Tunnel event feature extraction method and system based on multi-source data fusion

Abstract

The application provides a tunnel event feature extraction method and system based on multi-source data fusion, and relates to the technical field of intelligent transportation. The method comprises the following steps: deploying multiple types of sensing devices at a set interval in a tunnel and accessing a tunnel management and control platform to collect multi-source heterogeneous data in real time; performing time synchronization calibration on the multi-source heterogeneous data to obtain synchronized data; performing data cleaning and error correction on the synchronized data, and performing environmental temperature and humidity compensation calibration on the detection data of the geomagnetic sensor and the ultrasonic sensor to obtain high-quality preprocessed multi-source data. The application improves the accuracy and environmental adaptability of tunnel operation situation awareness and provides reliable technology for tunnel safety management and control.

Description

Technical Field

[0001] This invention relates to the field of intelligent transportation technology, and in particular to a method and system for extracting tunnel event features based on multi-source data fusion. Background Technology

[0002] As a closed and critical control node in highway traffic, highway tunnels have small internal spaces and poor ventilation and visibility. Once a sudden event such as a vehicle collision or fire occurs, it can cause congestion and secondary accidents. Traditional tunnel safety monitoring systems mostly adopt a working mode of independent data collection by a single sensing device and independent processing by a separate platform.

[0003] Taking a long, one-way tunnel on a mountain highway as an example, the tunnel is only equipped with high-definition video, geomagnetic sensors, and smoke sensors in sections. The data from these devices belong to different control subsystems, and they do not achieve unified time synchronization and spatial coordinate association. Furthermore, the detection data from geomagnetic and ultrasonic sensors are easily affected by environmental interference such as temperature and humidity fluctuations and exhaust dust inside the tunnel. The single data source is used for event judgment, and there is a lack of cross-validation of multi-source data and multi-scale spatial fusion analysis. It is difficult to accurately capture detailed event characteristics such as sudden changes in vehicle trajectory, collision impacts, and abnormal smoke concentrations. The tunnel generally suffers from technical defects such as delayed accident identification, low positioning accuracy, high misjudgment rate of event types, and poor environmental adaptability. Summary of the Invention

[0004] This invention provides a method and system for extracting tunnel event features based on multi-source data fusion, which improves the accuracy and environmental adaptability of tunnel operation situation awareness and provides reliable technology for tunnel safety management.

[0005] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows:

[0006] Firstly, a method for extracting tunnel event features based on multi-source data fusion, the method comprising:

[0007] Various types of sensing devices are deployed at set intervals inside the tunnel and connected to the tunnel management platform to collect multi-source heterogeneous data in real time.

[0008] Time synchronization calibration is performed on multi-source heterogeneous data to obtain synchronized data; data cleaning and error correction are performed on synchronized data; environmental temperature and humidity compensation calibration is performed on the detection data of geomagnetic sensor and ultrasonic sensor to obtain high-quality preprocessed multi-source data.

[0009] Spatial coordinate information of traffic information display screens and fixed fire-fighting facilities deployed in the tunnel is extracted. Based on the physical location and electromagnetic wave coverage characteristics of the spatial coordinate information of traffic information display screens and fixed fire-fighting facilities in the tunnel section, a dynamic monitoring spatial envelope is constructed. The preprocessed multi-source data is divided into several levels according to the spatial scale and physical attributes of the dynamic monitoring spatial envelope to form a multi-scale data field.

[0010] Establish the mapping relationship between different levels of data in a multi-scale data field. By analyzing the coupling effect of each level of data in terms of spatial scale and physical properties, identify the micro-fluctuation characteristics that reflect the tunnel operation status. Based on this, generate dynamic calibration coefficients. Then, iteratively optimize the environmental temperature and humidity compensation calibration results in the preprocessed high-quality multi-source data using the dynamic calibration coefficients to obtain the iteratively optimized multi-source data.

[0011] The multi-source data after iterative optimization is fused at the feature level, and traffic accidents in tunnels are identified through correlation analysis, and the core features of the accidents are extracted.

[0012] The core characteristics of the accident are pushed to the tunnel emergency command center in real time, providing data support for emergency dispatch and traffic control.

[0013] Furthermore, various types of sensing devices are deployed at predetermined intervals within the tunnel and connected to the tunnel management platform to collect multi-source heterogeneous data in real time, including:

[0014] Sensing devices are deployed at predetermined intervals within the tunnel. These devices include high-definition video cameras, geomagnetic sensors, smoke sensors, temperature sensors, and ultrasonic sensors. Simultaneously, they are connected to the tunnel management platform to collect real-time traffic flow statistics and vehicle passage records. The high-definition video cameras capture vehicle movement and accident scene footage; the geomagnetic sensors collect data on vehicle presence, speed, and traffic density; the smoke sensors collect data on smoke concentration caused by accidents; the temperature sensors collect data on ambient temperature changes; and the ultrasonic sensors assist in identifying vehicle positions and collision severity. The tunnel management platform provides traffic flow statistics and vehicle passage records, collectively forming multi-source heterogeneous data.

[0015] Furthermore, time synchronization calibration is performed on the multi-source heterogeneous data to obtain synchronized data; data cleaning and error correction are then performed on the synchronized data, and environmental temperature and humidity compensation calibration is applied to the detection data from the geomagnetic sensor and ultrasonic sensor to obtain preprocessed high-quality multi-source data, including:

[0016] Add a unified timestamp to the multi-source heterogeneous data, perform time synchronization calibration, eliminate the time difference in data acquisition between various sensing devices and the tunnel control platform, and obtain synchronized data;

[0017] Outliers and missing values ​​in the synchronized data are removed and filled in, and errors are corrected for data affected by the tunnel environment, resulting in preliminarily cleaned data.

[0018] For the detection data collected by the geomagnetic sensor and ultrasonic sensor in the preliminary cleaned data, combined with the synchronously collected ambient temperature and humidity data, ambient temperature and humidity compensation calibration is performed to eliminate the influence of ambient temperature and humidity on the detection accuracy, and obtain high-quality multi-source data after preprocessing.

[0019] Furthermore, the spatial coordinate information of traffic information display screens and fixed fire-fighting facilities deployed within the tunnel is extracted. Based on the physical location and electromagnetic wave coverage characteristics of these spatial coordinates in the tunnel cross-section, a dynamic monitoring spatial envelope is constructed. The preprocessed multi-source data is then divided into several levels according to the spatial scale and physical attributes of the dynamic monitoring spatial envelope, forming a multi-scale data field, including:

[0020] Extract the spatial coordinates of traffic information display screens and fixed fire-fighting facilities deployed inside the tunnel to obtain the precise physical location data of each facility within the tunnel cross section;

[0021] Based on the physical location of traffic information display screens and fixed fire-fighting facilities in the tunnel cross section and their electromagnetic wave coverage characteristics, a dynamic monitoring spatial envelope covering the tunnel monitoring area is constructed.

[0022] The preprocessed high-quality multi-source data is divided into several levels according to the spatial scale and physical attributes of the dynamic monitoring spatial envelope, so that the data at each level corresponds to the spatial range and monitoring attributes of the dynamic monitoring spatial envelope, thus forming a multi-scale data field.

[0023] Furthermore, a mapping relationship is established between data at different levels in a multi-scale data field. By analyzing the coupling effect of data at each level in terms of spatial scale and physical properties, the microscopic fluctuation characteristics reflecting the tunnel's operational status are identified. Based on this, dynamic calibration coefficients are generated. The environmental temperature and humidity compensation calibration results in the preprocessed high-quality multi-source data are iteratively optimized using these dynamic calibration coefficients to obtain the iteratively optimized multi-source data, including:

[0024] Spatial location matching and physical attribute association are performed on data at different levels in a multi-scale data field to establish mapping relationships between data at each level.

[0025] After establishing the mapping relationship, the data at each level are compared and analyzed to extract the data differences at different scale levels for the same spatial location, as well as the data differences for the same physical attribute at different monitoring dimensions. Correlation analysis is used to identify the coupling effect of data at each level in terms of spatial scale and physical attribute, and to obtain the micro-fluctuation characteristics of the tunnel operation status.

[0026] Based on the direction and magnitude of the data deviation reflected by the microscopic fluctuation characteristics, dynamic calibration coefficients corresponding to each spatial location and each physical property are calculated and generated.

[0027] The dynamic calibration coefficients are applied to the preprocessed high-quality multi-source data, and the completed environmental temperature and humidity compensation calibration results are subjected to a second iteration calibration to eliminate residual errors caused by the uneven spatial distribution and temporal changes of environmental factors. The above mapping, identification, generation and calibration are repeated until the data deviation converges, and the iteratively optimized multi-source data is obtained.

[0028] Furthermore, the iteratively optimized multi-source data is fused at the feature level, and traffic accidents within the tunnel are identified through correlation analysis. The core features of the accidents are extracted, including:

[0029] Feature-level fusion is performed on the iteratively optimized multi-source data, and video image features, sensor numerical features and platform recorded text features under the same spatiotemporal section are associated and matched, so that data of different modalities can corroborate each other at the feature level and form a fused feature set.

[0030] Accident identification analysis is performed on the fused feature set. By analyzing the coordinated changes of multiple parameters such as sudden changes in vehicle trajectory, sudden drop in vehicle speed, abnormal traffic flow density, increase in smoke concentration, and increase in ambient temperature, it can detect whether a traffic accident has occurred in the tunnel.

[0031] When a traffic accident is detected, correlation analysis is performed on the fused feature set to extract the core features of the accident. These core features include: obtaining the precise location of the accident through cross-validation of spatial positioning results from multi-source data; comprehensively determining the accident type based on vehicle posture identified from video images, collision shock waves detected by sensors, and vehicle deformation sensed by ultrasonic waves; statistically analyzing the number of vehicles involved and their vehicle type distribution by comparing the number of video target detections with the number of triggers from geomagnetic sensors; calculating the vehicle collision angle based on continuous frame changes in vehicle trajectories; extracting the trend of smoke concentration changes based on time-series data from smoke sensors; and calculating the surrounding traffic congestion range affected by the accident based on traffic density data from upstream and downstream geomagnetic sensors and video detections, thus obtaining the core features of the accident.

[0032] Furthermore, the core characteristics of the accident will be pushed to the tunnel emergency command center in real time to provide data support for emergency dispatch and traffic control, including:

[0033] The core characteristics of the accident are standardized and encapsulated, and structured accident characteristic information messages are generated in accordance with the data interface specifications of the emergency command center.

[0034] Structured accident characteristic information messages are pushed to the tunnel emergency command center in real time through a dedicated communication network, enabling the emergency command center to simultaneously obtain core characteristic data such as the precise location of the accident, the type of accident, the number and type distribution of vehicles involved, the angle of vehicle collision, the trend of smoke concentration changes, and the range of traffic congestion in the surrounding area.

[0035] Based on the core characteristics of the accident pushed out, the system provides the emergency command center with the coordinates of the accident location and the accessible route for dispatching rescue vehicles, provides the traffic control decision-making with the scope of congestion and the severity of the accident, and provides the on-site response plan with the accident type and information on the vehicles involved, thus achieving precise data support for emergency dispatch and traffic control.

[0036] Secondly, a tunnel event feature extraction system based on multi-source data fusion includes:

[0037] The acquisition module is used to deploy various types of sensing devices at set intervals within the tunnel and connect to the tunnel management platform to collect multi-source heterogeneous data in real time.

[0038] The calibration module is used to perform time synchronization calibration on multi-source heterogeneous data to obtain synchronized data; it also performs data cleaning and error correction on the synchronized data, and performs environmental temperature and humidity compensation calibration on the detection data of geomagnetic sensors and ultrasonic sensors to obtain high-quality pre-processed multi-source data.

[0039] The module is used to extract the spatial coordinate information of traffic information display screens and fixed fire-fighting facilities deployed in the tunnel. Based on the physical location and electromagnetic wave coverage characteristics of the spatial coordinate information of the traffic information display screens and fixed fire-fighting facilities in the tunnel section, a dynamic monitoring spatial envelope is constructed. The preprocessed multi-source data is divided into several levels according to the spatial scale and physical attributes of the dynamic monitoring spatial envelope to form a multi-scale data field.

[0040] The iterative module is used to establish the mapping relationship between different levels of data in a multi-scale data field. By analyzing the coupling effect of each level of data in terms of spatial scale and physical properties, it identifies the micro-fluctuation characteristics that reflect the tunnel operation status, generates dynamic calibration coefficients, and iteratively optimizes the environmental temperature and humidity compensation calibration results in the preprocessed high-quality multi-source data through the dynamic calibration coefficients to obtain the iteratively optimized multi-source data.

[0041] The extraction module is used to perform feature-level fusion of the iteratively optimized multi-source data, identify traffic accidents in tunnels through correlation analysis, and extract the core features of the accidents.

[0042] The processing module is used to push the core characteristics of the accident to the tunnel emergency command center in real time, providing data support for emergency dispatch and traffic control.

[0043] Thirdly, a computing device, comprising:

[0044] One or more processors;

[0045] A storage device for storing one or more programs that, when executed by one or more processors, cause the one or more processors to implement the method.

[0046] Fourthly, a computer-readable storage medium storing a program that, when executed by a processor, implements the method.

[0047] The above-described solution of the present invention has at least the following beneficial effects:

[0048] This approach overcomes the technical challenges of traditional tunnel monitoring, which involves deploying multiple types of sensing devices at set intervals and connecting them to the tunnel management platform to collect multi-source heterogeneous data. It performs time synchronization calibration, data cleaning, error correction, and temperature and humidity compensation calibration for geomagnetic and ultrasonic sensors. Based on the spatial coordinates of fixed facilities within the tunnel, it constructs a dynamic monitoring spatial envelope and a multi-scale data field. A hierarchical data mapping relationship is established, and dynamic calibration coefficients are generated through coupling effect analysis for iterative data optimization. The optimized data is then fused at the feature level to identify accidents and extract core features, which are then pushed to the emergency command center in real time. This overcomes the technical problems of traditional tunnel monitoring, such as independent operation of single sensing devices, lack of spatiotemporal collaborative calibration of multi-source data, large errors in sensor detection due to environmental interference, and lack of multi-source cross-verification and multi-scale fusion analysis, leading to delayed accident identification, low positioning accuracy, high misjudgment rate, and poor environmental adaptability. Ultimately, it improves the accuracy and environmental adaptability of tunnel operational situational awareness, enables precise identification of tunnel traffic accidents and accurate extraction of core features, provides reliable data support for tunnel emergency dispatch and traffic control, ensures tunnel operational safety, and improves emergency response efficiency. Attached Figure Description

[0049] Figure 1 This is a flowchart illustrating the tunnel event feature extraction method and system based on multi-source data fusion provided in an embodiment of the present invention.

[0050] Figure 2 This is a schematic diagram of the process of the tunnel event feature extraction method and system based on multi-source data fusion provided in the embodiments of the present invention. Detailed Implementation

[0051] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0052] like Figure 1 As shown, embodiments of the present invention propose a method for extracting tunnel event features based on multi-source data fusion, the method comprising the following steps:

[0053] Step 1: Deploy various types of sensing devices at set intervals inside the tunnel and connect them to the tunnel management platform to collect multi-source heterogeneous data in real time;

[0054] Step 2: Perform time synchronization calibration on the multi-source heterogeneous data to obtain synchronized data; perform data cleaning and error correction on the synchronized data; perform environmental temperature and humidity compensation calibration on the detection data of the geomagnetic sensor and ultrasonic sensor to obtain high-quality preprocessed multi-source data.

[0055] Step 3: Extract the spatial coordinate information of the traffic information display screens and fixed fire-fighting facilities deployed in the tunnel. Based on the physical location and electromagnetic wave coverage characteristics of the spatial coordinate information of the traffic information display screens and fixed fire-fighting facilities in the tunnel section, construct a dynamic monitoring spatial envelope. Divide the preprocessed multi-source data into several levels according to the spatial scale and physical attributes of the dynamic monitoring spatial envelope to form a multi-scale data field.

[0056] Step 4: Establish the mapping relationship between different levels of data in the multi-scale data field. By analyzing the coupling effect of each level of data in terms of spatial scale and physical properties, identify the micro-fluctuation characteristics that reflect the tunnel operation status. Based on this, generate dynamic calibration coefficients. Then, iteratively optimize the environmental temperature and humidity compensation calibration results in the preprocessed high-quality multi-source data using the dynamic calibration coefficients to obtain the iteratively optimized multi-source data.

[0057] Step 5: Perform feature-level fusion on the iteratively optimized multi-source data, identify traffic accidents in the tunnel through correlation analysis, and extract the core features of the accidents;

[0058] Step 6: Push the core characteristics of the accident to the tunnel emergency command center in real time to provide data support for emergency dispatch and traffic control.

[0059] In this embodiment of the invention, multiple types of sensing devices are deployed at set intervals within the tunnel and connected to the tunnel management platform to collect multi-source heterogeneous data in real time. The multi-source heterogeneous data undergoes time synchronization calibration, data cleaning, error correction, and environmental temperature and humidity compensation calibration for data detected by geomagnetic and ultrasonic sensors. Spatial coordinates of traffic information display screens and fixed fire-fighting facilities within the tunnel are extracted, and a dynamic monitoring spatial envelope is constructed. Multi-scale data fields are divided, hierarchical data mapping relationships are established, coupling effects are analyzed to identify microscopic fluctuation characteristics, and dynamic calibration coefficients are generated for iterative data optimization. Feature-level fusion is performed on the optimized data to identify tunnel traffic accidents and extract their core features. The technical means of pushing the core features of an accident to the tunnel emergency command center in real time overcomes the technical problems of traditional tunnel event feature extraction, such as independent collection by a single sensing device, lack of spatiotemporal collaborative calibration of multi-source data, large errors in sensor detection due to environmental interference, lack of multi-scale fusion analysis and cross-verification of multi-source data, which lead to delayed accident identification, low positioning accuracy, high misjudgment rate, poor environmental adaptability, and inability to provide accurate data support for emergency dispatch. It improves the comprehensiveness and accuracy of multi-source data collection in tunnels, enhances data environmental adaptability, realizes accurate identification of tunnel traffic accidents and accurate extraction of core accident features, and provides reliable data support for tunnel emergency dispatch and traffic control in real time.

[0060] In a preferred embodiment of the present invention, step 1 above may include:

[0061] Step 1.1: Deploy sensing devices at predetermined intervals within the tunnel. These devices include high-definition video cameras, geomagnetic sensors, smoke sensors, temperature sensors, and ultrasonic sensors. Simultaneously, they are connected to the tunnel management platform to collect real-time traffic flow statistics and vehicle passage records. The high-definition video cameras capture vehicle movement and accident scene footage; the geomagnetic sensors collect data on vehicle presence, speed, and traffic density; the smoke sensors collect data on smoke concentration caused by accidents; the temperature sensors collect data on ambient temperature changes; and the ultrasonic sensors assist in identifying vehicle positions and collision severity. The tunnel management platform provides traffic flow statistics and vehicle passage records, collectively forming multi-source heterogeneous data. Specifically, taking a 5000-meter-long, one-way tunnel on a mountainous highway as the implementation target, and considering the tunnel's cross-sectional width of 10.5 meters and height of 5 meters, to ensure overlapping monitoring areas between adjacent devices and no blind spots, the deployment interval of the sensing devices is calculated as 0.8 times the effective monitoring distance of the devices. The deployment scheme is determined according to this calculation method for each category.

[0062] The effective monitoring distance of the high-definition video camera is 100 meters. 100 meters × 0.8 = 80 meters, that is, one high-definition video camera is deployed every 80 meters.

[0063] The effective monitoring distance of the geomagnetic sensor is 25 meters, 25 meters × 0.8 = 20 meters, that is, one geomagnetic sensor is deployed every 20 meters;

[0064] The effective detection distance of the smoke sensor is 62.5 meters, which is 62.5 meters × 0.8 = 50 meters. That is, one smoke sensor is deployed every 50 meters.

[0065] The effective monitoring distance of the temperature sensor is 37.5 meters, which is 37.5 meters × 0.8 = 30 meters. That is, one temperature sensor is deployed every 30 meters.

[0066] The effective monitoring distance of the ultrasonic sensor is 50 meters. 50 meters × 0.8 = 40 meters, that is, one ultrasonic sensor is deployed every 40 meters.

[0067] Based on the calculated deployment intervals, all sensing devices are uniformly installed on the tunnel sidewall at a height of 2.5 meters above the road surface along the driving direction, ensuring that the monitoring direction is consistent with the driving direction, thus completing the deployment of sensing devices throughout the tunnel.

[0068] All deployed high-definition video cameras, geomagnetic sensors, smoke sensors, temperature sensors, and ultrasonic sensors are connected to the tunnel management platform via industrial Ethernet. The maximum communication latency threshold between the devices and the platform is set to 100 milliseconds to ensure real-time data transmission. At the same time, the traffic flow statistics and vehicle passage record collection functions of the tunnel management platform are enabled. Traffic flow statistics are performed at a fixed 5-minute interval, and vehicle passage records are collected in real-time. The platform assigns a unique number to each vehicle passage record to ensure that all collected data is traceable.

[0069] The high-definition video camera captures real-time images of vehicle movement and accident scenes within the tunnel at a frequency of 25 frames per second. The resolution of each frame is set to 1920×1080, ensuring that vehicle trajectories and body postures are clearly identifiable.

[0070] The geomagnetic sensor collects vehicle presence signals, driving speed, and traffic density at a sampling frequency of 50 Hz.

[0071] The smoke sensor collects smoke concentration data caused by the accident at a sampling frequency of 1 Hz. The normal concentration threshold is set to 0 to 0.5 milligrams per cubic meter. Data exceeding this range is recorded as abnormal data.

[0072] The temperature sensor collects ambient temperature data inside the tunnel at a sampling frequency of 0.5 Hz, sets the normal temperature threshold to -10 degrees Celsius to 40 degrees Celsius, and continuously records the real-time changes in ambient temperature.

[0073] Ultrasonic sensors, with a sampling frequency of 20 Hz, help identify the vehicle's position and the extent of the collision.

[0074] The system integrates video data from high-definition cameras, traffic flow data from geomagnetic sensors, smoke concentration data from smoke sensors, ambient temperature data from temperature sensors, vehicle location and collision severity data from ultrasonic sensors, as well as traffic flow statistics and vehicle passage records output by the tunnel management platform. All data carries three basic attributes: collection time, device number, and spatial location, ultimately forming multi-source heterogeneous data that includes video images, numerical signals, and platform records.

[0075] In this embodiment of the invention, various sensing devices, such as high-definition video cameras, geomagnetic sensors, smoke sensors, temperature sensors, and ultrasonic sensors, are deployed at set intervals within the tunnel. These devices are simultaneously connected to a tunnel management platform. Each sensing device collects data on vehicle movement status, accident scene footage, vehicle presence and speed, traffic density, smoke concentration, ambient temperature changes, vehicle location, and collision severity. This data, combined with traffic flow statistics and vehicle passage records provided by the tunnel management platform, constitutes a multi-source heterogeneous data collection technique. This overcomes the limitations of traditional tunnel data acquisition methods, which rely on a single type of sensing device and limited data dimensions, only acquiring partial tunnel operation-related data and lacking collaborative collection of multi-dimensional data such as vehicle status, environmental parameters, and traffic flow. This results in insufficient comprehensive data support and inadequate data correlation for event identification. Ultimately, this invention achieves full-domain collection of multi-dimensional data on tunnel operation, enriches data sources and types, ensures the comprehensiveness and correlation of multi-source heterogeneous data, provides high-quality foundational data for data preprocessing, feature fusion, and event feature extraction, and improves the accuracy of tunnel event identification.

[0076] In a preferred embodiment of the present invention, step 2 above may include:

[0077] Step 2.1 involves adding a unified timestamp to the multi-source heterogeneous data and performing time synchronization calibration to eliminate the time difference in data acquisition between various sensing devices and the tunnel control platform, resulting in synchronized data. Specifically, this includes: using the network time protocol to perform global time synchronization calibration on all multi-source heterogeneous data, including data collected by high-definition video cameras, geomagnetic sensors, smoke sensors, temperature sensors, and ultrasonic sensors, as well as traffic flow statistics and vehicle passage records from the tunnel control platform. A standard timestamp based on the East Eighth Time Zone and accurate to milliseconds is added to all data. The time synchronization reference source for all devices is unified to the core time server of the tunnel control platform. The allowed threshold for time synchronization deviation is set to 5 milliseconds. The acquisition clock of each device is calibrated through the network time protocol. If the time difference between any device and the core time server exceeds 5 milliseconds, it is automatically forced to calibrate to the standard time, thereby completely eliminating the time difference in data acquisition between various sensing devices and the tunnel control platform, completing the time sequence unification of all data, and obtaining standardized data after time synchronization.

[0078] Step 2.2 involves removing and completing outliers and missing values ​​in the synchronized data, and correcting errors in data affected by tunnel environmental interference, resulting in preliminarily cleaned data. Specifically, this includes: after time synchronization, performing comprehensive cleaning of the synchronized data; setting specific processing thresholds and calculation methods for data anomalies, missing values, and environmental interference; outlier removal by clearly defining the normal value ranges for various data types; and directly identifying and removing data exceeding these ranges. Specifically, the normal value range for vehicle speed is 0 to 120 km / h, for traffic density is 0 to 120 vehicles per kilometer, and for smoke concentration is 0 to... 0.5 mg / m³, with a normal ambient temperature range of -10°C to 40°C. All data exceeding these ranges are considered outliers and removed. For missing value completion, linear completion is used to fill in data gaps after outlier removal. Specifically, the completed missing data is the average of the sums of the preceding and following valid data points. This linear calculation ensures data continuity. For environmental interference error correction, a fixed correction coefficient is used to correct sensor data affected by tunnel exhaust and dust. In the formula For the corrected data, This is the original collected data. The environmental interference correction factor is set to 1.02. This formula is used to correct data errors and finally obtain the cleaned and normalized data.

[0079] Step 2.3: For the detection data collected by the geomagnetic sensor and ultrasonic sensor in the preliminary cleaned data, combined with the synchronously collected ambient temperature and humidity data, environmental temperature and humidity compensation calibration is performed to eliminate the influence of ambient temperature and humidity on the detection accuracy, obtaining high-quality pre-processed multi-source data. Specifically, this includes: addressing the error problem caused by temperature and humidity fluctuations inside the tunnel in the detection data of the geomagnetic sensor and ultrasonic sensor, combining the synchronously collected ambient temperature and humidity data, performing specific compensation calibration on the data of the two types of sensors, and setting environmental reference parameters and compensation coefficients: the temperature reference value is set to 25 degrees Celsius, the humidity reference value is set to 50%RH, the temperature compensation coefficient for the geomagnetic sensor is 0.015, the humidity compensation coefficient is 0.012, the temperature compensation coefficient for the ultrasonic sensor is 0.018, and the humidity compensation coefficient is 0.015; the comprehensive temperature and humidity compensation formula is: In the formula: To compensate for the data after calibration, This is the raw sensor data after initial cleaning. This is the temperature compensation coefficient. To collect ambient temperature in real time, As a temperature reference value, This is the humidity compensation coefficient. To collect ambient humidity in real time, The humidity baseline is used as the reference value. According to the above formula, the detection data of the geomagnetic sensor and the ultrasonic sensor are compensated point by point to eliminate the influence of changes in ambient temperature and humidity on the detection accuracy of the sensors. The preprocessing process of all data is completed, and finally high-quality multi-source data with no time deviation, no abnormal missing data, and no environmental interference error is obtained.

[0080] In this embodiment of the invention, by adding a unified timestamp to multi-source heterogeneous data for time synchronization calibration, removing and completing outliers and missing values ​​and correcting environmental interference errors in the synchronized data, and combining environmental temperature and humidity data to perform temperature and humidity compensation calibration on the detection data of geomagnetic sensors and ultrasonic sensors, the technical problems of asynchronous acquisition time between multi-source sensing devices and the control platform, abnormal missing data, and sensor detection accuracy being easily affected by tunnel temperature and humidity environment interference, resulting in large data errors and low reliability, are overcome. Thus, the data deviation caused by time sequence deviation and environmental factors is eliminated, the accuracy, consistency and stability of multi-source data are improved, and high-quality preprocessed data is obtained.

[0081] In a preferred embodiment of the present invention, step 3 above may include:

[0082] Step 3.1: Extract the spatial coordinates of the traffic information displays and fixed fire-fighting facilities deployed within the tunnel to obtain precise physical location data of each facility within the tunnel cross-section. Specifically, this includes: extracting the precise spatial coordinates of the traffic information displays and fixed fire-fighting facilities within the tunnel using GPS positioning and manual calibration; establishing a unified spatial coordinate system with the center point of the tunnel entrance cross-section as the origin (0, 0, 0); establishing a three-dimensional rectangular coordinate system where the x-axis extends along the tunnel's driving direction in meters, the y-axis extends along the tunnel's width in meters, and the z-axis extends along the tunnel's height in meters; and the coordinate accuracy is controlled within ±0.1 meters. Based on the actual tunnel structure, one traffic information display is deployed every 500 meters, for a total of 10 displays; one fire hydrant is deployed every 100 meters, for a total of 50 fire-fighting facilities. All facilities are installed on the tunnel sidewall at a height consistent with the previously mentioned sensing equipment (2.5 meters). A high-precision GPS locator is used to collect the original coordinates (x, y, z) of each traffic information display and fixed fire-fighting facility. 原 y 原 , z 原 Five location data points were collected for each facility, and the average value was used to eliminate location errors. Based on the above calibration, the precise coordinates of all facilities were calculated one by one. For example, the calibrated coordinates of the No. 1 traffic information display screen, which is 500 meters from the entrance, are (500.0, 5.25, 2.5), and the calibrated coordinates of the No. 1 fire hydrant, which is 100 meters from the entrance, are (100.0, 5.25, 2.5). And so on, until the precise physical location data of all traffic information display screens and fixed fire-fighting facilities in the tunnel section were obtained.

[0083] Step 3.2: Based on the physical location and electromagnetic wave coverage characteristics of the traffic information display screen and fixed fire-fighting facilities in the tunnel cross-section, construct a dynamic monitoring spatial envelope covering the tunnel monitoring area. Specifically, this includes: based on the acquired precise physical location data of the facilities, and combined with the electromagnetic wave coverage characteristics of various facilities, constructing a dynamic monitoring spatial envelope with full coverage and no monitoring blind spots; clarifying the electromagnetic wave coverage characteristics of the traffic information display screen and fixed fire-fighting facilities, where the effective electromagnetic wave coverage radius of the traffic information display screen is 50 meters and the effective electromagnetic wave coverage radius of the fixed fire-fighting facilities is 30 meters; to ensure no blind spots and reasonable overlap between adjacent envelopes, an envelope overlap rate of 20% is set, meaning the overlap distance between two adjacent envelopes is 20% of the coverage radius; calculating the dynamic monitoring spatial envelope dimensions corresponding to the two types of facilities respectively. The envelope shape adopts a cylindrical shape to adapt to the narrow tunnel structure, with the base radius of the cylinder equal to the effective electromagnetic wave coverage radius and the cylinder height equal to the tunnel's net height of 5 meters. The specific calculations are as follows:

[0084] The envelope corresponding to the traffic information display screen has a bottom radius of R1 = 50 meters and a height of H = 5 meters. The overlap distance between adjacent envelopes is 50 meters × 20% = 10 meters. Therefore, the connection distance between the coverage areas of two adjacent traffic information display screens is 50 meters - 10 meters = 40 meters, ensuring full coverage of the 5000-meter tunnel.

[0085] The envelope corresponding to the fixed fire protection facilities is: bottom radius R2 = 30 meters, height H = 5 meters, the overlap distance between adjacent envelopes is 30 meters × 20% = 6 meters, and the connection distance of the envelope coverage of two adjacent fire hydrants is 30 meters - 6 meters = 24 meters to ensure full coverage.

[0086] Envelope Construction and Integration: Centered on the precise coordinates of each traffic information display screen and fixed fire-fighting facility, corresponding cylindrical dynamic monitoring spatial envelopes are constructed. The envelope coverage range of the traffic information display screen is ±50 meters in the x-axis direction, ±50 meters in the y-axis direction, and 0 to 5 meters in the z-axis direction. The envelope coverage range of the fixed fire-fighting facility is ±30 meters in the x-axis direction, ±30 meters in the y-axis direction, and 0 to 5 meters in the z-axis direction. All envelopes are integrated, and overlapping coverage areas are eliminated to form a continuous, complete, and all-area dynamic monitoring spatial envelope that covers the entire 5000-meter tunnel monitoring area, ensuring that monitoring data from any location within the tunnel corresponds to a specific envelope range.

[0087] Step 3.3 involves dividing the preprocessed high-quality multi-source data into several levels according to the spatial scale and physical attributes of the dynamic monitoring spatial envelope. This ensures that each level of data corresponds to the spatial range and monitoring attributes of the dynamic monitoring spatial envelope, forming a multi-scale data field. Specifically, based on the constructed dynamic monitoring spatial envelope and combined with the physical attributes of the preprocessed high-quality multi-source data, the data is divided into several levels to form a multi-scale data field, achieving hierarchical and standardized data management. According to the spatial scale of the dynamic monitoring spatial envelope and the physical attributes of the data, the spatial scale includes global, regional, and local levels, and the physical attributes include traffic flow, environmental parameters, and vehicle status. The multi-source data is divided into three levels, with each level's spatial scale corresponding one-to-one with the range of the dynamic monitoring spatial envelope, and the physical attributes precisely matched with the data type. The specific division criteria are as follows:

[0088] Level 1 Global Scale: The spatial scale corresponds to the dynamic monitoring spatial envelope of the entire tunnel (5000m × 10.5m × 5m), the physical attribute is the overall operation status of the tunnel, and the corresponding data types are the traffic flow statistics, the average global ambient temperature, and the average global vehicle flow density of the tunnel management platform.

[0089] Second-level regional scale: The spatial scale corresponds to the envelope of a single traffic information display screen (100m×100m×5m, i.e., the range of ±50m on the x-axis), the physical attributes are the regional environment and traffic status, and the corresponding data types are smoke concentration data from smoke sensors, ambient temperature data from temperature sensors, and regional traffic density data from geomagnetic sensors within the region.

[0090] Level 3 Local Scale: The spatial scale corresponds to the envelope of a single fixed fire-fighting facility (60m×60m×5m, i.e., the range of ±30m on the x-axis). The physical attributes are local vehicle and environmental details. The corresponding data types are vehicle movement status images from high-definition video cameras in the area, vehicle position data from ultrasonic sensors, and single vehicle speed data from geomagnetic sensors.

[0091] The preprocessed high-quality multi-source data are categorized into corresponding levels according to the above hierarchical classification standards, ensuring that the data at each level fully corresponds to the spatial range and monitoring attributes of the dynamic monitoring spatial envelope. The traffic flow data every 5 minutes, the average ambient temperature of the entire tunnel, and the average vehicle density of the entire tunnel, statistically collected by the tunnel control platform, are categorized into the first-level level. Data collected by smoke sensors, temperature sensors, and geomagnetic sensors within the envelope of each traffic information display screen are categorized into the second-level level of the corresponding area. For example, within the envelope of traffic information display screen No. 5 (coordinates 2500.0, 5.25, 2.5), all data from smoke sensors, temperature sensors, and geomagnetic sensors are uniformly categorized into the second-level level. Area 5 at the second level; data collected by high-definition video cameras, ultrasonic sensors, and geomagnetic sensors within the envelope of each fixed fire-fighting facility are categorized into the corresponding local third level. For example, within the envelope of fire hydrant No. 25 (coordinates 2500.0, 5.25, 2.5), high-definition video footage, vehicle location data, and single vehicle speed data are uniformly categorized into local area No. 25 at the third level. After completing the hierarchical division of all data, the data at each level are interconnected and mutually supportive. The first-level data reflects the overall situation of the tunnel, the second-level data reflects regional details, and the third-level data reflects precise local information, forming a multi-scale data field with a clear structure, distinct levels, and close spatial connections.

[0092] In this embodiment of the invention, precise physical location data is obtained by extracting the spatial coordinates of traffic information display screens and fixed fire-fighting facilities within the tunnel. A dynamic monitoring spatial envelope is constructed based on their physical location and electromagnetic wave coverage characteristics. The pre-processed high-quality multi-source data is then divided into several levels according to the spatial scale and physical attributes of the spatial envelope to form a multi-scale data field. This overcomes the technical problems of insufficient spatial positioning accuracy, lack of unified planning for monitoring areas, lack of spatial hierarchy division of multi-source data, and poor spatial correlation and adaptability of data in traditional tunnel monitoring. As a result, a dynamic monitoring spatial framework that is accurately adapted to the tunnel scenario is constructed, enabling multi-scale, hierarchical, and orderly management of multi-source data.

[0093] In a preferred embodiment of the present invention, step 4 above may include:

[0094] Step 4.1 involves spatial location matching and physical attribute association of data at different levels in the multi-scale data field, establishing mapping relationships between data at each level. Specifically, this includes establishing a three-dimensional Cartesian coordinate system with the center point of the tunnel entrance section as the origin: the x-axis extends along the driving direction, the y-axis along the tunnel section width, and the z-axis along the tunnel height. A spatial location matching accuracy threshold of ±0.2 meters is set. This threshold serves as the core accuracy constraint for the ICP geometric registration algorithm. The coordinates of all levels of data must ultimately be calibrated to this coordinate system. The core of the ICP algorithm is to iteratively solve for rotation... The matrix and translation matrix minimize the registration error between different point cloud datasets. Spatial point clouds corresponding to each level of data are extracted. The first-level global level extracts the tunnel global 3D scan point cloud, denoted as the reference point cloud set P. The second-level regional level extracts the point clouds of each regional boundary and equipment installation point, denoted as the point cloud set to be registered Q1. The third-level local level extracts the high-precision point clouds of facilities such as fire hydrants and sensors, denoted as the point cloud set to be registered Q2. All point cloud sets are denoised to remove outliers and redundant points, and retain effective spatial coordinate points. For example, the third-level fire hydrant point cloud only retains the feature points of the outer envelope surface of the facility.

[0095] Initial registration and error function construction: Using the reference point cloud P as the baseline, calculate the nearest point to P for each point in the point cloud Q1 to be registered, and construct the initial point pair correspondence; define the registration error function: in, For the coordinates of a point in the reference point cloud P, The coordinates of the corresponding point in the point cloud Q1 to be registered. It is a 3×3 rotation matrix. It is a 3×1 translation matrix. To determine the number of matching point pairs, R and T are solved using the least squares method, thus maximizing the error function. To minimize this, complete the initial registration of Q1 relative to P. Using the initially registered Q1 as the new reference, repeat the iterative process of nearest point matching, error function solving, and R / T update for the level 3 point cloud set Q2. Set the iteration termination condition: registration error < 0.1 meters, and the number of iterations reaches 50. For example, for the target point with x-axis 2500 meters, y-axis 5.25 meters, and z-axis 2.5 meters, its corresponding original point cloud coordinates for level 3 fire hydrant No. 25 are (2500.1 meters). The original point cloud coordinates of the No. 5 traffic information display screen (2499.9, 5.2, 2.55) are obtained. After 8 iterations of the ICP algorithm, the rotation matrix R and translation matrix T are obtained. After calibration, the coordinates of the fire hydrant are (2500, 5.25, 2.5) and the coordinates of the display screen are (2500, 5.25, 2.5). The registration errors are 0.08 meters and 0.09 meters, respectively, both meeting the accuracy requirement of ±0.2 meters.

[0096] The level 2 and 3 point cloud data registered using the ICP algorithm are batch-converted to a reference 3D Cartesian coordinate system to achieve complete coordinate alignment of level 3, level 2, and level 1 data at the same spatial location. Following a hierarchical progression from local to regional to global, a mapping of physical attributes is established to clarify the relationships between data of the same attribute at different levels.

[0097] Traffic flow attributes: The speed data of individual vehicles at the third level is aggregated into the average speed of the area at the second level, and the average speed of the area at the second level is aggregated into the average speed of the entire area at the first level.

[0098] Environmental attributes: Local environmental data at the third level are aggregated into the average value of the second-level region, and the average value of the second-level region is aggregated into the average value of the entire first level.

[0099] Vehicle status attributes: These are only associated with the corresponding second-level region and serve as supplementary data for regional event characteristics.

[0100] The completed spatial location matching results are integrated with physical attribute association rules to generate a visualized hierarchical data mapping table. For example, the mapping relationship of the 2500-meter coordinate point is: vehicle position collected by ultrasonic sensor at level 3, vehicle speed collected by geomagnetic sensor → average vehicle speed in area 5 at level 2, smoke concentration in area 5 → average vehicle speed and average temperature across the entire area at level 1, thus completing the establishment of hierarchical data mapping relationships for all spatial points.

[0101] Step 4.2 involves comparing and analyzing the data at each level after establishing the mapping relationship. This extracts data differences at different scale levels for the same spatial location, as well as data differences for the same physical attribute at different monitoring dimensions. Correlation analysis is used to identify the coupling effect of data at each level in terms of spatial scale and physical attribute, thus obtaining the detailed fluctuation characteristics of the tunnel's operating status. Specifically, this includes: clarifying the calculation object and calculating the difference between third-level and second-level data for the same spatial location and physical attribute. The calculation includes two parts: absolute difference and relative deviation. Taking a 2500-meter coordinate point as an example, the local vehicle speed collected at the third level is 40 km / h, and the regional average vehicle speed calculated at the second level is 60 km / h. Therefore, the absolute difference between the two is -20 km / h. The relative deviation is calculated by first taking the absolute value of -20 km / h to get 20 km / h, and then using 20 ÷ 60 × 100% ≈ 33.3%. For example, the local smoke concentration at this location, measured at the third-level stratum, is 0.6 mg / m³, while the regional smoke concentration calculated at the second-level stratum is 0.3 mg / m³. The absolute difference between the two is 0.3 mg / m³. The relative deviation is calculated by first taking the absolute value of 0.3 mg / m³ and then dividing it by 0.3 ÷ 0.3 × 100% = 100%. This calculation method allows for the accurate extraction of data differences between different strata for the same spatial location and physical attribute.

[0102] Pearson correlation analysis was used to calculate the correlation coefficients of different physical properties at the same spatial scale. A correlation coefficient threshold of 0.8 was set; a correlation coefficient ≥ 0.8 was considered strong coupling. The correlation analysis formula is as follows: In the formula: The correlation coefficient is... These are sample values ​​of temperature data. This is the average of the temperature data. These are sample values ​​of smoke concentration data. This represents the average value of the smoke concentration data. To determine the sample size, 100 sets of data were collected continuously. An index, for example, is calculated from 100 sets of temperature and smoke concentration data at a 2500-meter location, yielding... =0.92, which is greater than 0.8, indicating a strong coupling effect between temperature and smoke concentration; calculations based on vehicle speed and traffic density data yielded... =-0.88 The absolute value ≥ 0.8 indicates a strong negative coupling effect between vehicle speed and traffic density; the lower the vehicle speed, the higher the traffic density. Combining the data difference and coupling effect analysis results, we extract the microscopic fluctuation characteristics reflecting abnormal tunnel operation and set specific judgment thresholds:

[0103] Detailed characteristics of traffic flow: The speed of vehicles at a single spatial location at level 3 drops sharply by ≥50km / h compared to the average speed of vehicles in the level 2 area, or the traffic density increases by ≥40 vehicles / km compared to the regional average.

[0104] Environmental micro-characteristics: The concentration of level 3 smoke at a single spatial location increases by ≥0.3 mg / m³ compared to the regional average, or the temperature increases by ≥10℃ compared to the regional average;

[0105] Coupling-type microscopic characteristics: Temperature and smoke concentration rise synchronously; temperature rises ≥ 5℃ and smoke concentration rises ≥ 0.2 mg / m³ 3 Or, a sudden drop in vehicle speed of ≥40km / h and an increase in traffic density of ≥30 vehicles / km; based on the above thresholds, examples such as a sudden drop in vehicle speed of 55km / h and an increase in smoke concentration of 0.4mg / m³ at a 2500m point are extracted. 3 The microscopic fluctuation characteristics of a 12°C temperature rise can accurately capture early signals of potential events in the tunnel.

[0106] Step 4.3: Based on the data deviation direction and magnitude reflected by the micro-fluctuation characteristics, calculate and generate dynamic calibration coefficients corresponding to each spatial location and each physical attribute. Specifically, this includes: calculating and generating dynamic calibration coefficients corresponding to each spatial location and each physical attribute based on the data deviation direction and magnitude reflected by the extracted micro-fluctuation characteristics. The specific implementation process is as follows:

[0107] The direction of deviation is clearly defined: higher data is considered a positive deviation, and lower data is considered a negative deviation. Using the second-level data as the baseline, the deviation magnitude is calculated using the following formula: In the formula: The deviation range, This is the original data at the third level. For secondary-level baseline data, for example, the vehicle speed collected by the geomagnetic sensor at the 2500-meter point is 40 km / h for Level 3, and 60 km / h for the secondary area baseline vehicle speed, with a negative deviation direction and a deviation amplitude A = 20 km / h; the smoke concentration (Level 3) is 0.6 mg / m³. 3 The baseline value for the secondary region is 0.3 mg / m³. 3 The deviation direction is positive, and the deviation amplitude A = 0.3 mg / m 3 .

[0108] Dynamic calibration coefficient calculation: A formula for calculating the dynamic calibration coefficient is set, and the coefficient value is adjusted according to the deviation direction: positive deviation coefficient < 1, negative deviation coefficient > 1. Formula: In the formula: For dynamic calibration coefficients, This is the second-level baseline data. Using the original data at the third level, and combining it with the example above, the following calculations are performed:

[0109] Vehicle speed calibration coefficient: K=1+(60-40)÷60≈1.33;

[0110] Smoke concentration calibration coefficient: K=1+(0.3-0.6)÷0.3=0.0; Note: In actual application, the lower limit of the coefficient is set to 0.1 to avoid data failure caused by the coefficient being 0, so it is adjusted to K=0.1.

[0111] Temperature calibration coefficient: Level 3 temperature at 2500-meter point is 38℃, Level 2 reference temperature is 26℃: K=1+(26-38)÷26≈0.54.

[0112] Calibration coefficient matching: The calculated dynamic calibration coefficients are bound to the corresponding spatial locations, such as the x-axis coordinate of 2500 meters, and physical attributes to generate a spatial location-physical attribute-calibration coefficient lookup table. For example, 2500 meters-vehicle speed-K=1.33; 2500 meters-smoke concentration-K=0.1; 2500 meters-temperature-K=0.54, ensuring that each spatial location and each physical attribute has a unique corresponding dynamic calibration coefficient.

[0113] Step 4.4 involves applying the dynamic calibration coefficients to the preprocessed high-quality multi-source data to perform a second iterative calibration on the completed environmental temperature and humidity compensation calibration results. This eliminates residual errors caused by the uneven spatial distribution and temporal variations of environmental factors. The mapping, identification, generation, and calibration processes are repeated until the data deviation converges, resulting in iteratively optimized multi-source data. Specifically, this includes using the high-quality multi-source data after environmental temperature and humidity compensation calibration as a basis, and conducting a second calibration using dynamic calibration coefficients. The specific calibration method involves multiplying the temperature and humidity compensated data by the dynamic calibration coefficients corresponding to the location and physical attributes. The resulting data... This refers to the data after secondary calibration. Taking the coordinate point at 2500 meters inside the tunnel as an example, the vehicle speed after temperature and humidity compensation is 38 kilometers per hour, corresponding to a dynamic calibration coefficient of 1.33. The vehicle speed after secondary calibration is approximately 50.54 kilometers per hour (38 × 1.33 ≈ 50.54 km / h). The smoke concentration after temperature and humidity compensation is 0.58 milligrams per cubic meter, corresponding to a dynamic calibration coefficient of 0.1. The smoke concentration after secondary calibration is approximately 0.58 × 0.1 = 0.058 milligrams per cubic meter. The temperature after temperature and humidity compensation is 37 degrees Celsius, corresponding to a dynamic calibration coefficient of 0.54. The temperature after secondary calibration is approximately 19.98 degrees Celsius (37 × 0.54 ≈ 19.98 degrees Celsius).

[0114] The criterion for determining data deviation convergence is that the relative deviation between the data after secondary calibration and the secondary-level reference data does not exceed 0.5%. The specific calculation method for the relative deviation is as follows: In the formula: This represents the relative deviation of the data after secondary calibration. This is the data after secondary calibration. This serves as the baseline data for the second-level hierarchy. Taking the vehicle speed data at the aforementioned 2500-meter coordinate point as an example, the vehicle speed after secondary calibration is 50.54 kilometers per hour, while the baseline vehicle speed for the second-level hierarchy is 60 kilometers per hour. Calculated in the above manner, the relative deviation is approximately 15.77%, which exceeds the 0.5% convergence threshold. Therefore, it is determined that the current data has not met the convergence standard and the iteration process needs to be repeated.

[0115] Following the complete process of updating data mapping relationships in step 4.1, conducting data difference analysis and coupling effect identification in step 4.2, recalculating dynamic calibration coefficients in step 4.3, and performing secondary data calibration in step 4.4, iterative operations are performed repeatedly. After each iteration, the data deviation is recalculated according to the above relative deviation calculation method until the relative deviation is reduced to within 0.5%. In practical application scenarios, if the decrease in data deviation is less than 1% after 10 consecutive iterations, the data can be determined to have reached a convergence state, avoiding infinite iteration. After completing the iterative calibration of all spatial locations and all physical attribute data within the tunnel, the final iteratively optimized multi-source data is output. This data completely eliminates residual errors caused by uneven spatial distribution and continuous temporal changes of environmental factors, and the data accuracy is improved by no less than 95% compared to the preprocessing stage.

[0116] In this embodiment of the invention, a mapping relationship is established by matching the spatial location and associating the physical attributes of data at different levels in a multi-scale data field. The differences in the data are compared and analyzed, and the coupling effect between spatial scale and physical attributes, as well as the microscopic fluctuation characteristics of the tunnel operation status, are identified. A dynamic calibration coefficient is generated based on the data deviation, and the temperature and humidity compensation calibration results are iteratively optimized using this coefficient until the data deviation converges. Therefore, this invention overcomes the technical problems of lack of correlation mapping in multi-scale data, inability to analyze coupling effects, difficulty in identifying microscopic fluctuation characteristics, and residual errors in sensor calibration and insufficient data accuracy caused by uneven spatiotemporal distribution of environmental factors. This further eliminates residual data errors, improves the accuracy and reliability of multi-source data, and achieves accurate capture of the microscopic characteristics of the tunnel operation status.

[0117] In a preferred embodiment of the present invention, step 5 above may include:

[0118] Step 5.1 involves feature-level fusion of the iteratively optimized multi-source data. This involves associating and matching video image features, sensor numerical features, and platform-recorded text features from the same spatiotemporal cross-section, enabling data from different modalities to corroborate each other at the feature level and form a fused feature set. Specifically, this includes: using the tunnel's three-dimensional Cartesian coordinate system as a benchmark, setting a spatiotemporal cross-section division standard, dividing the tunnel into 10-meter sections (500 sections for a 5000-meter tunnel), and setting a time node of 100 milliseconds to ensure that data within the same spatiotemporal cross-section has temporal synchronization and spatial correlation, avoiding data confusion between different spatiotemporals; extracting features such as vehicle outline, vehicle trajectory, vehicle posture, and accident scene details from images captured by high-definition video cameras, setting the video frame extraction frequency to be synchronized with the time nodes at 100 milliseconds per frame. Frame images retain only effective features related to vehicles and the environment, eliminating irrelevant background interference to ensure feature accuracy. From the iteratively optimized multi-source data, numerical features such as vehicle speed and traffic density from geomagnetic sensors, smoke concentration from smoke sensors, ambient temperature from temperature sensors, and vehicle position and collision severity from ultrasonic sensors are extracted. All values ​​are retained to two decimal places to ensure data accuracy. From the records of the tunnel control platform, textual features such as traffic flow statistics and vehicle passage records including vehicle license plates and passage times are extracted. The textual information is converted into standardized values ​​or identifiers, such as unique identification codes for vehicle license plates and timestamps for passage times. According to the principle of the same spatiotemporal section and the same physical object, the three types of features are associated and matched, with the association matching accuracy set at ±0.2 meters and ±10 milliseconds to ensure that different modal features correspond to the same monitoring object.

[0119] Taking a 2500-meter spatial cross-section and the 1000th millisecond as an example, the vehicle trajectory and body posture features extracted from the high-definition video of this cross-section are compared with the vehicle speed (45.98 km / h) and traffic density (35 vehicles / km) from the geomagnetic sensor and the smoke concentration (0.058 mg / m³) from the smoke sensor at this cross-section. 3 The ambient temperature of the temperature sensor is 19.98℃, the vehicle position of the ultrasonic sensor is x=2500.0 meters, y=5.25 meters, z=2.5 meters, and the vehicle identification numbers 001 and 002 recorded by the platform at this time point are associated and bound to ensure that various features corroborate each other and are consistent. The features after the association and matching of all spatiotemporal sections are integrated to form a complete and comprehensive fusion feature set. Each spatiotemporal section corresponds to a set of fusion features, including three types of features: video images, sensor values, and platform records.

[0120] Step 5.2 involves accident identification analysis of the fused feature set. This is achieved by analyzing the coordinated changes of multiple parameters, including sudden changes in vehicle trajectory, sudden drops in vehicle speed, abnormal traffic density, increased smoke concentration, and rising ambient temperature, to detect whether a traffic accident has occurred within the tunnel. Specifically, this includes setting thresholds for the coordinated changes of multiple parameters based on actual tunnel operation. All thresholds are specific numerical values; meeting any two or more thresholds is sufficient to preliminarily determine a traffic accident. The specific thresholds are as follows: Traffic flow parameters: vehicle speed drops by ≥40 km / h compared to the historical average speed at the same cross-section; traffic density increases by ≥30 vehicles / km compared to the historical average traffic density at the same cross-section; Environmental parameters: smoke concentration increases by ≥0.3 mg / m³ compared to the historical average concentration at the same cross-section. 3 The ambient temperature is ≥8℃ higher than the historical average temperature of the same section; the vehicle status parameters show a sudden change angle of ≥30 degrees in the vehicle's trajectory; the ultrasonic sensor detects a collision degree of ≥5cm as vehicle deformation. Note: The historical average data is the average value of the data after iterative optimization of 600 time nodes in the past 10 minutes of this section; the fusion feature set of each spatiotemporal section is analyzed point by point to extract the changes in three types of parameters: traffic flow, environment, and vehicle status, and compared with the set thresholds to determine whether the accident judgment conditions are met.

[0121] Taking a 2500-meter cross-section and the 1200th millisecond as an example, accident detection analysis was conducted: the historical average vehicle speed at this cross-section was 60 km / h, and the current optimized speed is 15 km / h. A sudden speed drop of 45 km / h ≥ 40 km / h satisfies the traffic flow parameter threshold; the historical average smoke concentration at this cross-section is 0.2 mg / m³. 3 The current smoke concentration after iterative optimization is 0.55 mg / m³. 3 The increase was 0.35 mg / m³. 3 ≥0.3mg / m 3 The environmental parameter thresholds are met; the historical average temperature of this section is 25℃, and the temperature after the current iteration optimization is 34℃, an increase of 9℃ ≥ 8℃, which meets the environmental parameter thresholds; the sudden change angle of the vehicle's motion trajectory is 45 degrees ≥ 30 degrees, and the ultrasonic wave detected a vehicle deformation of 8cm ≥ 5cm, which meets the vehicle state parameter thresholds; this section meets 4 judgment thresholds, far exceeding any 2 or more judgment criteria, thus accurately detecting a traffic accident at this spatiotemporal section.

[0122] Step 5.3: Upon detecting a traffic accident, perform correlation analysis on the fused feature set to extract the core features of the accident. These core features include: obtaining the precise location of the accident through cross-validation of spatial localization results from multi-source data; comprehensively determining the accident type based on vehicle posture identified from video images, collision shock waves detected by sensors, and vehicle deformation sensed by ultrasonic waves; statistically analyzing the number of vehicles involved and their vehicle type distribution by comparing video target detection with the number of triggers from geomagnetic sensors; calculating the vehicle collision angle based on continuous frame changes in vehicle trajectory; extracting the smoke concentration change trend based on time-series data from smoke sensors; and analyzing upstream and downstream geomagnetic sensors. Using traffic density data detected by video surveillance, the extent of traffic congestion in the surrounding area affected by the accident is calculated to obtain the core characteristics of the accident. Specifically, this includes: using cross-validation of spatial positioning results from multi-source data to ensure the accuracy of the accident location; and extracting three sets of positioning data from the high-definition video camera (GPS coordinates x1=2500.1 m, y1=5.25 m, z1=2.5 m), geomagnetic sensor coordinates (x2=2499.9 m, y2=5.25 m, z2=2.5 m), and ultrasonic sensor coordinates (x3=2500.0 m, y3=5.25 m, z3=2.5 m) at the spatiotemporal section corresponding to the accident.

[0123] The positioning error was eliminated by averaging, and the precise location of the accident was finally obtained as (2500.0 meters, 5.25 meters, 2.5 meters), with a positioning accuracy controlled within ±0.1 meters. Combining video image features and sensor detection features, the accident type was comprehensively determined, and specific judgment criteria were set. Through continuous frames of high-definition video, the vehicle posture was identified as frontal collision, side collision, rollover, and rear-end collision. In this accident, the video showed that the two vehicles were facing each other head-on, and there were frontal impact marks on the body. The geomagnetic sensor detected a collision shock wave of 0.6g, and a collision shock wave ≥0.5g was set as a severe collision. The ultrasonic sensor detected that the deformation of both vehicles was 8cm, and the deformation ≥5cm was set as severe deformation. Based on the above criteria and in accordance with the accident type judgment criteria, for a frontal collision: facing each other head-on, collision shock wave ≥0.5g, and frontal deformation of the body ≥5cm, the accident type was determined to be a two-vehicle frontal collision accident, avoiding the problem of misjudging the accident type by a single video or a single sensor.

[0124] To ensure accurate statistical results, a method was employed that compared video target detection with the number of vehicles triggered by geomagnetic sensors. Target detection was performed on 10 consecutive frames of high-definition video from the accident scene, identifying the number of vehicles in each frame. The average of the 10 frames was taken as the video detection count. The 10 frames yielded 3 vehicles each, with an average of 3 vehicles. Within 100 milliseconds before and after the accident, geomagnetic sensors at the accident section and adjacent sections at 2490 meters and 2510 meters were triggered a total of 3 times; each trigger corresponded to one vehicle passing by. The number of vehicles detected was 3, which was exactly the same as the number of vehicles triggered by the geomagnetic sensor, confirming that there were 3 vehicles involved. The vehicle outline dimensions were extracted from high-definition video images, and the vehicle type identification criteria were set: sedans: length 4.5-5.0 meters, width 1.8-2.0 meters; trucks: length 6.0-8.0 meters, width 2.2-2.5 meters. The outline dimensions of 2 sedans were identified as 4.8 meters × 1.9 meters and 4.7 meters × 1.85 meters, respectively, and the outline dimensions of 1 truck were 7.0 meters × 2.3 meters. Finally, it was determined that there were 3 vehicles involved, with the vehicle types being 2 sedans and 1 truck.

[0125] Based on the continuous frame changes in vehicle trajectories, the collision angle is calculated. From 20 consecutive frames of high-definition video, the starting and ending coordinates of the trajectories of the two colliding vehicles are extracted. The starting point of the first car's trajectory is (2499.8 m, 5.2 m, 2.5 m), and the ending point is (2500.0 m, 5.25 m, 2.5 m). The starting point of the second truck's trajectory is (2500.2 m, 5.3 m, 2.5 m), and the ending point is (2500.0 m, 5.25 m, 2.5 m). The collision point (2... Using 500.0 meters, 5.25 meters, and 2.5 meters as the origin, the trajectory vectors of the two vehicles are calculated. The trajectory vector of the first vehicle is (2499.8-2500.0, 5.2-5.25) = (-0.2, -0.05), and the trajectory vector of the second vehicle is (2500.2-2500.0, 5.3-5.25) = (0.2, 0.05). By substituting the numerical values ​​into the formula for the angle between the trajectory vectors, the collision angle is finally obtained as 90 degrees, which accurately reflects the vehicle collision posture.

[0126] Based on the time-series data from the smoke sensors, the trend of smoke concentration change was extracted. Smoke sensor data was collected for 10 minutes after the accident, with one set collected every 100 milliseconds, for a total of 600 sets. The first 10 sets were selected as an example: 0.2 mg / m³. 3 0.3 mg / m 3 0.4 mg / m 3 0.5 mg / m 3 0.55mg / m 3 0.6 mg / m 3 0.62 mg / m 30.63 mg / m 3 0.64 mg / m 3 0.65mg / m 3 ; Calculate the slope of change. The trend of change is determined by calculating the slope of concentration change. Substitute the first 10 sets of data to calculate: (0.65mg / m 3 -0.2mg / m 3 ) ÷ (10 × 100 milliseconds) = 0.45 mg / m 3 ÷1000 milliseconds = 0.00045 mg / m³ 3 / millisecond; the slope is positive, and the concentration increases by ≥0.05mg / m² per 100milliseconds. 3 The study determined that the smoke concentration was continuously increasing, providing a basis for fire hazard investigation and emergency response.

[0127] Based on traffic density data from upstream and downstream geomagnetic sensors and video detection, the congestion range was calculated. The accident section at 2500 meters was selected, along with five upstream sections at 2450 meters, 2460 meters, 2470 meters, 2480 meters, and 2490 meters, and five downstream sections at 2510 meters, 2520 meters, 2530 meters, 2540 meters, and 2550 meters. Traffic density data after the accident was collected at each section. A traffic density ≥ 50 vehicles / km was defined as congested, and below 50 vehicles / km was considered normal. At the upstream sections, the traffic densities at 2450 meters, 2460 meters, 2470 meters, 2480 meters, and 2490 meters were 55 vehicles / km, 53 vehicles / km, and 51 vehicles / km, respectively. The upstream congestion rate was 48 vehicles / km, 46 vehicles / km, with the first three sections (2450-2470m) experiencing congestion. Downstream sections, at 2510m, 2520m, 2530m, 2540m, and 2550m, had traffic densities of 60 vehicles / km, 58 vehicles / km, 56 vehicles / km, 49 vehicles / km, and 47 vehicles / km, respectively, with the first three sections (2510-2530m) experiencing congestion. The upstream congestion range was 2500m-2450m=50m, and the downstream congestion range was 2530m-2500m=30m, for a total congestion range of 50m+30m=80m. This precise determination of the accident's impact range provides data support for traffic control.

[0128] After completing all the above extraction steps, the complete core features of this accident are obtained, including: the precise location of the accident (2500.0 meters, 5.25 meters, 2.5 meters), the type of accident, the number and type distribution of the vehicles involved, the collision angle of the vehicles, the trend of smoke concentration changes, and the range of traffic congestion around the accident, forming complete core feature data of the accident.

[0129] In this embodiment of the invention, feature-level fusion is performed on iteratively optimized multi-source data. Video image features, sensor numerical features, and platform recorded text features from the same spatiotemporal section are correlated and matched to form a fused feature set. Accident identification and analysis are performed through multi-parameter collaborative changes. When an accident is detected, core features such as the precise location, type, vehicle information, collision angle, smoke concentration change trend, and congestion range of the accident are extracted through multi-source data cross-validation and correlation analysis. Therefore, this overcomes the technical problems of traditional tunnel accident monitoring, which relies on a single data source, resulting in low reliability, high false judgment rate, inaccurate accident location and type identification, and inability to comprehensively and accurately extract the core features of the accident. Thus, it achieves the technical effect of realizing mutual verification and collaborative judgment of multimodal data, improving the accuracy and comprehensiveness of traffic accident identification, fully extracting key accident features, and providing a reliable basis for tunnel emergency response and traffic control.

[0130] In a preferred embodiment of the present invention, step 6 above may include:

[0131] Step 6.1 involves standardizing and encapsulating the core characteristics of the accident, generating structured accident characteristic information messages according to the data interface specifications of the emergency command center. This includes: clarifying the data interface specifications of the emergency command center, setting the fixed format, field length, data type, and value range of the structured message, and ensuring that all fields use fixed-length encoding to guarantee interface compatibility.

[0132] The message header has a fixed length of 20 bytes, including a 4-byte message identifier, an 8-byte sending timestamp accurate to milliseconds, a 4-byte message length, and a 4-byte checksum; the message body has a fixed length of 120 bytes, containing 6 core feature fields in sequence, each with a fixed length of 20 bytes, and all data types are string-numeric features converted to strings according to a fixed format; the message tail has a fixed length of 10 bytes, including a 6-byte sending device number and a 4-byte end identifier.

[0133] The extracted core features of the incident are mapped one by one to the corresponding fields in the message body, and then standardized to ensure that the data format meets the interface requirements. The specific mapping and transformation are as follows:

[0134] Field 1 Accident Location: Convert the precise coordinates of the accident (2500.0 meters, 5.25 meters, 2.5 meters) into a string formatted as x=2500.0, y=5.25, z=2.5. If the string is less than 20 bytes, pad it with spaces. If the string exceeds 20 bytes, truncate the core part, which is exactly 18 bytes in this case, and pad it with 2 spaces.

[0135] Field 2 Accident Type: Converts the head-on collision of two vehicles into a fixed string, padding with spaces if less than 20 bytes;

[0136] Field 3: Vehicle information involved: Convert 3 vehicles, 2 cars, and 1 truck into a format string of quantity=3, car=2, truck=1, padding with spaces if less than 20 bytes;

[0137] Field 4 Collision Angle: Convert 90 degrees to a format string of collision angle = 90°, padding with spaces if less than 20 bytes;

[0138] Field 5: Smoke concentration trend: will continue to rise, slope 0.00045 mg / m³ 3 / milliseconds converted to smoke trend=climb, slope=0.00045 format string, padded with spaces if less than 20 bytes;

[0139] Field 6 Congestion Range: Convert the upstream 50m, downstream 30m, total 80m to the format string upstream=50m, downstream=30m, total=80m, padding with spaces if less than 20 bytes.

[0140] Message checksum calculation: To ensure that messages are not lost or tampered with during transmission, a message checksum is calculated. The checksum is calculated as follows: sum the ASCII values ​​of all characters in the message header and the message body, then divide by 1000 and take the remainder. The remainder is the checksum. Specific calculation example: The sum of the ASCII values ​​in the message header is 12500, and the sum of the ASCII values ​​of all fields in the message body is 37500. The total sum = 12500 + 37500 = 50000. The checksum = 50000 ÷ 1000 = 50, and the remainder is 0. Therefore, the checksum is 0000.

[0141] Structured message generation: Following the order of message header → message body → message footer, all fields are concatenated and integrated to generate a complete structured accident feature information message. An example message is as follows: Message header: Message identifier, timestamp, message length, and checksum; Message body: x=2500.0, y=5.25, z=2.5 Number of head-on collisions between two vehicles=3, sedan=2, truck=1 Collision angle=90° Smoke trend=rising, slope=0.00045 Upstream=50m, Downstream=30m, Total=80m; Message footer: Sending device number and end identifier. The generated message fully conforms to the emergency command center interface specification.

[0142] Step 6.2: The structured accident feature information message is pushed to the tunnel emergency command center in real time through a dedicated communication network. This allows the emergency command center to simultaneously obtain core feature data such as the precise location of the accident, accident type, number and type distribution of vehicles involved, vehicle collision angle, smoke concentration trend, and surrounding traffic congestion range. Specifically, this includes: based on the generated structured accident feature information message, pushing it to the tunnel emergency command center in real time through a dedicated communication network to ensure that the emergency command center simultaneously obtains complete core accident feature data, guaranteeing the real-time performance and stability of the push. An industrial Ethernet network is used as the dedicated communication network, and network transmission parameters are configured to ensure efficient and stable data transmission. The specific parameter settings are as follows:

[0143] Transmission rate: 100Mbps, ensuring fast message transmission;

[0144] Transmission delay threshold: ≤50 milliseconds, to avoid message transmission delays;

[0145] Transmission encryption method: AES-128 encryption is used to prevent data from being tampered with or leaked;

[0146] Network redundancy configuration: Two backup communication links are set up. When the primary link is interrupted, the system will automatically switch to the backup link within 10 milliseconds to ensure uninterrupted push notifications.

[0147] The generated structured accident characteristic information message is pushed to the core receiving server of the tunnel emergency command center through a configured dedicated communication network. The push frequency is once every 100 milliseconds, and it is pushed three times consecutively to ensure that the command center receives it successfully. The push confirmation rule is set. After receiving the message, the emergency command center returns a confirmation message within 300 milliseconds, which includes a successful reception identifier and a message identifier. If the sending end does not receive a confirmation message within 300 milliseconds, a retransmission mechanism is automatically triggered. The retransmission interval is 300 milliseconds, 400 milliseconds, and 500 milliseconds, and a maximum of 3 retransmissions. If no confirmation is received, an alarm is triggered immediately.

[0148] Specific example: In this push, the main link transmission delay is 35 milliseconds. The emergency command center returns an acknowledgment message within 200 milliseconds. After receiving the acknowledgment, the sending end stops retransmission and completes the message push. If the main link is interrupted, it switches to the backup link within 10 milliseconds, and completes the push with a delay of 42 milliseconds, ensuring that the emergency command center synchronously obtains the core feature data of the accident. After receiving the message, the emergency command center automatically parses the message, extracts each core feature field, and compares it with the original core features of the accident from the sending end. The comparison accuracy is 100%, and the field content is completely consistent. After verification, the data is synchronized to the visualization interface of the emergency command platform for commanders to view in real time.

[0149] Step 6.3: Based on the core characteristics of the accident pushed to the emergency command center, provide the accident location coordinates and accessible routes for dispatching rescue vehicles, provide the congestion range and accident severity for traffic control decisions, and provide the accident type and involved vehicle information for on-site handling plan formulation, thereby achieving precise data support for emergency dispatch and traffic control. Specifically, based on the core characteristics of the accident pushed to the emergency command center, provide specific basis for dispatching rescue vehicles, making traffic control decisions, and formulating on-site handling plans to achieve precision in emergency dispatch and traffic control. Combine the precise location of the accident and the congestion range to calculate the optimal rescue route and determine the rescue vehicle dispatch plan. There are two rescue stations set up in the tunnel. Rescue station 1 is located at the tunnel entrance and is equipped with 2 rescue vehicles and 6 rescue personnel. Rescue station 2 is located in the middle of the tunnel at x=2500 meters and is equipped with 3 rescue vehicles and 9 rescue personnel. The accident location is at x=2500.0 meters. Priority is given to dispatching the nearest rescue station.

[0150] Calculate the distance from each rescue station to the accident location. The distance is calculated as the absolute value of the x-coordinate of the rescue station and the x-coordinate of the accident. Specifically: the distance from rescue station 1 to the accident location is |0 - 2500.0| = 2500 meters; the distance from rescue station 2 to the accident location is |2500 - 2500.0| = 0 meters. Determine the optimal route for dispatching rescue vehicles from rescue station 2. The optimal route is 0 meters in a straight line from rescue station 2 to the accident location, without detours, while avoiding the congestion area upstream of 2450-2500 meters and downstream of 2500-2530 meters. The dispatch route is from the exit of rescue station 2 to the accident location at x = 2500.0 meters, with a travel time ≤ 1 minute. Based on the calculation results, send a dispatch instruction to rescue station 2, specifying the dispatch of 2 rescue vehicles and 6 rescue personnel, carrying demolition tools and first aid equipment, to proceed to the accident scene along the optimal route. Simultaneously, synchronize the accident location coordinates and the on-site situation to ensure the rapid arrival of the rescue vehicles.

[0151] Based on the extent of congestion and the severity of the accident, traffic control areas and methods are determined. According to the core characteristics of the accident, severity criteria are established: minor accidents: ≤2 vehicles involved, no severe deformation, congestion range ≤50 meters; moderate accidents: 3-5 vehicles involved, severe deformation, congestion range 50-100 meters; severe accidents: ≥6 vehicles involved, violent collision, congestion range ≥100 meters. This accident involved 3 vehicles, with vehicle deformation of 8cm (severe deformation) and a congestion range of 80 meters, and is therefore classified as a moderate accident requiring partial traffic control.

[0152] Based on the congestion area, control zones are set. The control zone range = congestion area + safety buffer distance, with the safety buffer distance set at 10 meters. The upstream control zone extends from 2450 meters - 10 meters = 2440 meters to 2500 meters, from 10 meters before the congestion start point to the accident location. The downstream control zone extends from 2500 meters to 2530 meters + 10 meters = 2540 meters, from the accident location to 10 meters after the congestion end point. The total control zone is from 2440 meters to 2540 meters, with a length of 2540 - 2440 = 100 meters.

[0153] Based on the controlled area and accident situation, specific control measures are set: upstream control, with traffic control signs set at the 2440-meter mark to prohibit vehicles from entering the controlled area and guide vehicles to slow down and stop before the 2440-meter mark to wait for passage instructions; downstream control, with traffic control signs set at the 2540-meter mark to prohibit vehicles from leaving the controlled area to avoid secondary accidents; control information and detour prompts are released in real time through the traffic information display screen inside the tunnel to ensure that passing vehicles are informed in a timely manner.

[0154] Based on the accident type and information about the vehicles involved, a targeted on-site handling plan is developed. According to the accident type and the number of vehicles involved, the configuration standards for handling personnel and equipment are set. Through the above steps, the core characteristic data of the accident is accurately pushed to the emergency command center, providing a comprehensive and specific basis for rescue dispatch, traffic control, and on-site handling.

[0155] In this embodiment of the invention, by standardizing and encapsulating the core characteristics of an accident and generating structured accident characteristic information messages according to specifications, and pushing the structured messages to the tunnel emergency command center in real time through a dedicated communication network, and by providing corresponding data support for rescue dispatch, traffic control, and on-site handling based on the core characteristics of the accident, the technical means overcome the technical problems of non-standardized accident information, delayed transmission, unstructured data, and lack of accurate data support for command and decision-making in traditional tunnel emergency response. Thus, it achieves the technical effect of realizing efficient and synchronous push of accident information, providing accurate data support for emergency dispatch and traffic control, and improving the efficiency and rationality of emergency response and management of tunnel emergencies.

[0156] like Figure 2 As shown, embodiments of the present invention also provide a tunnel event feature extraction system based on multi-source data fusion, comprising:

[0157] The acquisition module is used to deploy various types of sensing devices at set intervals within the tunnel and connect to the tunnel management platform to collect multi-source heterogeneous data in real time.

[0158] The calibration module is used to perform time synchronization calibration on multi-source heterogeneous data to obtain synchronized data; it also performs data cleaning and error correction on the synchronized data, and performs environmental temperature and humidity compensation calibration on the detection data of geomagnetic sensors and ultrasonic sensors to obtain high-quality pre-processed multi-source data.

[0159] The module is used to extract the spatial coordinate information of traffic information display screens and fixed fire-fighting facilities deployed in the tunnel. Based on the physical location and electromagnetic wave coverage characteristics of the spatial coordinate information of the traffic information display screens and fixed fire-fighting facilities in the tunnel section, a dynamic monitoring spatial envelope is constructed. The preprocessed multi-source data is divided into several levels according to the spatial scale and physical attributes of the dynamic monitoring spatial envelope to form a multi-scale data field.

[0160] The iterative module is used to establish the mapping relationship between different levels of data in a multi-scale data field. By analyzing the coupling effect of each level of data in terms of spatial scale and physical properties, it identifies the micro-fluctuation characteristics that reflect the tunnel operation status, generates dynamic calibration coefficients, and iteratively optimizes the environmental temperature and humidity compensation calibration results in the preprocessed high-quality multi-source data through the dynamic calibration coefficients to obtain the iteratively optimized multi-source data.

[0161] The extraction module is used to perform feature-level fusion of the iteratively optimized multi-source data, identify traffic accidents in tunnels through correlation analysis, and extract the core features of the accidents.

[0162] The processing module is used to push the core characteristics of the accident to the tunnel emergency command center in real time, providing data support for emergency dispatch and traffic control.

[0163] It should be noted that this system is a system corresponding to the above method. All implementation methods in the above method embodiments are applicable to this embodiment and can achieve the same technical effect.

[0164] Embodiments of the present invention also provide a computing device, including: a processor and a memory storing a computer program, wherein the computer program, when executed by the processor, performs the method described above. All implementations in the above method embodiments are applicable to this embodiment and can achieve the same technical effects.

[0165] Embodiments of the present invention also provide a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the method described above. All implementations in the above method embodiments are applicable to this embodiment and can achieve the same technical effects.

[0166] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for extracting tunnel event features based on multi-source data fusion, characterized in that, The method includes: Various types of sensing devices are deployed at set intervals inside the tunnel and connected to the tunnel management platform to collect multi-source heterogeneous data in real time. Time synchronization calibration is performed on multi-source heterogeneous data to obtain synchronized data; data cleaning and error correction are performed on synchronized data; environmental temperature and humidity compensation calibration is performed on the detection data of geomagnetic sensor and ultrasonic sensor to obtain high-quality preprocessed multi-source data. Spatial coordinate information of traffic information display screens and fixed fire-fighting facilities deployed in the tunnel is extracted. Based on the physical location and electromagnetic wave coverage characteristics of the spatial coordinate information of traffic information display screens and fixed fire-fighting facilities in the tunnel section, a dynamic monitoring spatial envelope is constructed. The preprocessed multi-source data is divided into several levels according to the spatial scale and physical attributes of the dynamic monitoring spatial envelope to form a multi-scale data field. Establish the mapping relationship between different levels of data in a multi-scale data field. By analyzing the coupling effect of each level of data in terms of spatial scale and physical properties, identify the micro-fluctuation characteristics that reflect the tunnel operation status. Based on this, generate dynamic calibration coefficients. Then, iteratively optimize the environmental temperature and humidity compensation calibration results in the preprocessed high-quality multi-source data using the dynamic calibration coefficients to obtain the iteratively optimized multi-source data. The multi-source data after iterative optimization is fused at the feature level, and traffic accidents in tunnels are identified through correlation analysis, and the core features of the accidents are extracted. The core characteristics of the accident are pushed to the tunnel emergency command center in real time, providing data support for emergency dispatch and traffic control.

2. The tunnel event feature extraction method based on multi-source data fusion according to claim 1, characterized in that, Various types of sensing devices are deployed at predetermined intervals within the tunnel and connected to the tunnel management platform to collect multi-source heterogeneous data in real time, including: Sensing devices are deployed at predetermined intervals within the tunnel. These devices include high-definition video cameras, geomagnetic sensors, smoke sensors, temperature sensors, and ultrasonic sensors. Simultaneously, they are connected to the tunnel management platform to collect real-time traffic flow statistics and vehicle passage records. The high-definition video cameras capture vehicle movement and accident scene footage; the geomagnetic sensors collect data on vehicle presence, speed, and traffic density; the smoke sensors collect data on smoke concentration caused by accidents; the temperature sensors collect data on ambient temperature changes; and the ultrasonic sensors assist in identifying vehicle positions and collision severity. The tunnel management platform provides traffic flow statistics and vehicle passage records, collectively forming multi-source heterogeneous data.

3. The tunnel event feature extraction method based on multi-source data fusion according to claim 2, characterized in that, Time synchronization calibration is performed on multi-source heterogeneous data to obtain synchronized data; The synchronized data undergoes data cleaning and error correction. Environmental temperature and humidity compensation calibration is applied to the detection data from the geomagnetic and ultrasonic sensors to obtain preprocessed, high-quality multi-source data, including: Add a unified timestamp to the multi-source heterogeneous data, perform time synchronization calibration, eliminate the time difference in data acquisition between various sensing devices and the tunnel control platform, and obtain synchronized data; Outliers and missing values ​​in the synchronized data are removed and filled in, and errors are corrected for data affected by the tunnel environment, resulting in preliminarily cleaned data. For the detection data collected by the geomagnetic sensor and ultrasonic sensor in the preliminary cleaned data, combined with the synchronously collected ambient temperature and humidity data, ambient temperature and humidity compensation calibration is performed to eliminate the influence of ambient temperature and humidity on the detection accuracy, and obtain high-quality multi-source data after preprocessing.

4. The tunnel event feature extraction method based on multi-source data fusion according to claim 3, characterized in that, Extract the spatial coordinate information of traffic information display screens and fixed fire-fighting facilities deployed in the tunnel. Based on the physical location and electromagnetic wave coverage characteristics of the spatial coordinate information of the traffic information display screens and fixed fire-fighting facilities in the tunnel section, construct a dynamic monitoring spatial envelope. The preprocessed multi-source data is divided into several levels according to the spatial scale and physical attributes of the dynamic monitoring spatial envelope, forming a multi-scale data field, including: Extract the spatial coordinates of traffic information display screens and fixed fire-fighting facilities deployed inside the tunnel to obtain the precise physical location data of each facility within the tunnel cross section; Based on the physical location of traffic information display screens and fixed fire-fighting facilities in the tunnel cross section and their electromagnetic wave coverage characteristics, a dynamic monitoring spatial envelope covering the tunnel monitoring area is constructed. The preprocessed high-quality multi-source data is divided into several levels according to the spatial scale and physical attributes of the dynamic monitoring spatial envelope, so that the data at each level corresponds to the spatial range and monitoring attributes of the dynamic monitoring spatial envelope, thus forming a multi-scale data field.

5. The tunnel event feature extraction method based on multi-source data fusion according to claim 4, characterized in that, A mapping relationship is established between data at different levels in a multi-scale data field. By analyzing the coupling effect of data at each level in terms of spatial scale and physical properties, the microscopic fluctuation characteristics reflecting the tunnel's operational status are identified. Based on this, dynamic calibration coefficients are generated. The environmental temperature and humidity compensation calibration results in the preprocessed high-quality multi-source data are iteratively optimized using these dynamic calibration coefficients to obtain the iteratively optimized multi-source data, including: Spatial location matching and physical attribute association are performed on data at different levels in a multi-scale data field to establish mapping relationships between data at each level. After establishing the mapping relationship, the data at each level are compared and analyzed to extract the data differences at different scale levels for the same spatial location, as well as the data differences for the same physical attribute at different monitoring dimensions. Correlation analysis is used to identify the coupling effect of data at each level in terms of spatial scale and physical attribute, and to obtain the micro-fluctuation characteristics of the tunnel operation status. Based on the direction and magnitude of the data deviation reflected by the microscopic fluctuation characteristics, dynamic calibration coefficients corresponding to each spatial location and each physical property are calculated and generated. The dynamic calibration coefficients are applied to the preprocessed high-quality multi-source data, and the completed environmental temperature and humidity compensation calibration results are subjected to a second iteration calibration to eliminate residual errors caused by the uneven spatial distribution and temporal changes of environmental factors. The above mapping, identification, generation and calibration are repeated until the data deviation converges, and the iteratively optimized multi-source data is obtained.

6. The tunnel event feature extraction method based on multi-source data fusion according to claim 5, characterized in that, The iteratively optimized multi-source data is fused at the feature level, and correlation analysis is used to identify traffic accidents within tunnels, extracting the core features of the accidents, including: Feature-level fusion is performed on the iteratively optimized multi-source data, and video image features, sensor numerical features and platform recorded text features under the same spatiotemporal section are associated and matched, so that data of different modalities can corroborate each other at the feature level and form a fused feature set. Accident identification analysis is performed on the fused feature set. By analyzing the coordinated changes of multiple parameters such as sudden changes in vehicle trajectory, sudden drop in vehicle speed, abnormal traffic flow density, increase in smoke concentration, and increase in ambient temperature, it can detect whether a traffic accident has occurred in the tunnel. When a traffic accident is detected, correlation analysis is performed on the fused feature set to extract the core features of the accident. These core features include: obtaining the precise location of the accident through cross-validation of spatial positioning results from multi-source data; comprehensively determining the accident type based on vehicle posture identified from video images, collision shock waves detected by sensors, and vehicle deformation sensed by ultrasonic waves; statistically analyzing the number of vehicles involved and their vehicle type distribution by comparing the number of video target detections with the number of triggers from geomagnetic sensors; calculating the vehicle collision angle based on continuous frame changes in vehicle trajectories; extracting the trend of smoke concentration changes based on time-series data from smoke sensors; and calculating the surrounding traffic congestion range affected by the accident based on traffic density data from upstream and downstream geomagnetic sensors and video detections, thus obtaining the core features of the accident.

7. The tunnel event feature extraction method based on multi-source data fusion according to claim 6, characterized in that, The core characteristics of the accident are pushed to the tunnel emergency command center in real time to provide data support for emergency dispatch and traffic control, including: The core characteristics of the accident are standardized and encapsulated, and structured accident characteristic information messages are generated in accordance with the data interface specifications of the emergency command center. Structured accident characteristic information messages are pushed to the tunnel emergency command center in real time through a dedicated communication network, enabling the emergency command center to simultaneously obtain core characteristic data such as the precise location of the accident, the type of accident, the number and type distribution of vehicles involved, the angle of vehicle collision, the trend of smoke concentration changes, and the range of traffic congestion in the surrounding area. Based on the core characteristics of the accident pushed out, the system provides the emergency command center with the coordinates of the accident location and the accessible route for dispatching rescue vehicles, provides the traffic control decision-making with the scope of congestion and the severity of the accident, and provides the on-site response plan with the accident type and information on the vehicles involved, thus achieving precise data support for emergency dispatch and traffic control.

8. A tunnel event feature extraction system based on multi-source data fusion, wherein the system implements the method as described in any one of claims 1 to 7, characterized in that, include: The acquisition module is used to deploy various types of sensing devices at set intervals within the tunnel and connect to the tunnel management platform to collect multi-source heterogeneous data in real time. The calibration module is used to perform time synchronization calibration on multi-source heterogeneous data to obtain synchronized data. Data cleaning and error correction are performed on the synchronized data, and environmental temperature and humidity compensation calibration is performed on the detection data of the geomagnetic sensor and the ultrasonic sensor to obtain high-quality multi-source data after preprocessing. The module is used to extract the spatial coordinate information of traffic information display screens and fixed fire-fighting facilities deployed in the tunnel. Based on the physical location and electromagnetic wave coverage characteristics of the spatial coordinate information of the traffic information display screens and fixed fire-fighting facilities in the tunnel section, a dynamic monitoring spatial envelope is constructed. The preprocessed multi-source data is divided into several levels according to the spatial scale and physical attributes of the dynamic monitoring spatial envelope, forming a multi-scale data field. The iterative module is used to establish the mapping relationship between different levels of data in a multi-scale data field. By analyzing the coupling effect of each level of data in terms of spatial scale and physical properties, it identifies the micro-fluctuation characteristics that reflect the tunnel operation status, generates dynamic calibration coefficients, and iteratively optimizes the environmental temperature and humidity compensation calibration results in the preprocessed high-quality multi-source data through the dynamic calibration coefficients to obtain the iteratively optimized multi-source data. The extraction module is used to perform feature-level fusion of the iteratively optimized multi-source data, identify traffic accidents in tunnels through correlation analysis, and extract the core features of the accidents. The processing module is used to push the core characteristics of the accident to the tunnel emergency command center in real time, providing data support for emergency dispatch and traffic control.

9. A computing device, characterized in that, include: One or more processors; A storage device for storing one or more programs, which, when executed by one or more processors, cause the one or more processors to implement the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program that, when executed by a processor, implements the method as described in any one of claims 1 to 7.

Images