieve real-time monitoring of some road sections, they have monitoring blind spots due to the installation location and number of cameras, and poor adaptability to severe weather (such as heavy rain, heavy fog, and night), which may result in missed or false alarms. Driver alarm relies on the initiative of drivers, and the timeliness and accuracy of the alarm information cannot be guaranteed, and in emergency situations, drivers may not have time to report.

[0004] Some existing highway danger warning systems often use a single information collection method, such as relying only on cameras or sensors for monitoring, resulting in insufficient comprehensiveness and reliability of information collection. At the same time, these systems have limited data processing capabilities, making it difficult to quickly analyze and process massive amounts of monitoring data, and unable to identify danger events and issue warnings in a timely manner. In terms of information transmission, traditional wireless networks are often used, which have large transmission delay and unstable bandwidth, affecting the timely transmission of danger warning information. In addition, the linkage mechanism between the existing system and the emergency disposal department is not perfect, and after a danger event occurs, it is difficult to quickly notify the relevant departments for disposal, resulting in low disposal efficiency.

[0005] For example, the patent with the publication number CN108765432A "Highway Abnormal Event Monitoring and Warning System" collects information through cameras and sensors, but does not use multi-modal data fusion technology, resulting in low accuracy of danger event identification in complex scenarios. The patent with the publication number CN110232156B "Highway Danger Warning System Based on Vehicle-Road Cooperation" mainly relies on vehicle-mounted devices for information collection, which has limited coverage, and the information transmission does not use a dedicated communication network, which needs to be improved in terms of transmission reliability.

[0006] Therefore, it has important practical significance and application value to develop a highway emergency danger warning transmission system capable of realizing multi-source information fusion collection, quickly and accurately identifying dangerous events, low-latency transmission of warning information, and efficient linkage with emergency departments. SUMMARY

[0007] In view of the deficiencies of the prior art, the present application provides a highway emergency danger warning transmission system to solve the problems raised in the above background art.

[0008] To achieve the above object, the present application is implemented by the following technical scheme: a highway emergency danger warning transmission system, comprising a danger information collection module, a data processing center, an information transmission network, a vehicle-mounted receiving terminal and an emergency linkage unit, each module and unit realizes data interaction and cooperative work through a preset communication protocol;

[0009] The danger information collection module includes multi-source heterogeneous sensing devices, specifically a high-definition camera group, a millimeter wave radar array, an infrared thermal imaging sensor, a road surface state sensor, a meteorological monitoring station and a user reporting terminal.

[0010] Specifically, the high-definition camera group adopts a 360° panoramic shooting mode, monitors points are arranged at intervals, industrial-grade cameras are equipped, night vision function is provided, road surface dynamics can be clearly captured in low light environment, an intelligent analysis chip is built-in the camera group, vehicle trajectory recognition, abnormal behavior detection and dangerous target classification can be realized in real time, the identification algorithm adopts an improved YOLOv8 model, and the feature map is weighted processed by introducing an attention mechanism module; the millimeter wave radar array adopts a specific frequency band, radar base stations are arranged, can cover a certain area, has the ability to track multiple moving targets at the same time, and a constant false alarm rate detection algorithm is adopted for radar data processing; the infrared thermal imaging sensor is mainly used for detecting high-temperature objects on the road surface, can realize clear imaging of the target in a lightless environment, and the thermal image data output by the sensor is analyzed in real time through an edge computing node; the road surface state sensor is buried on both sides of the highway lane line, monitoring sections are arranged, and is used for measuring road surface temperature, humidity and ice thickness; the meteorological monitoring station is arranged along the highway, and can monitor meteorological parameters such as wind speed, wind direction, precipitation and visibility in real time; the user reporting terminal is an interactive interface integrated in a vehicle-mounted navigation system or a mobile phone APP, users can report emergency danger events in various forms, the reported information is automatically attached with related information, and an intelligent verification module is built-in the terminal.

[0011] The data processing center adopts a cloud computing architecture, and includes a data storage server, a distributed computing node, an artificial intelligence analysis platform and a database system.

[0012] Specifically, the data storage server adopts a distributed file system and has the ability to process data writing, and can store original monitoring data and analysis results for a certain period of time; the distributed computing node adopts a GPU cluster architecture and includes multiple computing nodes, which can realize parallel processing of multi-source data; the artificial intelligence analysis platform is equipped with a multi-modal fusion model based on the Transformer architecture, and the feature vectors output by each sub-module are fused through an attention mechanism; the database system adopts a combination of a relational database and a time series database;

[0013] The information transmission network adopts a hybrid architecture of "edge computing + 5G private network + optical backbone network"; the edge computing node is deployed in a communication base station or a monitoring room along the expressway and is connected to the nearby collection equipment through a wired manner, and is responsible for preprocessing the collected data; the edge computing node and the data processing center communicate through a 5G private network or an optical backbone network;

[0014] The vehicle-mounted receiving terminal is integrated into the vehicle-mounted information entertainment system or the head-up display device of the vehicle and supports the 5G-V2X communication protocol; the vehicle-mounted receiving terminal is built-in with a positioning module and a map engine, which can localize the dangerous warning information according to the current position and driving direction of the vehicle, and perform multi-level early warning through various ways when the vehicle is about to enter the dangerous area;

[0015] The emergency linkage unit is connected to the systems of the expressway management department, the traffic police department, the fire department, the medical rescue institution and other emergency disposal units, and realizes information sharing and collaborative linkage through a preset interface protocol; the emergency linkage unit can automatically send event details to the related emergency disposal units, generate preliminary disposal suggestions according to the event type and severity, and simultaneously feedback the emergency disposal progress to the data processing center in real time, forming a closed-loop management.

[0016] Optionally, the high-definition camera group sets a monitoring point every 400-600 m, and each monitoring point is equipped with 3-4 industrial-grade cameras with a resolution of 8K and starlight night vision function, which can clearly capture the road surface in an environment with an illumination intensity lower than 0.1 lux, and the specific formula of its identification algorithm is:

[0017] Attention(F)=σ(W2·ReLU(W1·F))·F

[0018] Wherein, F is an input feature map, W1 and W2 are convolutional layer weight parameters respectively, and sigma is a Sigmoid activation function. After processing by the algorithm, the abnormal behavior recognition response time is less than 200ms. The intelligent analysis chip adopts a heterogeneous computing architecture, integrating CPU, GPU and NPU cores, wherein the computing power of the NPU core reaches 20TOPS, can parallelly process multiple video stream data, and supports dynamic adjustment of computing resource allocation. When a suspected dangerous target is detected in the picture, the image processing priority of the region is automatically improved.

[0019] Optionally, the millimeter wave radar array adopts a 77GHz frequency band, and one radar base station is arranged every 800-1000m. A single base station can cover an area with a radius of 500m and has the ability to simultaneously track more than 200 moving targets. The ranging accuracy can reach ±0.1m, and the speed measurement accuracy can reach ±0.05m / s. The MIMO (Multiple Input Multiple Output) technology is adopted, including 12 transmitting antennas and 16 receiving antennas. The three-dimensional positioning (distance, azimuth angle, elevation angle) of the target is realized through digital beamforming technology. The azimuth angle measurement range is -60° to +60°, and the accuracy is ±0.5°. The elevation angle measurement range is -10° to +10°, and the accuracy is ±1°. The vehicle on the road surface and the fixed object on the roadside can be effectively distinguished.

[0020] Optionally, the infrared thermal imaging sensor detects a wavelength range of 8-14pm, a temperature measurement range of -20℃ to 150℃, and a temperature measurement accuracy of ±2℃. Clear imaging of the target is realized in a completely light-free environment. The thermal image data output by the sensor is analyzed in real time by an edge computing node. An abnormal area segmentation algorithm based on temperature gradient is adopted. When a region with a temperature exceeding a preset threshold and an area greater than 0.5㎡ is detected, a danger warning is triggered.

[0021] Optionally, the road surface state sensor is buried on both sides of the highway lane line, and one monitoring section is arranged every 2km. Each section includes three sensor units for measuring road surface temperature, humidity and ice thickness. The road surface temperature measurement range is -40℃ to 80℃, and the accuracy is ±0.5℃. The humidity measurement range is 0% to 100%RH, and the accuracy is ±3%RH. The ice thickness measurement range is 0-50mm, and the accuracy is ±1mm. The sensor adopts digital signal output and is connected to the nearby edge computing node through a wired mode. The data sampling frequency is 1Hz.

[0022] Optionally, the meteorological monitoring station is set every 5km along the expressway, which monitors the wind speed, wind direction, precipitation, and visibility meteorological parameters in real time. The wind speed measurement range is 0-60m / s, the accuracy is ±0.5m / s, the wind direction measurement range is 0-360°, the accuracy is ±5°, the precipitation measurement resolution is 0.2mm, the visibility measurement range is 10m to 10km, the accuracy is ±10%, and the meteorological data is uploaded to the data processing center in real time through the 4G / 5G wireless network.

[0023] Optionally, the user reporting terminal is an interactive interface integrated into a vehicle navigation system or a mobile phone APP. Users can report sudden dangerous events in the form of text, pictures, and videos. The reporting information automatically includes the reporting time, geographic location (positioned by GPS / Beidou, accuracy ±5m), and the reporting user identifier. The terminal has a built-in intelligent verification module that can preliminarily filter the reporting content and remove obviously false or duplicate information.

[0024] Optionally, the data storage server of the data processing center uses a distributed file system (such as HDFS) with the ability to process more than 10GB of data per second. It can store at least 3 months of raw monitoring data and analysis results. The distributed computing node uses a GPU cluster architecture, which includes more than 100 computing nodes, each node equipped with 8 NVIDIA A100 graphics cards, with a total computing capacity of more than 10 PFlops, which can realize parallel processing of multi-source data. The artificial intelligence analysis platform is equipped with a multi-modal fusion model based on the Transformer architecture. This model includes a text processing submodule, an image processing submodule, a radar data processing submodule, and a sensor data processing submodule. The feature vectors output by each submodule are fused through an attention mechanism. The specific fusion formula is:

[0025]

[0026] where F i is the feature vector of the i-th data source, α i is the corresponding attention weight, and satisfies α iThe comprehensive recognition accuracy of the model to the dangerous event can reach more than 99% and the false positive rate is less than 0.1% after the model is trained by the historical data, the model has self-learning ability, receives new dangerous event samples and disposal results in real time through an online learning module, updates the multi-modal fusion model by using an incremental training algorithm, the model parameters updated each time are synchronized to each edge computing node by using a federated learning method, and the recognition ability of the entire system is continuously improved; the database system adopts a combination of a relational database and a time series database, the relational database is used to store structured data such as system configuration information, user data and event processing records, and the time series database is used to store time series data generated by devices such as sensors, cameras and radars, and supports high-concurrency read-write and fast query.

[0027] Optionally, the edge computing node of the information transmission network is arranged in a communication base station or a monitoring room along the expressway, is connected with the nearby collection devices in a wired manner, is responsible for pre-processing (such as data cleaning, feature extraction and preliminary analysis) of the collected data, adopts an edge cloud architecture, supports containerized deployment, can dynamically expand computing resources according to actual business needs, the processing capacity of a single edge computing node can support simultaneous access of 50 high-definition video streams, 20 millimeter wave radars and 100 various sensors, and has a local data caching function, when communication with the data processing center is interrupted, the edge computing node can store key data locally, and after the communication is restored, the key data is automatically supplemented; the edge computing node and the data processing center communicate through a 5G private network or an optical fiber backbone network, the 5G private network adopts a slicing technology, allocates dedicated bandwidth resources (the uplink bandwidth is not less than 100 Mbps and the downlink bandwidth is not less than 1 Gbps) for the system, the end-to-end transmission delay is less than 10 ms, and the optical fiber backbone network adopts a redundant link design to ensure the reliability of data transmission.

[0028] Optionally, the vehicle-mounted receiving terminal is integrated into a vehicle-mounted information entertainment system or a head-up display (HUD) device of the vehicle, supports a 5G-V2X communication protocol, can receive dangerous warning information from an information transmission network, is provided with a positioning module (GPS / Beidou dual-mode) and a map engine, can process the dangerous warning information locally according to a current position and a driving direction of the vehicle, and can perform multi-level prewarning through various modes such as sound, image and vibration when the vehicle is about to enter a dangerous area, wherein the prewarning levels are divided into three levels of prompt (green), warning (yellow) and emergency (red), different levels correspond to different prewarning intensities and response measures, and deep integration with a vehicle control system is supported, so that, when emergency-level dangerous warning information is received, a control signal can be automatically sent to an electronic stability program (ESP), an adaptive cruise system (ACC) and the like of the vehicle to suggest the vehicle to slow down, maintain a distance or perform a lane deviation prewarning, and a specific control strategy is dynamically adjusted according to a dangerous type and a current state of the vehicle; the emergency linkage unit is provided with an event level evaluation model, the model is based on parameters such as a type (such as a traffic accident, a road collapse, a vehicle self-ignition and the like) of a dangerous event, an influence range, a duration and a potential risk, an analytic hierarchy process (AHP) is used to calculate a comprehensive level of the event, the event is divided into four levels of general (level I), relatively large (level II), major (level III) and particularly major (level IV), different levels correspond to different emergency response plans and linkage units, and the system further includes a self-checking and fault-tolerant module, the module periodically detects states of various collection devices, a transmission network, a data processing center and terminal devices, identifies fault points and performs hierarchical alarming through modes such as sending a diagnosis data packet, monitoring device operating parameters (such as temperature, voltage, operating duration and the like) and checking data integrity, automatically starts a backup link or a recovery program for slight faults (such as temporary communication interruption of a sensor) that can be self-recovered, and timely informs maintenance personnel to handle serious faults, so that the availability of the system is ensured to be above 99.99%.

[0029] The present application provides a highway sudden danger warning transmission system, which has the following beneficial effects:

[0030] The highway sudden danger warning transmission system adopts a multi-source heterogeneous information collection mode, integrates a high-definition camera group, a millimeter wave radar array, an infrared thermal imaging sensor, a road surface state sensor, a meteorological monitoring station and a user reporting terminal and the like, realizes omnibearing and multidimensional monitoring of highway sudden danger events, the multi-source information collection mode can make up for the shortcomings of a single device, the high-definition camera can clearly identify vehicle and road surface conditions in good light, but its performance decreases in bad weather, the millimeter wave radar and the infrared thermal imaging sensor are not affected by weather and can stably work in rainy, snowy, foggy and night environments, the devices complement each other, significantly improve the comprehensiveness and reliability of danger information collection, and effectively reduce the occurrence of monitoring blind spots and missed reporting situations;

[0031] Secondly, the data processing center of the system adopts a cloud computing architecture and a multi-modal fusion model based on Transformer, which has strong data processing and analysis capabilities. The multi-modal fusion model can deeply fuse and intelligently analyze information from different devices, and through the attention mechanism, it can weight process the features of different data sources, improving the accuracy of dangerous event identification. At the same time, the model has self-learning ability and can continuously optimize through online learning, adapting to various complex scenarios, further improving the identification accuracy and response speed of the system to dangerous events, effectively reducing the false alarm rate, and providing reliable decision basis for subsequent warning and disposal.

[0032] Thirdly, the information transmission network adopts a hybrid architecture of "edge computing + 5G private network + optical fiber backbone network", realizing low-latency and high-reliability transmission of dangerous information. The edge computing nodes preprocess the collected data, reducing the amount of data uploaded to the data processing center and reducing network transmission pressure. The 5G private network allocates dedicated bandwidth to the system through slicing technology, ensuring the real-time and stability of data transmission, and the end-to-end transmission delay is less than 10ms, meeting the demand of fast transmission of dangerous warning information. The optical fiber backbone network provides high-bandwidth guarantee for the transmission of massive data, and adopts redundant link design to improve the reliability of the network, avoiding the interruption of information transmission caused by network failure.

[0033] In addition, the vehicle-mounted receiving terminal can provide timely and accurate dangerous warning for the driver through a multi-level early warning mechanism according to the distance and driving direction of the vehicle from the dangerous area. The multi-level early warning (prompt, warning, emergency) adopts different warning methods and intensities, which can attract the attention of the driver without causing excessive interference to the driver, helping the driver to take correct measures in time and reducing the incidence of traffic accidents.

[0034] Finally, the emergency linkage unit is used to realize efficient cooperation with various emergency disposal departments. After a dangerous event occurs, the event information can be quickly transmitted to the relevant departments, and preliminary disposal suggestions can be provided, shortening the emergency response time and improving the efficiency of emergency disposal. At the same time, through the closed-loop management mechanism, the disposal progress is fed back in real time, ensuring that the dangerous event is handled in time and effectively, further ensuring the safety and smoothness of the expressway. DETAILED DESCRIPTION

[0035] The technical solutions in the embodiments of the present application will be described clearly and completely below. Obviously, the described embodiments are only part of the embodiments of the present application, not all.

[0036] The application discloses a highway emergency danger warning transmission system, which comprises a danger information collection module, a data processing center, an information transmission network, a vehicle-mounted receiving terminal and an emergency linkage unit, and each module and unit realizes data interaction and cooperative work through a preset communication protocol.

[0037] The danger information collection module comprises multiple source heterogeneous sensing devices, specifically, a high-definition camera group, a millimeter wave radar array, an infrared thermal imaging sensor, a road surface state sensor, a meteorological monitoring station and a user reporting terminal.

[0038] Specifically, the high-definition camera group adopts a 360-degree panoramic shooting mode, the monitoring points are arranged at intervals, industrial-grade cameras are equipped, night vision function is provided, the road surface dynamic can be clearly captured in a low-light environment, an intelligent analysis chip is arranged in the camera group, vehicle trajectory recognition, abnormal behavior detection and dangerous target classification can be realized in real time, the identification algorithm adopts an improved YOLOv8 model, and a feature map is weighted processed through the introduction of an attention mechanism module; the millimeter wave radar array adopts a specific frequency band, radar base stations are arranged, a certain area can be covered, the millimeter wave radar array has the ability to simultaneously track multiple moving targets, and a constant false alarm rate detection algorithm is adopted for radar data processing; the infrared thermal imaging sensor is mainly used for detecting high-temperature objects on the road surface, clear imaging of the target can be realized in a lightless environment, and thermal image data output by the sensor is analyzed in real time through an edge computing node; the road surface state sensor is buried on both sides of the highway lane line, monitoring sections are arranged, and the road surface temperature, humidity and ice thickness are measured; the meteorological monitoring station is arranged along the highway, and meteorological parameters such as wind speed, wind direction, precipitation and visibility can be monitored in real time; the user reporting terminal is an interactive interface integrated in a vehicle-mounted navigation system or a mobile phone APP, and users can report emergency danger events in various forms, the reported information is automatically attached with related information, and an intelligent verification module is arranged in the terminal.

[0039] The high-definition camera group is provided with a monitoring point every 400-600 m, each monitoring point is provided with 3-4 industrial-grade cameras with a resolution of 8K and starlight night vision function, the road surface dynamic can be clearly captured in an environment with an illumination intensity lower than 0.1 lux, and the specific formula of the identification algorithm is as follows:

[0040] Attention (F) = sigma (W2 ReLU (W1 F)) F

[0041] Wherein, F is an input feature map, W1 and W2 are convolutional layer weight parameters respectively, and sigma is a Sigmoid activation function. After processing by the algorithm, the abnormal behavior recognition response time is less than 200 ms. The intelligent analysis chip adopts a heterogeneous computing architecture, integrating CPU, GPU and NPU cores, wherein the NPU core has a computing power of 20 TOPS, can parallelly process multiple video stream data, and supports dynamic adjustment of computing resource allocation. When a suspected dangerous target is detected in the picture, the image processing priority of the region is automatically improved;

[0042] The millimeter wave radar array adopts a 77GHz frequency band, and one radar base station is arranged every 800-1000m. A single base station can cover an area with a radius of 500m and has the ability to simultaneously track more than 200 moving targets. Its ranging accuracy can reach ±0.1m, and its speed measurement accuracy can reach ±0.05m / s. The MIMO (Multiple Input Multiple Output) technology is adopted, including 12 transmitting antennas and 16 receiving antennas. Through digital beamforming technology, three-dimensional positioning (distance, azimuth, elevation) of the target is realized. The azimuth measurement range is -60° to +60°, and the accuracy is ±0.5°. The elevation measurement range is -10° to +10°, and the accuracy is ±1°. It can effectively distinguish vehicles on the road surface from fixed objects on the roadside.

[0043] The infrared thermal imaging sensor detects a wavelength range of 8-14um, and the temperature measurement range is -20℃ to 150℃. The temperature measurement accuracy is ±2℃. It can realize clear imaging of the target in a completely lightless environment. The thermal image data output by the sensor is analyzed in real time by the edge computing node. An abnormal area segmentation algorithm based on temperature gradient is adopted. When a region with a temperature exceeding a preset threshold and an area greater than 0.5㎡ is detected, a danger warning is triggered.

[0044] The road surface state sensor is buried on both sides of the highway lane line. One monitoring section is arranged every 2km. Each section includes 3 sensor units for measuring road surface temperature, humidity and ice thickness. The road surface temperature measurement range is -40℃ to 80℃, and the accuracy is ±0.5℃. The humidity measurement range is 0% to 100% RH, and the accuracy is ±3% RH. The ice thickness measurement range is 0-50mm, and the accuracy is ±1mm. The sensor adopts digital signal output and is connected to the nearby edge computing node through a wired way. The data sampling frequency is 1Hz.

[0045] A meteorological monitoring station is set up every 5 km along the expressway to monitor wind speed, wind direction, precipitation, and visibility meteorological parameters in real time. The wind speed measurement range is 0-60 m / s, with an accuracy of ±0.5 m / s. The wind direction measurement range is 0-360°, with an accuracy of ±5°. The precipitation measurement resolution is 0.2 mm. The visibility measurement range is 10 m to 10 km, with an accuracy of ±10%. Meteorological data is uploaded to the data processing center in real time through a 4G / 5G wireless network.

[0046] The user reports the terminal as an interactive interface integrated into the vehicle navigation system or the mobile phone APP. The user can report sudden dangerous events in the form of text, pictures, and videos. The reported information automatically includes the reporting time, geographic location (positioned by GPS / Beidou, accuracy ±5 m), and the reporting user identifier. The terminal has a built-in intelligent verification module that can preliminarily filter the reported content and remove obviously false or duplicate information.

[0047] The data processing center uses a cloud computing architecture, including data storage servers, distributed computing nodes, artificial intelligence analysis platforms, and database systems.

[0048] Specifically, the data storage server uses a distributed file system and has the ability to process data writing, storing raw monitoring data and analysis results for a certain period of time. The distributed computing node uses a GPU cluster architecture, including multiple computing nodes, which can realize parallel processing of multi-source data. The artificial intelligence analysis platform is equipped with a multi-modal fusion model based on the Transformer architecture, and the feature vectors output by each sub-module are fused through an attention mechanism. The database system uses a combination of relational databases and time series databases.

[0049] The data storage server of the data processing center uses a distributed file system (such as HDFS) and has the ability to process more than 10 GB of data writing per second, storing raw monitoring data and analysis results for at least 3 months. The distributed computing node uses a GPU cluster architecture, including more than 100 computing nodes, each equipped with 8 NVIDIA A100 graphics cards, with a total computing capacity of more than 10 PFlops, capable of parallel processing of multi-source data. The artificial intelligence analysis platform is equipped with a multi-modal fusion model based on the Transformer architecture, which includes a text processing sub-module, an image processing sub-module, a radar data processing sub-module, and a sensor data processing sub-module. The feature vectors output by each sub-module are fused through an attention mechanism, and the specific fusion formula is:

[0050]

[0051] where F i is the feature vector of the i-th data source, and α i is the corresponding attention weight, and satisfies a i The comprehensive identification accuracy rate of the model to the dangerous event can reach more than 99% after the model is trained by the historical data, and the false positive rate is less than 0.1%, the model has self-learning ability, receives new dangerous event samples and disposal results in real time through the online learning module, updates the multi-modal fusion model by using the incremental training algorithm, and synchronizes the model parameters of each update to each edge computing node through the federated learning method, so as to ensure the continuous improvement of the identification ability of the whole system; the database system adopts the combination of relational database and time series database, the relational database is used to store structured data such as system configuration information, user data, event processing records, and the time series database is used to store time series data generated by sensors, cameras, radars and other devices, which supports high-concurrency read and write and fast query;

[0052] The information transmission network adopts a hybrid architecture of "edge computing + 5G private network + optical backbone network"; the edge computing nodes are deployed in the communication base stations or monitoring rooms along the expressway, connected with the nearby collection devices through wired mode, responsible for preprocessing the collected data, and the edge computing nodes and the data processing center communicate through 5G private network or optical backbone network;

[0053] The edge computing nodes of the information transmission network are deployed in the communication base stations or monitoring rooms along the expressway, connected with the nearby collection devices through wired mode, responsible for preprocessing the collected data (such as data cleaning, feature extraction, preliminary analysis, etc.), adopting edge cloud architecture, supporting containerized deployment, and dynamically expanding computing resources according to actual business needs, the processing capacity of a single edge computing node can support simultaneous access of 50 channels of high-definition video streams, 20 millimeter wave radars and 100 various sensors, and has local data caching function, when the communication with the data processing center is interrupted, it can store the key data locally, and automatically supplement the transmission after the communication is restored; the edge computing nodes and the data processing center communicate through 5G private network or optical backbone network, the 5G private network adopts slicing technology to allocate dedicated bandwidth resources (uplink bandwidth not less than 100Mbps, downlink bandwidth not less than 1Gbps) for the system, the end-to-end transmission delay is less than 10ms, and the optical backbone network adopts redundant link design to ensure the reliability of data transmission;

[0054] The vehicle-mounted receiving terminal is integrated into the vehicle-mounted information entertainment system or the head-up display device of the vehicle, and supports 5G-V2X communication protocol; the vehicle-mounted receiving terminal is built-in with a positioning module and a map engine, which can localize the dangerous warning information according to the current position and driving direction of the vehicle, and give multi-level warning through various ways when the vehicle is about to enter the dangerous area;

[0055] The vehicle-mounted receiving terminal is integrated in a vehicle-mounted information entertainment system or a head-up display (HUD) device of a vehicle, supports a 5G-V2X communication protocol, can receive dangerous warning information from an information transmission network, is provided with a positioning module (GPS / Beidou dual mode) and a map engine, can perform localized processing on the dangerous warning information according to a current position and a driving direction of the vehicle, and can perform multi-level prewarning in multiple modes such as sound, image and vibration when the vehicle is about to enter a dangerous area, the prewarning levels are divided into three levels of prompt (green), warning (yellow) and emergency (red), different levels correspond to different prewarning intensities and response measures, and deep integration with a vehicle control system is supported, when emergency-level dangerous warning information is received, control signals can be automatically sent to an electronic stability program (ESP), an adaptive cruise control system (ACC) and the like of the vehicle, to suggest the vehicle to slow down, maintain a distance or perform a lane deviation prewarning, and a specific control strategy is dynamically adjusted according to a dangerous type and a current state of the vehicle; the emergency linkage unit is provided with an event level evaluation model, the model is based on parameters such as a type (such as a traffic accident, a road collapse, a vehicle self-ignition and the like) of a dangerous event, an influence range, a duration and a potential risk, an analytic hierarchy process (AHP) is used to calculate a comprehensive level of the event, the event is divided into four levels of general (level I), relatively large (level II), major (level III) and particularly major (level IV), different levels correspond to different emergency response plans and linkage units, and the system further includes a self-checking and fault-tolerant module, the module periodically performs state detection on each collection device, a transmission network, a data processing center and a terminal device, identifies fault points and performs hierarchical alarm through the sending of a diagnosis data packet, the monitoring of device operation parameters (such as temperature, voltage, operation duration and the like) and the checking of data integrity, automatically starts a backup link or a recovery program for slight faults (such as temporary communication interruption of a sensor) that can be self-recovered, and timely informs maintenance personnel to handle serious faults, so that the availability of the system is ensured to be more than 99.99%.

[0056] The emergency linkage unit is connected with systems of emergency disposal units such as a highway management department, a traffic police department, a fire department and a medical rescue institution, information sharing and collaborative linkage are realized through a preset interface protocol, the emergency linkage unit can automatically send event details to the related emergency disposal units, generate preliminary disposal suggestions according to an event type and a severity, and simultaneously feedback emergency disposal progress to the data processing center in real time, to form a closed-loop management.

[0057] The above merely describes a preferred specific embodiment of the application, but the protection scope of the application is not limited to this, any person skilled in the art can make equivalent replacement or change according to the technical solution and the inventive concept of the application within the technical range disclosed by the application, which should be covered in the protection scope of the application.

Claims

1. A highway emergency hazard warning transmission system, characterized by, The system comprises a dangerous information collection module, a data processing center, an information transmission network, a vehicle-mounted receiving terminal and an emergency linkage unit, and each module and unit realizes data interaction and collaborative work through a preset communication protocol. The dangerous information collection module comprises multiple source heterogeneous sensing devices, specifically, a high-definition camera group, a millimeter wave radar array, an infrared thermal imaging sensor, a road surface state sensor, a meteorological monitoring station and a user reporting terminal. Specifically, the high-definition camera group adopts a 360° panoramic shooting mode, monitors points are arranged at intervals, industrial-grade cameras are equipped, night vision function is provided, road surface dynamics can be clearly captured in low light environment, an intelligent analysis chip is built-in, vehicle trajectory recognition, abnormal behavior detection and dangerous target classification can be realized in real time, the identification algorithm adopts an improved YOLOv8 model, and a attention mechanism module is introduced to perform weighted processing on feature maps; the millimeter wave radar array adopts a specific frequency band, radar base stations are arranged, a certain area is covered, and the ability to track multiple moving targets at the same time is provided, and a constant false alarm rate detection algorithm is adopted for radar data processing; the infrared thermal imaging sensor is mainly used for detecting high-temperature objects on the road surface, and clear imaging of the target is realized in a lightless environment, and thermal image data output by the sensor is analyzed in real time through an edge computing node; the road surface state sensor is buried on both sides of the highway lane line, monitoring sections are arranged, and the road surface temperature, humidity and ice thickness are measured; the meteorological monitoring station is arranged along the highway, and real-time monitoring of wind speed, wind direction, precipitation and visibility meteorological parameters is realized; the user reporting terminal is an interactive interface integrated in a vehicle navigation system / mobile phone APP, the user reports a sudden dangerous event in multiple forms, the reported information is automatically attached with related information, and an intelligent verification module is built-in; The data processing center adopts a cloud computing architecture, and comprises a data storage server, a distributed computing node, an artificial intelligence analysis platform and a database system; Specifically, the data storage server adopts a distributed file system, has the ability to process data writing, stores original monitoring data and analysis results; the distributed computing node adopts a GPU cluster architecture, comprises multiple computing nodes, and realizes parallel processing of multiple source data; The artificial intelligence analysis platform is loaded with a multi-modal fusion model based on a Transformer architecture, and feature vectors output by each sub-module are fused through an attention mechanism; the database system adopts a combination of a relational database and a time series database; The information transmission network adopts a hybrid architecture of "edge computing + 5G private network + optical fiber backbone network"; the edge computing node is deployed in a communication base station or a monitoring room along the highway, is connected with the nearby collection equipment through a wired mode, is responsible for preprocessing the collected data, and communicates with the data processing center through a 5G private network or an optical fiber backbone network; The vehicle-mounted receiving terminal is integrated in a vehicle-mounted information entertainment system or a head-up display device of a vehicle, and supports a 5G-V2X communication protocol; the vehicle-mounted receiving terminal is internally provided with a positioning module and a map engine, can perform localized processing on dangerous warning information according to a current position and a driving direction of the vehicle, and performs multi-level early warning through various ways when the vehicle is about to enter a dangerous area; The emergency linkage unit is connected with systems of a highway management department, a traffic police department, a fire department and emergency disposal units such as a medical rescue institution, information sharing and cooperative linkage are realized through a preset interface protocol, the emergency linkage unit sends event details to the related emergency disposal units, generates a preliminary disposal suggestion according to an event type and a severity, and simultaneously feeds back an emergency disposal progress to the data processing center in real time, forming a closed-loop management.

2. The highway hazard warning transmission system of claim 1, wherein: The high-definition camera group is provided with a monitoring point every 400-600 m, a single monitoring point is equipped with 3-4 industrial-grade cameras with a resolution of 8K and night vision function, and can clearly capture the road surface dynamics in an environment with light intensity lower than 0.1 lux, and the specific formula of the identification algorithm is: Attention(F)=σ(W2·ReLU(W1·F))·F Wherein, F is an input feature map, W1 and W2 are convolution layer weight parameters, and sigma is a Sigmoid activation function, after the algorithm processing, the abnormal behavior identification response time is less than 200 ms, the intelligent analysis chip adopts a heterogeneous computing architecture, integrates CPU, GPU and NPU cores, the computing power of the NPU core reaches 20TOPS, can parallelly process multiple video stream data, and supports dynamic adjustment of computing resource allocation, when a suspected dangerous target appears in the picture, the image processing priority of the region is automatically improved.

3. The highway hazard warning transmission system of claim 1, wherein: The millimeter wave radar array adopts a 77GHz frequency band, and a radar base station is arranged every 800-1000 m, a single base station covers an area with a radius of 500 m, and has the ability to simultaneously track more than 200 moving targets, the ranging accuracy is ±0.1 m, the speed measurement accuracy is ±0.05 m / s, the MIMO technology is adopted, 12 transmitting antennas and 16 receiving antennas are included, three-dimensional positioning of the target is realized through digital beamforming technology, the azimuth angle measurement range is-60° to +60°, the accuracy is ±0.5°, the elevation angle measurement range is-10° to +10°, the accuracy is ±1°, and the vehicle on the road surface and the fixed object on the roadside can be effectively distinguished.

4. The highway hazard warning transmission system of claim 1, wherein: The infrared thermal imaging sensor detects a wavelength range of 8-14um, a temperature measurement range of-20℃ to 150℃, a temperature measurement accuracy of ±2℃, and can clearly image the target in a lightless environment, the thermal image data output by the sensor is analyzed in real time through an edge computing node, an abnormal area segmentation algorithm based on temperature gradient is adopted, when a region with a temperature exceeding a preset threshold and an area greater than 0.5㎡ is detected, a dangerous early warning is triggered.

5. The highway hazard warning transmission system of claim 1, wherein: The road surface state sensor is buried on both sides of the highway lane line, one monitoring section is arranged every 2 km, each section contains 3 sensor units for measuring road surface temperature, humidity and ice thickness, the road surface temperature measurement range is-40℃ to 80℃, the accuracy is ±0.5℃, the humidity measurement range is 0% to 100% RH, the accuracy is ±3% RH, the ice thickness measurement range is 0-50mm, the accuracy is ±1mm, the sensor adopts digital signal output, and is connected to the nearby edge computing node through a wired mode, and the data sampling frequency is 1Hz.

6. The highway hazard warning transmission system of claim 1, wherein: The meteorological monitoring station is arranged every 5 km along the highway, and real-time monitoring of wind speed, wind direction, precipitation and visibility meteorological parameters is realized, the wind speed measurement range is 0-60m / s, the accuracy is ±0.5m / s, the wind direction measurement range is 0-360°, the accuracy is ±5°, the precipitation measurement resolution is 0.2mm, the visibility measurement range is 10m to 10km, the accuracy is ±10%, and the meteorological data is uploaded to the data processing center in real time through a 4G / 5G wireless network.

7. The highway hazard warning transmission system of claim 1, wherein: The user reporting terminal is an interactive interface integrated in a vehicle navigation system or a mobile phone APP, and the user reports a sudden dangerous event in the form of text, picture or video, the reporting information is automatically attached with reporting time, geographical position and reporting user identification, and an intelligent checking module is built in the terminal to preliminarily screen the reporting content and eliminate obviously false or repeated information.

8. The highway hazard warning transmission system of claim 1, wherein: The data storage server of the data processing center adopts a distributed file system, has the ability of processing more than 10GB of data writing per second, and stores original monitoring data and analysis results for at least 3 months; the distributed computing node adopts a GPU cluster architecture, contains more than 100 computing nodes, each node is equipped with 8 NVIDIA A100 graphics cards, the total computing capacity reaches more than 10PFlops, and parallel processing of multi-source data is realized; the artificial intelligence analysis platform is loaded with a multi-modal fusion model based on a Transformer architecture, the model contains a text processing sub-module, an image processing sub-module, a radar data processing sub-module and a sensor data processing sub-module, and feature vectors output by each sub-module are fused through an attention mechanism, and the specific fusion formula is: wherein, F i is the feature vector of the i-th data source, α i is the corresponding attention weight, and satisfies α i The weight matrix obtained by training is calculated, and after the model is trained by the historical data, the model has self-learning ability. New dangerous event samples and disposal results are received in real time through the online learning module, an incremental training algorithm is used to update the multi-modal fusion model, the model parameters updated each time are synchronized to each edge computing node through the federated learning mode, and the identification ability of the entire system is continuously improved. The database system adopts the combination of a relational database and a time series database. The relational database is used for storing system configuration information, user data, and event processing record structured data. The time series database is used for storing time series data generated by sensors, cameras, and radar devices, and supports high-concurrency read-write and fast query.

9. The highway hazard warning transmission system of claim 1, wherein: The edge computing node of the information transmission network is arranged in a communication base station or a monitoring room along the highway, is connected to the nearby collection equipment through a wired mode, is responsible for preprocessing of the collected data, adopts an edge cloud architecture, supports containerized deployment, dynamically expands computing resources according to actual business requirements, the processing capacity of a single edge computing node supports simultaneous access of 50 high-definition video streams, 20 millimeter wave radars and 100 various sensors, and has a local data caching function, when communication with the data processing center is interrupted, key data is stored locally, and after communication is restored, the key data is automatically transmitted. The edge computing node and the data processing center communicate through a 5G private network or a fiber backbone network. The 5G private network uses slicing technology to allocate dedicated bandwidth resources for the system, and the end-to-end transmission delay is less than 10 ms. The fiber backbone network uses redundant link design to ensure the reliability of data transmission.

10. The highway hazard warning transmission system of claim 1, wherein: The vehicle-mounted receiving terminal is integrated into the vehicle's infotainment system or head-up display device, supports 5G-V2X communication protocol, receives dangerous warning information from the information transmission network, and has a built-in positioning module and map engine. The terminal can process the dangerous warning information locally according to the current position and driving direction of the vehicle. When the vehicle is about to enter a dangerous area, it will issue multi-level warnings through sound, image, and vibration in different levels, including prompt, warning, and emergency. Different levels correspond to different warning intensities and response measures. The system supports deep integration with the vehicle control system. When receiving an emergency-level dangerous warning, it automatically sends control signals to the vehicle's electronic stability program and adaptive cruise control system, suggesting the vehicle to slow down, maintain a distance, or issue a lane departure warning. The specific control strategy is dynamically adjusted according to the type of danger and the current state of the vehicle. The emergency linkage unit has an event level evaluation model. Based on the type, impact range, duration, and potential risk of the dangerous event, the model uses the analytic hierarchy process to calculate the comprehensive level of the event, and divides the event into four levels: general, large, major, and extremely major. Different levels correspond to different emergency response plans and linkage units. The system also includes a self-checking and fault-tolerant module that periodically checks the status of each collection device, transmission network, data processing center, and terminal device. By sending diagnostic data packets, monitoring device operating parameters, and verifying data integrity, the module identifies fault points and issues graded alarms. For minor faults that can be self-healing, it automatically starts a backup link or recovery program. For serious faults, it notifies maintenance personnel for processing in a timely manner.

Images