A Computer Vision-Based Method and System for Monitoring Highway Hazardous Liquid Leakage Accidents

Abstract

This invention discloses a computer vision-based method and system for monitoring highway hazardous chemical liquid leaks. The method includes: installing markings configured for hazardous chemical vehicle identification on vehicles transporting hazardous chemicals; the markings include optical markings configured to improve the visual recognition of hazardous chemical vehicles and digital markings configured to enable encrypted storage and rapid retrieval of vehicle information; deploying a lightweight recognition model configured to detect the status of the optical and digital markings in real time within an existing road monitoring system; visually locating the projected pattern on the road surface using a camera, and calculating the real-time vehicle location using map coordinates to supplement GPS positioning blind spots in tunnels and mountainous areas; calculating the probability of an accident based on the marking status to determine whether the vehicle is suspected of being involved in an accident; and triggering an accident response process when an accident is suspected. This significantly improves accident response efficiency.

Description

Technical Field

[0001] This invention belongs to the field of computer vision technology, and more specifically, relates to a method and system for monitoring highway hazardous chemical liquid leakage accidents based on computer vision. Background Technology

[0002] In the field of road transportation, although hazardous chemical transport accidents are sporadic events, the damage they cause is extremely serious. Once a hazardous liquid leaks, the liquid will rapidly flow into the surrounding environment through the road drainage system, polluting water bodies, soil, and the atmosphere. The treatment of such pollution is technically challenging and costly, especially in environmentally sensitive areas such as water source protection zones, where the influx of hazardous chemical spills can lead to local water supply difficulties, affecting residents' lives and ecological safety. Currently, there are numerous types of chemicals, which can be classified into multiple categories according to their hazards, numbering in the tens of thousands, posing a significant challenge to the monitoring of hazardous chemical accidents.

[0003] Traditional hazardous chemical monitoring primarily relies on the intermittent deployment of automatic online water quality monitoring systems at locations such as road drainage pipes and ditches. These systems monitor parameters like water temperature, pH, conductivity, and turbidity to determine the presence of hazardous chemical leaks. However, this technology has significant shortcomings in practical applications: First, automatic online water quality monitoring equipment cannot cover all types of hazardous chemicals, making comprehensive monitoring difficult when faced with a large number of chemicals of varying properties. Second, the equipment has a detection limit; when the concentration of a hazardous chemical solution decreases to a certain level after mixing with rainwater, the equipment cannot accurately identify it, leading to monitoring gaps. Third, equipment operation and maintenance face numerous challenges. Water-sensitive areas are often located in mountainous regions, reservoirs, and other inaccessible locations, resulting in delayed equipment maintenance. Furthermore, these precision instruments require specialized technicians to operate, and the highway system often lacks such personnel. Additionally, the cost of regularly adding specialized reagents is high. Finally, the automatic online water quality monitoring equipment and its associated power and communication facilities are expensive, increasing the construction and operation costs of highways.

[0004] From the perspective of monitoring timeliness, traditional technologies rely on changes in water quality parameters to determine accidents. However, it takes time for hazardous chemical liquids to leak and flow into the drainage system, resulting in a lag in monitoring results and making it difficult to provide timely warnings in the early stages of an accident. For example, when a hazardous chemical vehicle overturns or leaks on a highway, traditional systems may need several hours or even longer to detect the problem through abnormal water quality parameters. By then, the contamination area may have expanded, increasing the difficulty of emergency response. In addition, highway drainage systems are widely distributed, and the intermittently installed monitoring equipment cannot cover all road sections, creating monitoring blind spots. If an accident occurs outside the coverage area of ​​the equipment, it cannot be detected in time. Summary of the Invention

[0005] The computer vision-based method and system for monitoring hazardous chemical spills on highways proposed in this invention are designed to address the aforementioned challenges. This solution utilizes existing highway surveillance cameras, constructs a recognition model to determine whether a hazardous chemical accident has occurred, and uses the camera positions to pinpoint the accident location. This method enables real-time monitoring of hazardous chemical vehicles, allowing for timely detection of anomalies in the early stages of an accident. It improves the efficiency and accuracy of monitoring and alarming for hazardous chemical spills, reduces the amount of hazardous chemical liquids that need to be disposed of after an accident, and provides technical support for environmental protection along highways.

[0006] To address the aforementioned deficiencies or improvement needs of existing technologies, as a first aspect of this invention, the present invention provides a computer vision-based method for monitoring highway hazardous chemical liquid spill accidents, comprising:

[0007] S1. Installing markings configured for identifying hazardous chemical vehicles on vehicles carrying and transporting hazardous chemicals; the markings include optical markings configured to enhance the visual recognition of hazardous chemical vehicles and digital markings configured to enable encrypted storage and rapid retrieval of vehicle information.

[0008] S2. Based on the existing road monitoring system, deploy and load a lightweight recognition model configured to detect the status of optical and digital signs in real time; at the same time, bind the RFID card reader and camera signal at the toll station entrance to complete the comparison and verification of vehicle electronic identity and visual information;

[0009] S3. Verify the validity of the tag by periodically scanning with a camera and cross-validating the data with an RFID reader; use a camera to visually locate the projected pattern on the road surface and combine it with map coordinates to calculate the real-time location of the vehicle to supplement the GPS positioning blind spots in tunnels and mountainous areas; and calculate the probability of an accident based on the tag status to determine whether the vehicle is suspected of being involved in an accident.

[0010] S4. When an incident is determined to be suspected, the incident response process is triggered.

[0011] Furthermore, the optical markings in S1 include fluorescent reflective composite stickers and dynamic light and shadow projection devices;

[0012] The fluorescent reflective composite sticker is affixed to both sides and the tail of the can. The pattern is the national standard dangerous goods symbol and includes a fluorescent background. The sticker adopts a three-layer composite structure, including a base layer, a fluorescent layer and a reflective layer. The base layer is made of weather-resistant material; the fluorescent layer contains rare earth activated fluorescent pigments; and the reflectivity of the reflective layer is ≥500cd / lx / m².

[0013] The dynamic light and shadow projection device is installed at the rear of the vehicle and powered by the vehicle. It projects dangerous goods symbols onto the road surface in real time. The projected pattern of the dynamic light and shadow projection device is dynamically adjusted with the vehicle speed: when the vehicle speed is >60km / h, the projection is reduced to avoid interfering with following vehicles, and when the vehicle speed is ≤60km / h, it is enlarged to no more than 1.5m × 1.5m to form a visual ground marking.

[0014] Furthermore, the digital identifier in S1 includes dynamic QR code identifiers and passive RFID identifiers;

[0015] The QR code dynamic identifier has a built-in NFC chip and is detachably set below the license plate. The QR code includes information on the type of goods and transportation route for each transportation task of the transportation company. The RFID passive identifier is set on the windshield of the vehicle, and the vehicle VIN code and dangerous goods transportation permit number are written in the label.

[0016] Furthermore, the model for real-time detection of optical markers in S2 is as follows:

[0017] For fluorescent reflective stickers, the marking area is extracted by threshold segmentation in the HSV color space:

[0018] ,

[0019] in, These are the pixel's hue, saturation, and brightness values. The threshold range for color hue selection; This represents the saturation filtering threshold range; The threshold range for filtering brightness values; Represents a set of pixels that satisfy a specific HSV range condition; Represents a single pixel in an image;

[0020] Dynamically adjust via adaptive threshold The parameters complete the separation of the fluorescent yellow background and the reflective border;

[0021] For the projected pattern at the rear of the vehicle, through Optical flow method for calculating pixel motion vector field function :

[0022] ,

[0023] In the formula, Indicates time ,coordinate Pixel brightness at; where , , respectively pixels in The speed of movement in the direction; Indicates a time interval; Represents a time variable;

[0024] For each pixel The optical flow constraint condition expands to:

[0025] ,

[0026] In the formula, Represents the image gradient; Represents the time derivative; for pixels Given a neighborhood W, construct a least-squares optimization problem:

[0027] ,

[0028] In the formula, For pixels A pixel in the neighborhood W; by... Taking the partial derivative and setting it to 0 yields the average motion vector within the neighborhood. ;

[0029] Then, by quantifying the proportion of "dynamic light spot pixels" in the road surface projection area, the validity of the projection is determined, i.e., the probability of projection existence is calculated. ,as follows:

[0030] ,

[0031] Represents pixel coordinates, Indicates the area of ​​interest; This is an indicator function that outputs 1 if the condition is met, and 0 otherwise. The threshold for motion speed; Indicates the magnitude of the speed of motion; express Total number of pixels in the region; when And vehicle speed An alarm for projection failure is triggered at that time.

[0032] Furthermore, the detection model for the digital identifier status is as follows:

[0033] For QR code labels, the corner points of the QR code are located using Hough transform, and a perspective transformation matrix is ​​constructed. Correcting the image:

[0034] ,

[0035] In the formula, Represents the pixel coordinates in the original image; Indicates the corrected target pixel coordinates; This is the perspective transformation matrix;

[0036] Simultaneously define module contrast The mean grayscale difference of the black and white modules after correction:

[0037] ,

[0038] In the formula, Represents the number of module logarithms. This indicates the module's index. Indicates the first The average gray level of each black module Indicates the first The average gray level of each white module; when The larger the value, the more significant the difference between black and white modules; when If the QR code is deemed blurry, the QR code is automatically deemed invalid. Indicates the threshold for fuzzy judgment;

[0039] For passive RFID tags, the VIN code and dangerous goods transport permit number information in the vehicle's RFID tag are collected by the RFID card reader at the toll station entrance. At the same time, the vehicle's optical tag is identified by the camera, and the tag information is directly matched with the data collected by the card reader to verify the correspondence between the VIN code, permit number and vehicle, and to determine the legality of the tag. If the rule is met, 1 is output, indicating that the vehicle can pass; if the rule is not met, 0 is output, indicating that the vehicle cannot enter the highway.

[0040] Furthermore, the specific process in S2 for determining whether a vehicle is suspected of being involved in an accident by calculating the accident probability based on the identifier status is as follows:

[0041] Constructing the identifier state feature vector ,in Indicates the integrity of the fluorescent reflective sticker; A clear quantification of the QR code; Indicates the effective proportion of the dynamic projection of the road surface;

[0042] Calculating accident probability based on Bayes' theorem:

[0043] ,

[0044] in, Indicates the prior probability of the accident; This indicates that the status was as follows when the accident occurred. The probability of; This indicates that the flag state is in all scenarios. The probability of; Indicates the state of a given identifier At that time, the probability of the accident occurring is the ultimate goal to be calculated;

[0045] Calculate Then, a decision threshold is needed. To determine whether an accident warning has been triggered; when If the system determines that an incident is suspected, it will automatically report it to the monitoring center and mark the location; otherwise, it will maintain normal monitoring status.

[0046] Furthermore, the decision threshold The method for determining it is as follows:

[0047] Finding the optimal threshold using maximum likelihood estimation This ensures the highest possible match between the predicted results and the actual accident labels.

[0048] ,

[0049] In the formula, Indicates the first The actual accident labels for each sample; Indicates the first The identifier feature vector of each sample; Indicates a given and At that time, the prediction was The probability of.

[0050] Furthermore, the accident response process in S4 is specifically as follows:

[0051] Once the warning is triggered, the system automatically retrieves data from other nearby cameras, RFID readers, and vehicle sensors to cross-verify the accident risk. If the verification confirms an anomaly, the warning level is upgraded from "suspected" to "confirmed," and a detailed accident risk report is generated.

[0052] Based on the severity of the accident and its potential impact, the early warning is divided into different levels: Level 1 Emergency, Level 2 Serious, and Level 3 Attention. The Level 1 Emergency warning will immediately notify the public security, fire, emergency management departments and hazardous chemical disposal teams; the Level 2 Serious warning will simultaneously notify the highway administration and environmental protection departments; and the Level 3 Attention warning will be pushed to transportation companies and toll station managers, requiring them to strengthen monitoring.

[0053] After receiving the early warning information, the emergency command center activates the emergency plan according to the type and location of the accident; dispatches nearby resources, including fire trucks, hazardous chemical spill response equipment, and ambulances, to the scene; at the same time, it coordinates with highway administration to implement traffic control and guide the accident vehicles to exit the highway or park in the emergency lane to avoid secondary accidents.

[0054] After the incident is handled, a review and analysis of the entire event data is conducted to summarize the causes of the incident, response efficiency, and handling effectiveness. The analysis results are then fed back to the model optimization module to update the Bayesian model parameters and decision thresholds. .

[0055] As a second aspect of the present invention, a computer vision-based highway hazardous chemical liquid leakage accident monitoring system is also provided, comprising:

[0056] A hazardous materials identification placement unit is used to place and install identification configured for the identification of hazardous materials vehicles on vehicles carrying and transporting hazardous materials; the identification includes optical identification configured to improve the visual recognition of hazardous materials vehicles and digital identification configured to enable encrypted storage and rapid retrieval of vehicle information.

[0057] The lightweight model deployment unit is used to deploy and load a lightweight recognition model configured to detect the status of optical and digital signs in real time within an existing road monitoring system; at the same time, it binds the RFID card reader and camera signal at the toll station entrance to complete the comparison and verification of vehicle electronic identity and visual information.

[0058] The identification status verification unit is used to verify the validity of identification by periodically scanning with a camera and cross-verifying data with an RFID reader; it uses a camera to visually locate the road surface projection pattern and combines it with map coordinates to calculate the real-time location of the vehicle to supplement the GPS positioning blind spots in tunnels and mountainous areas; and it calculates the probability of accidents based on the identification status to determine whether the vehicle is suspected of being involved in an accident.

[0059] The incident response triggering unit is used to trigger the incident response process when an incident is determined to be suspected.

[0060] As a third aspect of the invention, a computer-readable storage medium is also provided, on which a computer program is stored, which is executed by a processor of any step of the computer vision-based method for monitoring road hazardous chemical liquid spills.

[0061] In summary, compared with the prior art, the above-described technical solutions conceived by this invention can achieve the following beneficial effects:

[0062] 1. The computer vision-based method for monitoring hazardous chemical liquid leaks on highways, as described in this invention, constructs a dual identification system of visualization and digitization by deploying optical and digital tags on vehicles. The three-layer composite fluorescent reflective sticker uses rare-earth activated fluorescent pigments and high-reflectivity materials, maintaining high visibility even in complex lighting environments such as strong light and tunnels. The dynamic projection device intelligently adjusts the projection size according to vehicle speed, forming dynamic warning signs on the road surface and enhancing the visual characteristics of hazardous chemical vehicles. The digital identification system composed of QR codes and RFID enables encrypted storage and rapid retrieval of transportation task information, providing digital evidence for transportation compliance verification and solving the problems of easy soiling and low information reading efficiency of traditional tags.

[0063] 2. The computer vision-based method for monitoring hazardous chemical liquid leaks on highways, as described in this invention, achieves intelligent detection of label status through a lightweight model. It utilizes HSV color segmentation and adaptive thresholding technology to accurately extract fluorescent sticker areas, quantifies the motion features of projected pixels based on optical flow to determine projection validity, evaluates QR code clarity using perspective transformation and contrast analysis, and cross-verifies vehicle electronic identity and visual information through RFID and camera data. This technical solution overcomes the environmental limitations of traditional single-label detection, maintaining a high recognition rate even in rainy, foggy, and bumpy conditions, providing reliable data support for accident early warning and effectively reducing the risk of false alarms and missed alarms.

[0064] 3. The computer vision-based method for monitoring hazardous chemical liquid leaks on highways, as described in this invention, dynamically optimizes the early warning threshold through maximum likelihood estimation, enabling accurate prediction of suspected accidents. Upon triggering an early warning, the system automatically links multi-source data for verification and initiates a tiered response mechanism, efficiently dispatching emergency resources and implementing traffic control. Post-event analysis optimizes model parameters. This solution supplements GPS positioning blind spots in tunnels and mountainous areas through visual positioning, upgrading hazardous chemical transportation monitoring from single-identity identification to a closed-loop management process encompassing "status detection - risk prediction - emergency response," significantly improving accident response efficiency, building an intelligent safety defense line for highway hazardous chemical transportation, and helping to reduce the incidence of transportation safety accidents. Attached Figure Description

[0065] Figure 1 This is a flowchart of a computer vision-based method for monitoring highway hazardous chemical liquid leaks according to an embodiment of the present invention.

[0066] Figure 2 This is a system unit diagram of an embodiment of the present invention. Detailed Implementation

[0067] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0068] Example 1

[0069] Please refer to Figure 1 This embodiment 1 provides a computer vision-based method for monitoring highway hazardous chemical liquid spill accidents, including:

[0070] S1. Installing markings configured for identifying hazardous chemical vehicles on vehicles carrying and transporting hazardous chemicals; the markings include optical markings configured to enhance the visual recognition of hazardous chemical vehicles and digital markings configured to enable encrypted storage and rapid retrieval of vehicle information.

[0071] S2. Based on the existing road monitoring system, deploy and load a lightweight recognition model configured to detect the status of optical and digital signs in real time; at the same time, bind the RFID card reader and camera signal at the toll station entrance to complete the comparison and verification of vehicle electronic identity and visual information;

[0072] S3. Verify the validity of the tag by periodically scanning with a camera and cross-validating the data with an RFID reader; use a camera to visually locate the projected pattern on the road surface and combine it with map coordinates to calculate the real-time location of the vehicle to supplement the GPS positioning blind spots in tunnels and mountainous areas; and calculate the probability of an accident based on the tag status to determine whether the vehicle is suspected of being involved in an accident.

[0073] S4. When an incident is determined to be suspected, the incident response process is triggered.

[0074] This embodiment 1 further elaborates on the above steps.

[0075] (1) Hazardous chemical signage arrangement

[0076] To ensure the effective identification of hazardous chemical transport vehicles in the road environment, it is necessary to systematically deploy a signage system on the vehicles that combines visual warning and information storage functions. This system includes two types: optical signs and digital signs, which work together to improve vehicle identification and information traceability efficiency.

[0077] The first step in setting up the optical signage is to apply the fluorescent reflective composite sticker. Flat areas on both sides and the rear of the can are selected as the application sites. Before application, the surface is thoroughly cleaned with a specialized cleaner to remove oil, dust, and other impurities, ensuring a clean base surface to enhance adhesion between the sticker and the can. The sticker employs a three-layer composite structure: the base layer uses weather-resistant materials to adapt to different climates, preventing aging and cracking during high and low temperature changes; the fluorescent layer contains rare-earth activated fluorescent pigments, giving the sticker a bright orange color that maintains high visual visibility both day and night; the reflective layer has a reflectivity ≥500cd / lx / m² and uses microprism reflective materials, efficiently reflecting incident light under low-light conditions such as rain, fog, and nighttime, ensuring clear identification by monitoring equipment or vehicles behind. The sticker surface is printed with national standard hazardous materials symbols, and the fluorescent background contrasts sharply with the reflective layer. To prevent edge curling and maintain effectiveness, the sticker edges are also bound.

[0078] The dynamic light and shadow projection device is fixed to the chassis crossbeam at the rear of the vehicle by a bracket. Adjusting the bracket's tilt angle ensures the lens is vertically downwards, pointing directly at the road surface, guaranteeing accurate projection of the pattern onto the road behind the vehicle. The device's power cable is connected to the vehicle's battery using a waterproof connector to prevent short circuits caused by rain or moisture. To enable dynamic adjustment of the projected pattern according to vehicle speed, a vehicle speed signal acquisition device is installed in the driver's cab. This device acquires real-time vehicle speed data via a CAN bus and transmits the data to the projection device's control chip. When the vehicle speed exceeds 60 km / h, the control chip automatically reduces the size of the projected pattern to avoid strong light interference with the vision of drivers behind. When the vehicle speed is 60 km / h or less, the projected pattern enlarges to no more than 1.5 meters x 1.5 meters, forming a clear and conspicuous visual mark on the ground, enhancing the warning effect when the vehicle is traveling at low speeds.

[0079] The digital identification system aims for accurate information storage and rapid retrieval. The dynamic QR code identifier has a built-in NFC chip and is detachably installed below the license plate. Its carrier is made of waterproof and scratch-resistant material, covered with a transparent protective film to prevent distortion and blurring caused by vehicle bumps and rain. The QR code contains key information such as the type of goods and transportation route for each transport mission, facilitating rapid reading by monitoring systems or inspection personnel.

[0080] The passive RFID tag is installed inside the vehicle's windshield, ensuring stability without obstructing the driver's view. The tag contains the vehicle's VIN code and dangerous goods transport permit number, enabling encrypted storage and contactless, rapid reading of vehicle qualification information. After installation, a dedicated reader must be used to verify the information written on the RFID tag to ensure that the tag information matches the vehicle's actual qualifications.

[0081] The entire process of deploying hazardous chemical signage, from the visual warning function of optical signs to the information storage and traceability of digital signs, strictly adheres to installation specifications at every stage to ensure the functionality and reliability of the signs. By strategically placing these signs, not only can highway monitoring systems promptly identify hazardous chemical vehicles, but they also provide accurate vehicle information to traffic management departments and emergency rescue teams, laying a solid foundation for the safe management of hazardous chemical transportation and effectively improving safety and traceability during transport.

[0082] (2) Deployment of lightweight models

[0083] Deploying a lightweight recognition model in the existing road monitoring system requires the systematic integration of optical and digital sign detection modules, as well as the construction of a multi-dimensional information verification system at toll station entrances, to achieve real-time and accurate detection of the status of hazardous chemical vehicle identification.

[0084] 2.1 Optical Marking Inspection

[0085] To detect fluorescent reflective stickers, an HSV color space threshold segmentation method is first embedded in the edge computing nodes of the monitoring system. During initial configuration, specific hue, saturation, and brightness threshold ranges are set according to the color standards for hazardous materials labeling in China to extract the label area. Simultaneously, an adaptive thresholding method is deployed. By analyzing image samples from different time periods (such as midday strong light and rainy night low light), the threshold parameters are adjusted in real time. For example, the lower limit of the brightness threshold is automatically increased on rainy days to ensure accurate segmentation of the fluorescent yellow background and reflective border.

[0086] For detecting the projected pattern at the rear of a vehicle, an optical flow method module is integrated into the video stream processing pipeline of the monitoring system. Specifically, by analyzing the motion characteristics of pixels in consecutive video frames, the motion speed of the pixels at different times is calculated. A reasonable motion speed threshold is set, and the effectiveness of the projection is determined by calculating the proportion of dynamic light spot pixels in the road projection area. When the proportion of dynamic light spot pixels is lower than the set threshold and the vehicle is in motion, a projection failure alarm mechanism is triggered to ensure timely alarm response.

[0087] In a specific preferred embodiment, the model for real-time detection of optical markers is as follows:

[0088] For fluorescent reflective stickers, the marking area is extracted by threshold segmentation in the HSV color space:

[0089] ,

[0090] in, These are the pixel's hue, saturation, and brightness values. The threshold range for color hue selection; This represents the saturation filtering threshold range; The threshold range for filtering brightness values; Represents a set of pixels that satisfy a specific HSV range condition; Represents a single pixel in an image;

[0091] Dynamically adjust via adaptive threshold The parameters complete the separation of the fluorescent yellow background and the reflective border;

[0092] For the projected pattern at the rear of the vehicle, through Optical flow method for calculating pixel motion vector field function :

[0093] ,

[0094] In the formula, Indicates time ,coordinate Pixel brightness at; where , , respectively pixels in The speed of movement in the direction; Indicates a time interval; Represents a time variable;

[0095] For each pixel The optical flow constraint condition expands to:

[0096] ,

[0097] In the formula, Represents the image gradient; Represents the time derivative; for pixels Given a neighborhood W, construct a least-squares optimization problem:

[0098] ,

[0099] In the formula, For pixels A pixel in the neighborhood W; by... Taking the partial derivative and setting it to 0 yields the average motion vector within the neighborhood. ;

[0100] Then, by quantifying the proportion of "dynamic light spot pixels" in the road surface projection area, the validity of the projection is determined, i.e., the probability of projection existence is calculated. ,as follows:

[0101] ,

[0102] Represents pixel coordinates, Indicates the area of ​​interest; This is an indicator function that outputs 1 if the condition is met, and 0 otherwise. The threshold for motion speed; Indicates the magnitude of the speed of motion; express Total number of pixels in the region; when And vehicle speed An alarm for projection failure is triggered at that time.

[0103] 2.2 Digital Identification Detection

[0104] The deployment of the QR code identification detection module needs to be carried out in steps: First, the corner points of the QR code are located using a specific method, and the deformed image is corrected using a perspective transformation matrix, converting it into a standard rectangular area for subsequent recognition. Simultaneously, a module contrast index is defined to measure the grayscale difference between the black and white modules of the corrected QR code. A blur threshold is set; when the module contrast falls below this threshold, the QR code is considered blurry and invalid, ensuring the accuracy of information reading.

[0105] For passive RFID tags, RFID readers and cameras are hardware-linked at tollbooth entrances. Specifically, the RFID reader collects the VIN code and dangerous goods transport permit number from the vehicle's RFID tag, while the camera identifies the vehicle's optical identification. The system matches the tag information with the data collected by the reader in real time to verify the correspondence. If the information matches, the tag is deemed valid, and the vehicle is allowed to enter the highway; if the information does not match, the vehicle is denied entry, achieving strict verification of vehicle identity.

[0106] In a preferred embodiment, the specific method for QR code identification detection is as follows:

[0107] For QR code labels, the corner points of the QR code are located using Hough transform, and a perspective transformation matrix is ​​constructed. Correcting the image:

[0108] ,

[0109] In the formula, Represents the pixel coordinates in the original image; Indicates the corrected target pixel coordinates; This is the perspective transformation matrix;

[0110] Simultaneously define module contrast The mean grayscale difference of the black and white modules after correction:

[0111] ,

[0112] In the formula, Represents the number of module logarithms. This indicates the module's index. Indicates the first The average gray level of each black module Indicates the first The average gray level of each white module; when The larger the value, the more significant the difference between black and white modules; when If the QR code is deemed blurry, the QR code is automatically deemed invalid. This represents the threshold for fuzzy judgment.

[0113] (3) Identification status verification

[0114] Cross-verification of identification status is achieved through periodic scanning by cameras and data from RFID readers. Specifically, road surveillance cameras capture images of the optical markings (fluorescent reflective stickers, road surface projection patterns) of hazardous chemical vehicles at a preset frequency (e.g., 2 frames per second). Simultaneously, RFID readers deployed at tollbooth entrances or along the route read the digital information from the vehicle's RFID tags in real time. The system compares the visual features identified by the camera, such as the integrity of the sticker and the validity of the projection, with the vehicle identification information (VIN code, hazardous materials transport permit number) in the RFID tag. If the sticker is damaged, resulting in missing visual features, or if RFID data reading fails, the identification validity is determined to be abnormal, and an initial warning is triggered.

[0115] To address GPS signal blind spots in areas such as tunnels and mountainous regions, a camera-based method is used to visually locate vehicles projected onto road surfaces. The specific steps are as follows: First, the camera identifies the outline of hazardous material symbols projected onto the road surface. Combining this with the installation height and angle parameters of the projection device, a mapping relationship is established between the pixel coordinates of the projected pattern and the actual geographic coordinates. Then, through cross-view positioning using multiple cameras, the relative position of the vehicle in the two-dimensional image is calculated. Finally, using coordinate matching methods from the road electronic map, the pixel coordinates are converted into latitude and longitude coordinates. For example, when a vehicle enters a tunnel and GPS signal is lost, the system uses the position of the projected pattern captured by a camera inside the tunnel, combined with preset coordinate reference points within the tunnel, to calculate the vehicle's real-time position.

[0116] In the accident probability determination stage, the system first extracts key features from optical and digital markings. Image analysis technology is used to assess the integrity of fluorescent reflective stickers, calculate the contrast of the QR code's black and white modules, and analyze the effective display area of ​​the road surface projection. These features are then integrated into a marking status feature vector. Based on historical data, the system analyzes the differences in marking features during normal driving and accidents, constructing a probability model. The likelihood of an accident is estimated based on the current vehicle's marking features. Finally, the system learns from historical samples to determine a reasonable judgment standard. When the calculated accident probability exceeds this standard, the system determines that the vehicle has a suspected accident risk, immediately sends an alert to the monitoring center, and marks the vehicle's location; if the probability is below the standard, routine monitoring continues.

[0117] In a preferred embodiment, the specific process of determining whether a vehicle is suspected of being involved in an accident by calculating the accident probability based on the identification status is as follows:

[0118] Constructing the identifier state feature vector ,in Indicates the integrity of the fluorescent reflective sticker; A clear quantification of the QR code; Indicates the effective proportion of the dynamic projection of the road surface;

[0119] Calculating accident probability based on Bayes' theorem:

[0120] ,

[0121] in, Indicates the prior probability of the accident; This indicates that the status was as follows when the accident occurred. The probability of; This indicates that the flag state is in all scenarios. The probability of; Indicates the state of a given identifier At that time, the probability of the accident occurring is the ultimate goal to be calculated;

[0122] Calculate Then, a decision threshold is needed. To determine whether an accident warning has been triggered; when If the system determines that an incident is suspected, it will automatically report it to the monitoring center and mark the location; otherwise, it will maintain normal monitoring status.

[0123] The decision threshold The method for determining it is as follows:

[0124] Finding the optimal threshold using maximum likelihood estimation This ensures the highest possible match between the predicted results and the actual accident labels.

[0125] ,

[0126] In the formula, Indicates the first The actual accident labels for each sample; Indicates the first The identifier feature vector of each sample; Indicates a given and At that time, the prediction was The probability of.

[0127] (4) Accident response triggering

[0128] Once the system determines a vehicle's status as "suspected accident," it will immediately initiate the accident response process, forming a closed-loop emergency handling mechanism through multi-level verification, tiered early warning, resource scheduling, and post-mortem optimization.

[0129] Risk review and early warning upgrade: After an early warning is triggered, the system automatically retrieves data from other nearby cameras, RFID readers and vehicle sensors to cross-verify the accident risk; if the review confirms that there is an anomaly, the warning level will be upgraded from "suspected" to "confirmed", and a detailed report containing the accident time, location, vehicle information and risk type will be generated.

[0130] Tiered early warning and multi-departmental coordination: Based on the severity and potential impact of the accident, the system classifies early warnings into three levels. Level 1 Emergency Warning is for accidents that may cause serious consequences such as explosions or large-scale leaks. It immediately sends alerts to public security, fire, emergency management departments, and hazardous chemical disposal teams to ensure rapid deployment of professional rescue forces. Level 2 Severe Warning is applicable to medium-risk scenarios and simultaneously notifies highway administration and environmental protection departments to prepare for traffic control and environmental protection in advance. Level 3 Attention Warning is pushed to transportation companies and toll station managers, requiring them to strengthen real-time monitoring of the vehicles involved and, if necessary, guide the vehicles off the highway.

[0131] Emergency Resource Dispatch and Traffic Control: Upon receiving the early warning information, the emergency command center immediately activates the corresponding emergency plan based on the type of accident and on-site needs. It rapidly dispatches nearby fire trucks, hazardous chemical spill response equipment, ambulances, and other rescue resources to the scene, ensuring timely arrival of rescue supplies and personnel. Simultaneously, it coordinates with highway administration departments to implement traffic control measures, using electronic displays and lane closures to guide accident vehicles to exit the highway at the nearest exit or park in the emergency lane, and directs surrounding vehicles to detour, minimizing the risk of secondary accidents.

[0132] Data Review and Model Optimization: After an incident is handled, the system conducts an in-depth review of the entire event's data, analyzing the causes of the incident, the time taken in the response process, and the effectiveness of the measures taken. For example, it analyzes whether the early warning response was timely, whether the allocation of rescue resources was reasonable, and whether multi-departmental collaboration was smooth. The review results are fed back to the system's model optimization module to update the parameters of the Bayesian probability model and decision thresholds, further improving the accuracy of subsequent incident assessments and response efficiency, forming a complete closed loop of "monitoring-early warning-response-optimization".

[0133] Example 2

[0134] Please refer to Figure 2 This embodiment 2 provides a computer vision-based monitoring system for highway hazardous chemical liquid spill accidents, including:

[0135] A hazardous materials identification placement unit is used to place and install identification configured for the identification of hazardous materials vehicles on vehicles carrying and transporting hazardous materials; the identification includes optical identification configured to improve the visual recognition of hazardous materials vehicles and digital identification configured to enable encrypted storage and rapid retrieval of vehicle information.

[0136] The lightweight model deployment unit is used to deploy and load a lightweight recognition model configured to detect the status of optical and digital signs in real time within an existing road monitoring system; at the same time, it binds the RFID card reader and camera signal at the toll station entrance to complete the comparison and verification of vehicle electronic identity and visual information.

[0137] The identification status verification unit is used to verify the validity of identification by periodically scanning with a camera and cross-verifying data with an RFID reader; it uses a camera to visually locate the road surface projection pattern and combines it with map coordinates to calculate the real-time location of the vehicle to supplement the GPS positioning blind spots in tunnels and mountainous areas; and it calculates the probability of accidents based on the identification status to determine whether the vehicle is suspected of being involved in an accident.

[0138] The incident response triggering unit is used to trigger the incident response process when an incident is determined to be suspected.

[0139] Example 3

[0140] This embodiment 3 also provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it can implement any step of a computer vision-based method for monitoring road hazardous chemical liquid leakage accidents.

[0141] The computer-readable storage medium may include various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0142] For a description of the computer-readable storage medium provided in this application, please refer to the above method embodiments; further details will not be repeated here.

[0143] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A computer vision-based highway dangerous chemical liquid leakage accident monitoring method, characterized in that, Comprise: S1. Arranging an installation configured for dangerous goods vehicle identification on a vehicle loaded with dangerous goods; the installation comprises an optical mark configured to improve the visual recognition of the dangerous goods vehicle and a digital mark configured to achieve the encrypted storage and rapid reading of vehicle information; S2. Based on the existing road monitoring system, a lightweight identification model configured to detect the state of the optical mark and the digital mark in real time is deployed therein; at the same time, the RFID card reader and camera signal at the toll station entrance are bound to complete the comparison and verification of the vehicle electronic identity and visual information; S3. The validity of the mark is verified by periodic scanning of the camera and cross verification of the RFID card reader data; the real-time position of the vehicle is calculated by visual positioning of the road projection pattern by the camera combined with the map coordinates to supplement the positioning blind area of GPS in tunnels or mountainous areas; the accident probability calculation based on the state of the mark completes the determination of whether the vehicle is suspected of accident; S4. When the accident is suspected, the accident response process is triggered; The optical mark in S1 includes a fluorescent and reflective composite sticker and a dynamic light projection device; The model for real-time detection of the optical mark in S2 is: for the fluorescent and reflective composite sticker, the mark area is extracted by HSV color space threshold segmentation: ， wherein, is a pixel hue, saturation, lightness value; is a screening threshold interval for hue; is a screening threshold interval for saturation; is a screening threshold interval for lightness; denotes a set of pixels satisfying a certain HSV range condition; denotes a single pixel point in an image; Dynamic adjustment by adaptive threshold Parameters complete the segmentation of the fluorescent yellow background color and the reflective frame; With regard to the vehicle rear projection pattern, by Optical flow method to calculate pixel motion vector field function : ， wherein denotes the time instant , the coordinates of the pixel; wherein , are the motion velocities of the pixel in the directions , respectively; denotes the time interval; denotes the time variable; For each pixel The optical flow constraint equation expands to: ， wherein denotes the image gradient; denotes the time derivative; To the neighborhood W of the pixel a least squares optimization problem is constructed: ， In the formula, For pixels A pixel in the neighborhood W; by... Taking the partial derivative and setting it to 0 yields the average motion vector in the neighborhood. ; Then, the projection is judged to be effective or not by quantifying the proportion of "dynamic light spot pixels" in the projection area on the road surface, i.e. the existence probability of the projection is calculated As follows: ， represents the pixel coordinate, represents the region of interest; is an indicator function, output 1 when the condition is met, otherwise output 0; is the motion speed threshold value; represents the motion speed size; represents total number of pixels in the region; when and the vehicle speed trigger the projection failure alarm. 2.The computer vision-based highway dangerous chemical liquid leakage accident monitoring method according to claim 1, characterized in that, The fluorescent and reflective composite sticker is pasted on both sides and the tail of the tank, the pattern is the national standard dangerous goods symbol and includes a fluorescent background color, and adopts a three-layer composite sticker, including a base layer, a fluorescent layer and a reflective layer, the base layer is a weather-resistant material; the fluorescent layer adds rare earth activated fluorescent pigment; the reflective coefficient of the reflective layer is ≥500 cd / lx / m²; The dynamic light projection device is set at the tail of the vehicle, which is powered by the vehicle, which projects the dangerous goods symbol to the road surface in real time; the projection pattern of the dynamic light projection device is dynamically adjusted with the vehicle speed: when the vehicle speed is >60 km / h, the projection is reduced to avoid interference with the following vehicle, and when the vehicle speed is ≤60 km / h, the projection is enlarged to not more than 1.5 meters x 1.5 meters, forming a ground visual mark. 3.The computer vision-based highway dangerous chemical liquid leakage accident monitoring method according to claim 1, characterized in that, The digital mark in S1 includes a dynamic two-dimensional code and an RFID passive mark; The dynamic two-dimensional code includes the information of the type of goods and the transportation route of each transportation task of the transportation enterprise, and the RFID passive mark is set on the front windshield of the vehicle, and the label writes the vehicle VIN code and the dangerous goods transportation permit number. The detection model of the state of the digital mark in S2 is: 4.The computer vision-based highway dangerous chemical liquid leakage accident monitoring method according to claim 1, wherein, For the RFID passive mark, the VIN code and the dangerous goods transportation permit number information in the vehicle RFID tag are collected by the RFID card reader at the toll station entrance; at the same time, the vehicle optical mark is identified by the camera, and the label information is directly matched with the data collected by the card reader to verify the correspondence between the VIN code, the permit number and the vehicle, and to determine the legality of the mark; the rule-compliant output is 1, indicating that it can be released, and the rule-incompliant output is 0, indicating that it cannot be admitted to the highway. For two-dimensional code identification, the Hough transform is used to locate the corner points of the two-dimensional code, and a perspective transformation matrix is constructed Corrected image: ， In the formula, represents the pixel coordinates in the original image; represents the corrected target pixel coordinates; is a perspective transformation matrix; the module contrast is defined simultaneously is the average gray scale difference of the black and white modules after correction; ， wherein, represents the module log, represents the module index, represents the average gray level of the th black module, represents the average gray level of the th white module; When The greater, the more significant black and white module difference is; when The two-dimensional code is determined to be blurred, and at this time the two-dimensional code is determined to be invalid. The blurred determination threshold value is represented. The specific process of determining whether the vehicle is suspected of accident based on the state of the mark in S3 is: 5.The computer vision-based highway dangerous chemical liquid leakage accident monitoring method according to claim 1, wherein, ​ Constructing an identification state feature vector Wherein Indicate the completeness of the fluorescent reflective composite sticker; Indicate the clarity quantization value of the two-dimensional code; Indicate the effective proportion of the road surface dynamic projection; The accident probability is calculated based on a Bayesian formula: ， wherein, represents the accident prior probability; represents the probability that the identification state is when the accident occurs; represents the probability that the identification state is in all scenarios; represents the probability that the accident occurs given the identification state ; i.e. the target to be calculated ultimately. Calculate Then, a decision threshold is needed. To determine whether an accident warning has been triggered; When the system determines that it is a suspected accident, automatically reports to the monitoring center and marks the location, otherwise it maintains normal monitoring state. 6.The computer vision-based highway dangerous chemical liquid leakage accident monitoring method according to claim 5, characterized in that, The decision threshold The determination method is that: Finding the optimal threshold by maximum likelihood estimation Maximizing the match between the predicted outcome and the actual incident label: ， wherein, represents the actual incident label of the th sample; represents the identification feature vector of the th sample; represents the probability that the given and are predicted to be . 7.The computer vision-based highway dangerous chemical liquid leakage accident monitoring method according to claim 1, wherein, The accident response process in the S4 is specifically: After the early warning is triggered, the system automatically retrieves the data of other cameras, RFID card readers and vehicle-mounted sensors in the surrounding area, cross- verifies the accident risk, and if the review confirms that there is an anomaly, the early warning level is upgraded from "suspected" to "confirmed", and an accident risk report is generated; According to the severity and potential impact of the accident, the early warning is divided into different levels: first-level emergency, second-level serious and third-level attention; the first-level emergency early warning immediately notifies the public security, fire, emergency management departments and hazardous chemical substance disposal professional teams; The second-level serious early warning synchronously informs the highway administration and environmental protection departments; the third-level attention early warning is pushed to the transportation enterprises and toll station managers, requiring them to strengthen monitoring; After receiving the early warning information, the emergency command center starts the emergency plan according to the type and location of the accident; dispatches nearby resources including fire vehicles, hazardous chemical substance leakage treatment equipment and ambulances to the scene; at the same time, coordinates the highway administration to conduct traffic control and guide the accident vehicles to exit the highway or park in the emergency lane as soon as possible to avoid secondary accidents; After the accident is handled, the whole-process data of the event is analyzed, the accident cause, response efficiency and disposal effect are summarized, and the analysis result is fed back to the model optimization module to update the Bayesian model parameters and the decision threshold .

8. A computer vision-based highway dangerous chemical liquid leakage accident monitoring system, characterized in that, It includes: A hazardous chemical substance identification arrangement unit is configured to arrange and install an identification configured for hazardous chemical substance vehicle identification on a vehicle carrying and transporting hazardous chemicals; the identification includes an optical identification configured to improve the visual recognition of the hazardous chemical substance vehicle and a digital identification configured to realize the encrypted storage and rapid reading of vehicle information; A lightweight model deployment unit is configured to deploy and load a lightweight identification model configured to detect the state of the optical identification and the digital identification in real time in the existing road monitoring system; at the same time, the RFID card reader and camera signals at the entrance of the toll station are bound to complete the comparison and verification of the vehicle electronic identity and visual information; An identification state verification unit is configured to complete the identification validity verification through periodic scanning by the camera and cross-verification of the RFID card reader data; through visual positioning of the projection pattern on the road surface by the camera, the real-time position of the vehicle is calculated in combination with the map coordinates to supplement the positioning blind area of GPS in tunnels or mountainous areas; the accident probability is calculated based on the identification state to determine whether the vehicle is suspected of an accident; An accident response triggering unit is configured to trigger the accident response process when the accident is determined to be suspected; The optical identification in the hazardous chemical substance identification arrangement unit includes a fluorescent reflective composite sticker and a dynamic light projection device; The model for real-time detection of the optical identification in the lightweight model deployment unit is as follows: for the fluorescent reflective composite sticker, the identification area is extracted by threshold segmentation in the HSV color space: ， wherein, is a pixel hue, saturation, lightness value; is a screening threshold interval for hue; is a screening threshold interval for saturation; is a screening threshold interval for lightness value; denotes a set of pixels satisfying a certain HSV range condition; denotes a single pixel point in an image; by adaptive thresholding Parameters complete the segmentation of the fluorescent yellow background and the reflective frame; With regard to the vehicle rear projection pattern, by Optical flow method calculates the pixel motion vector field function : ， In the formula, denotes the time instant , the coordinate of the pixel; wherein , , respectively, the motion speed of the pixel in the direction; denotes the time interval; denotes the time variable; For each pixel The optical flow constraint expands to: ， wherein denotes the image gradient; denotes the time derivative; for a pixel of a neighborhood W, a least squares optimization problem is constructed: ， wherein is a pixel in the neighborhood W of the pixel ; the average motion vector within the neighborhood is obtained by taking the partial derivative with respect to and setting it to zero. Then, the projection is judged to be effective or not by quantifying the proportion of "dynamic light spot pixels" in the projection area on the road surface, i.e. the existence probability of the projection is calculated As follows: ， represents the pixel coordinate, represents the region of interest; is an indicator function, output 1 when the condition is met, otherwise output 0; is a motion speed threshold value; represents the motion speed size; represents total number of pixels in the region; when and the vehicle speed trigger the projection failure alarm.

9. A computer readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to perform the highway hazardous chemical substance leakage accident monitoring method based on computer vision according to any one of claims 1-7.

Images