Eye tracking-based detection and management method for hazardous chemicals transport personnel and related equipment

Abstract

This invention provides a method and related equipment for the detection and management of hazardous chemical transport personnel based on eye tracking. The method includes: acquiring eye-tracking behavior data of the current transport personnel and environmental status data of the current transport scenario; fusing the eye-tracking behavior data and environmental status data using a dynamic fusion algorithm to determine the current transport personnel's attention state data and corresponding high-risk target area data; determining transport risk score data based on the attention state data and corresponding high-risk target area data; determining the corresponding risk management level based on the transport risk score data, and performing intervention operations corresponding to the risk level on the transport personnel. Through the above steps, the method can identify operator fatigue, distraction, and other states, and proactively intervene through audible and visual alarms, interface optimization, and strategy adjustments to reduce accident risks and ensure the safety and controllability of the entire hazardous chemical transport process.

Description

Technical Field

[0001] This invention relates to the field of intelligent inspection, and in particular to a method, device, system, electronic device and storage medium for the inspection and management of personnel transporting hazardous chemicals based on eye tracking. Background Art

[0002] With the widespread use of hazardous chemicals in industries, energy, and other fields, their transportation safety has become an increasingly important concern.

[0003] Existing hazardous chemical transportation supervision systems typically rely on fixed monitoring equipment and manual inspections, which cannot identify driver attention levels and operational risks in real time. Furthermore, traditional systems focus primarily on monitoring vehicle status or cargo physical parameters, lacking effective modeling and intervention capabilities for "human-caused risks." Especially in complex scenarios such as nighttime, high-noise conditions, or severe weather, operator cognitive fatigue and distraction are difficult to detect in a timely manner, easily leading to safety accidents. Environmental changes and multi-source sensor data during transportation require efficient fusion and real-time response mechanisms; traditional static, rule-based judgment methods are insufficient to meet the demands of efficient safety supervision in complex, dynamic scenarios.

[0004] Therefore, existing methods for monitoring and managing personnel transporting hazardous chemicals have limitations in integrating multimodal data, dynamically identifying personnel status, and implementing risk level interventions. Summary of the Invention

[0005] This invention provides a method for monitoring and managing hazardous chemical transport personnel based on eye tracking, in order to solve the problems of existing methods for monitoring and managing hazardous chemical transport personnel that cannot integrate multimodal data, dynamically identify personnel status, and implement risk level intervention.

[0006] In a first aspect, embodiments of the present invention provide a method for the detection and management of hazardous chemical transport personnel based on eye tracking, the method comprising the following steps:

[0007] Acquire eye-tracking data of current transport personnel and environmental status data of the current transport scenario;

[0008] The eye-tracking behavior data and environmental state data are fused together using a dynamic fusion algorithm to determine the current attention state data of the transport personnel and the corresponding high-risk target area data.

[0009] Based on the attention state data and the corresponding high-risk target area data, the transportation risk score data is determined;

[0010] Based on the transportation risk score data, the corresponding risk management level is determined, and intervention operations corresponding to the risk level are carried out on the transportation personnel.

[0011] Optionally, acquiring the eye-tracking behavior data of the current transport personnel and the environmental status data of the current transport scenario includes:

[0012] Using a non-invasive eye-tracking device, the eye movement behavior data of the current transport personnel is identified by analyzing their gaze point data, gaze duration data, saccade path data, and pupil diameter change data.

[0013] By using environmental sensing devices, data on light intensity, humidity, temperature, noise, and tank status are collected and identified to determine the environmental status data of the current transportation scenario.

[0014] Optionally, the step of fusing the eye-tracking behavior data and environmental state data using a dynamic fusion algorithm to determine the current attention state data of the transport personnel and the corresponding high-risk target area data includes:

[0015] The eye-tracking behavior data and environmental state data are fused according to preset weighting parameters to obtain fused feature data;

[0016] The fused feature data is input into a preset attention state recognition model for recognition processing to obtain the current attention state data of the transport personnel;

[0017] The fused feature data is input into a preset risk area identification model for identification processing to obtain the corresponding high-risk target area data.

[0018] Optionally, determining the transportation risk score data based on the attention state data and the corresponding high-risk target area data includes:

[0019] The attention state data and the corresponding high-risk target area data are input into a preset transportation risk assessment model for matching processing to determine the distribution characteristic data and gaze matching data of the attention state data and the corresponding high-risk target area data.

[0020] Based on the distribution feature data and gaze matching data, the transportation risk score data is determined.

[0021] Optionally, the step of inputting the attention state data and the corresponding high-risk target area data into a preset transportation risk assessment model for matching processing, and determining the distribution characteristic data and gaze matching data of the attention state data and the corresponding high-risk target area data, includes:

[0022] Based on the attention state data, the corresponding gaze heatmap is determined;

[0023] Based on the gaze heatmap, determine the corresponding gaze distribution density data, area coverage data, and saccade path variation range;

[0024] Hazardous area features are extracted from the high-risk target area data to determine hazardous material identification area data and trajectory offset area data;

[0025] The gaze distribution density data, regional coverage data, and saccade path change amplitude are compared with the hazardous material identification area data and trajectory offset area data to generate gaze matching data.

[0026] Optionally, determining the corresponding risk management level based on the transportation risk score data and performing corresponding risk level intervention operations on the transportation personnel includes:

[0027] If the transportation risk score data is at the first risk level, then the interface will highlight the identified high-risk target area and output a voice warning message as an intervention operation.

[0028] If the transportation risk score data is at the second risk level, then the intervention operation of sending a shift reminder instruction and generating a fatigue warning report will be executed.

[0029] If the transportation risk score data is at the third risk level, then the operating speed of the current transport vehicle will be limited to a preset safe speed threshold, and the transport personnel will be contacted.

[0030] Secondly, embodiments of the present invention also provide a hazardous chemical transport personnel detection and management device based on eye tracking, the hazardous chemical transport personnel detection and management device based on eye tracking comprising:

[0031] The first acquisition module is used to acquire the eye movement behavior data of the current transportation personnel and the environmental status data of the current transportation scenario;

[0032] The first determining module is used to fuse the eye-tracking behavior data and environmental state data through a dynamic fusion algorithm to determine the current attention state data of the transport personnel and the corresponding high-risk target area data.

[0033] The second determining module is used to determine transportation risk score data based on the attention state data and the corresponding high-risk target area data;

[0034] The intervention module is used to determine the corresponding risk management level based on the transportation risk score data, and to perform intervention operations on the transportation personnel according to the corresponding risk level.

[0035] Thirdly, embodiments of the present invention provide an eye-tracking-based hazardous chemical transport personnel detection and management system, which includes: an eye-tracking-based hazardous chemical transport personnel detection and management device, a server, and intelligent devices.

[0036] Fourthly, embodiments of the present invention provide an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps in the eye-tracking-based hazardous chemical transport personnel detection and management method provided in embodiments of the present invention.

[0037] Fifthly, embodiments of the present invention provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps in the eye-tracking-based method for detecting and managing hazardous chemical transport personnel provided in the embodiments of the present invention.

[0038] In this embodiment of the invention, eye-tracking behavior data of the current transport personnel and environmental status data of the current transport scenario are acquired. A dynamic fusion algorithm is used to fuse the eye-tracking behavior data and environmental status data to determine the current transport personnel's attention status data and corresponding high-risk target area data. Based on the attention status data and the corresponding high-risk target area data, transport risk score data is determined. Based on the transport risk score data, the corresponding risk management level is determined, and intervention operations corresponding to the risk level are performed on the transport personnel. Through the above method steps, operator fatigue, distraction, and other states can be identified, and proactive intervention can be implemented through audible and visual alarms, interface optimization, and strategy adjustments to reduce accident risks and ensure the safety and controllability of the entire hazardous chemical transport process. Attached Figure Description

[0039] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0040] Figure 1 This is a system architecture diagram of a personnel detection and management system for pipeline hazardous chemical transportation based on eye tracking, provided by an embodiment of the present invention.

[0041] Figure 2 This is a system architecture diagram of another personnel detection and management system for pipeline hazardous chemical transportation based on eye tracking, provided by an embodiment of the present invention;

[0042] Figure 3This is an architecture diagram of a decision tree model with a three-level feedback mechanism provided in an embodiment of the present invention;

[0043] Figure 4 This is a flowchart of a CAN bus protocol interaction provided by an embodiment of the present invention;

[0044] Figure 5 This is a flowchart of a decision tree model optimization provided in an embodiment of the present invention;

[0045] Figure 6 This is a flowchart of a method for detecting and managing personnel transporting hazardous chemicals based on eye tracking, provided in an embodiment of the present invention;

[0046] Figure 7 This is a schematic diagram of another eye-tracking-based hazardous chemical transport personnel detection and management device provided in this embodiment of the invention;

[0047] Figure 8 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention.

[0048] Explanation of reference numerals in the attached figures: 100-Hazardous materials transport personnel detection and management system based on eye tracking; 700-Hazardous materials transport personnel detection and management device based on eye tracking; 701-First acquisition module; 702-First determination module; 703-Second determination module; 704-Intervention module; 101-Server; 102-Intelligent device. Detailed Implementation

[0049] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0050] like Figure 1 As shown, Figure 1This is an architecture diagram of an eye-tracking-based hazardous chemical transport personnel detection and management system 100 provided in an embodiment of the present invention. The system includes an eye-tracking-based hazardous chemical transport personnel detection and management device 700, a server 101, and a smart device 102. The eye-tracking-based hazardous chemical transport personnel detection and management device 700 further includes a first acquisition module, used to acquire the eye-tracking behavior data of the current transport personnel and the environmental state data of the current transport scenario; a first determination module, used to fuse the eye-tracking behavior data and environmental state data using a dynamic fusion algorithm to determine the current transport personnel's attention state data and corresponding high-risk target area data; a second determination module, used to determine transport risk score data based on the attention state data and the corresponding high-risk target area data; and an intervention module, used to determine the corresponding risk management level based on the transport risk score data and perform intervention operations corresponding to the risk level on the transport personnel.

[0051] Specifically, the aforementioned eye-tracking behavior data can refer to eye movement data of transport personnel collected and processed by eye-tracking devices, which are used to reflect their attention distribution and gaze behavior characteristics, including but not limited to fixation point coordinates, fixation duration, saccade path, and pupil diameter change rate. Among these, the fixation point coordinates can be the position of the gaze on the screen or scene, the fixation duration can be the duration for which the gaze stays on the same point, and the saccade path can be the trajectory of the gaze moving between different targets.

[0052] More specifically, the aforementioned eye-tracking data can be collected in real time at high frequency using a non-invasive head-mounted eye tracker, and after preprocessing, sent to a dynamic data processing unit to assess whether the driver's attention status is normal. For example, a sudden increase in pupil diameter or an abnormally shortened fixation time may indicate driver fatigue or distraction.

[0053] The aforementioned environmental status data can be a set of information data reflecting the environmental conditions and working environment under the current transportation scenario. It is used to help determine the complexity of the external environment in which the driver is located and the degree of impact on human eye attention. Specifically, it may include, but is not limited to, light intensity, such as daytime, nighttime or tunnel light brightness; temperature and humidity, such as the ambient temperature and humidity inside and outside the monitoring room or driver's cab, which will affect personnel comfort and fatigue level; and noise level, such as environmental background noise, wind and rain noise or mechanical noise, measured in dB.

[0054] More specifically, when dealing with hazardous chemical transportation, the aforementioned environmental status data can also collect specific hazardous environmental elements, such as tank status information of the transport tank, including sensor data such as internal pressure, liquid level, and external leak detection signals. In this embodiment, comprehensive environmental status data can also be collected, and the environmental interference in the eye-tracking behavior data can be calibrated through the aforementioned eye-tracking-based hazardous chemical transportation personnel detection and management system. For example, when the light is too dim and the aperture is magnified, it is not determined to be pupil dilation caused by fatigue. Furthermore, environmental factors are assigned certain weights according to the aforementioned dynamic fusion algorithm, thereby eliminating the risk impact of environmental status data on driver attention assessment.

[0055] The aforementioned dynamic fusion algorithm can be any multi-source data fusion algorithm used to perform weighted fusion, feature extraction, and comprehensive analysis of eye-tracking behavior data and environmental state data according to a certain strategy. Specifically, the algorithm can optimize and adjust the fusion strategy according to different scenarios or historical data, so that the fusion process is adaptive to environmental changes and individual differences. In this embodiment, features from eye-tracking devices, such as the average gaze duration and pupil change rate, can be linearly or nonlinearly combined with data from environmental sensors, such as temperature and illumination, after normalization processing using preset weight parameters or fusion formulas to obtain fused feature data.

[0056] The corresponding fusion formula and weight parameters can be obtained through the following equations:

[0057]

[0058] in, and The mean and standard deviation of the eye-tracking data are used; α=0.7 and β=0.3 are weighting coefficients, which can be optimized based on historical accident datasets.

[0059] For example, when transport personnel are in a rainstorm environment, the weight of environmental data can be increased and weighted fusion processing can be performed through a dynamic fusion algorithm: for example, using the weight coefficients obtained by optimizing historical data, eye-tracking data is given a weight of α=0.7 and environmental data is given a weight of β=0.3, and fusion features are calculated in real time. Through the above methods and steps, the influence of environmental factors on attention in rainstorm scenarios can be dynamically adapted.

[0060] In one possible embodiment, the aforementioned eye-tracking-based hazardous chemical transport personnel detection and management system performs time-series synchronization, scale normalization, and feature-level fusion calculations on eye-tracking behavior data and environmental status data to generate comprehensive features that can be used for risk assessment. This includes, but is not limited to, weighting data synthesis based on the importance of each data source to risk prediction, or extracting key fusion features through data dimensionality reduction and feature extraction algorithms (such as principal component analysis, PCA).

[0061] For example, when two indicators from different sources—the duration of a driver's continuous gaze at the road ahead and the current degree of road slipperiness—are combined into a composite indicator that better represents the danger, and then input into the attention state recognition model and the risk area recognition model, it can ensure that multidimensional information from the human eye and the environment is comprehensively utilized within the same decision-making framework, making risk analysis more accurate and improving its robustness.

[0062] The aforementioned attention state data can refer to the quantitative representation or classification results of the current attention distribution and concentration level of transportation personnel obtained after fusion analysis. Generally, the aforementioned attention state data can be output by a preset attention state recognition model. The preset attention state recognition model can be an AI model that uses a practical random forest classifier to classify the aforementioned eye-tracking behavior data and environmental state data.

[0063] The aforementioned high-risk target area data can be a set of information on target areas in the current scenario that have high safety risks and require special attention, identified by the aforementioned eye-tracking-based hazardous chemical transport personnel detection and management system. These areas are crucial to transport safety in the monitoring screen or on-site environment, such as valve interfaces of hazardous chemical tanks, warning labels for hazardous chemicals, road areas where the vehicle's trajectory may deviate, obstacles ahead, or potential leak sources.

[0064] Specifically, the aforementioned high-risk target area data can be extracted by a risk area identification model (such as an image recognition algorithm based on CNN). This data can be represented as image coordinates or spatial location and a description of its risk attributes, such as "leakage marker location (x,y), vehicle trajectory deviation area range, abnormal tank pressure area," and other information.

[0065] The data on the high-risk target areas will be matched and analyzed with the driver's attention status data. For example, it will check whether the driver's gaze point and scanning trajectory cover these high-risk areas. If a certain high-risk area is not gazed at continuously, it will be identified as a monitoring blind spot, and corresponding interventions will be taken, such as highlighting the area on the interface or using other sensor methods for inspection.

[0066] The aforementioned transportation risk score data can refer to the quantitative risk assessment result calculated after comprehensively considering driver attention status data and high-risk target area data. The transportation risk score can be generated by a preset transportation risk assessment model. For example, by analyzing features such as attention distribution density, area coverage, and changes in saccade paths, and calculating the area overlap with the characteristics of dangerous areas (hazardous goods signs, trajectory deviation areas, etc.), "gaze matching degree" data is generated. Based on this data, the final risk score is calculated. The aforementioned transportation risk score data can be a score of 0-100 (the larger the value, the higher the risk) or a risk probability in the form of a percentage. For example, when "attention distraction risk score 85%" means that the risk assessment of an accident in the current state is 85%, which is already at a high risk level, the aforementioned hazardous chemical transportation personnel detection and management system based on eye tracking will compare the score with a preset threshold or level classification standard to determine the corresponding risk management level.

[0067] The aforementioned risk management levels can be based on the severity of risks as determined by transportation risk scoring data, and are used to decide the intensity of regulatory and intervention measures. Generally, multiple risk levels can be established, each corresponding to a predetermined set of handling strategies. Intervention actions can be adopted from lenient to stringent based on the different risk management levels. This embodiment can utilize a three-level response mechanism, triggered step-by-step, for example:

[0068] 1. Level 1 Response: The risk area is highlighted on the interface (flashing red box), and a voice prompt of ≥70dB is given simultaneously;

[0069] 2. Level 2 Response: Mandate personnel rotation and transmit fatigue reports to the monitoring terminal;

[0070] 3. Level 3 response: The vehicle's ESC module is linked via the CAN bus protocol to limit the vehicle speed to a safe threshold (e.g., 40 km / h).

[0071] Among these features, dynamic strategy adjustments are possible:

[0072] For example, if the average viewing time of an operator for a hazardous chemical label (such as a corrosive label) is less than 1 second (the experimental verification threshold), the label will be automatically enlarged to 150% and warning text will be superimposed.

[0073] If the same area is not monitored three times in a row, a drone inspection will be initiated and the footage will be simultaneously transmitted to the monitoring terminal.

[0074] Understandably, the determination of different risk management levels can be made and modified according to the specific implementation plan in order to obtain transportation risks that are more in line with the current environmental conditions.

[0075] The aforementioned interventions can be specific control and guidance measures implemented based on the determined risk management level, aimed at reducing risk, correcting driver condition, or strengthening external protection. Generally, these interventions can be proactive response actions to detected anomalies or hazards, encompassing various forms such as human-machine interface prompts, alarm notifications, and direct control of people or vehicles.

[0076] For example, when the risk is low, the intervention can be a prominent visual prompt on the interface (such as a bright flashing box to mark the risk area) or an auditory warning (a voice prompt to the driver to pay attention); when severe driver fatigue and distraction are detected (medium to high risk), the intervention is upgraded to management measures: such as sending a shift rest instruction, generating a fatigue alarm report to notify the supervisor to intervene, or triggering a contingency plan (such as activating a backup driver to take over).

[0077] In extremely high-risk emergencies, intervention measures also include automated control measures: for example, by integrating with the vehicle control system to forcibly reduce vehicle speed or adjust driving parameters; or by activating external equipment such as drones or robots for on-site investigation. It is important to emphasize that intervention measures are characterized by tiered triggering and dynamic adjustment: the system will select the appropriate level of intervention in real time based on the development of the risk, and can evaluate the effectiveness of the intervention through continuous monitoring, escalating or reversing measures as necessary.

[0078] In one possible embodiment, when the transportation environment is at night, the above-mentioned hazardous chemical transportation personnel detection and management system based on eye tracking detects insufficient light intensity through the environmental perception module, automatically activates the thermal imaging camera and overlays the image onto the monitoring interface, and at the same time integrates eye tracking data to identify situations where the driver does not pay enough attention to key parts (such as tank valves), thereby triggering high-brightness warnings and voice reminders, and dispatching drones for close-range inspections when necessary.

[0079] It can be done as follows Figure 2 The architecture diagram of another eye-tracking-based hazardous chemical transport personnel detection and management system illustrates the logical architecture of this embodiment, which can be divided into four parts: data acquisition layer, data processing layer, application layer, and external system linkage interface. In the data acquisition layer, the system collects multi-source data from eye-tracking devices, environmental sensors, and hazardous chemical status sensors through the Jetson AGX Xavier edge computing unit, and transmits it through CAN bus or wireless means.

[0080] The data processing layer is responsible for median filtering, feature extraction and fusion processing of eye-tracking and image data, and for identifying the driver's attention state and risk target areas through random forest and CNN models;

[0081] The application layer is based on the recognition results and links the dynamic response system (triggering voice prompts and speed limits), the drone platform (conducting blind spot inspections), and the enterprise management platform (generating risk reports). It also achieves closed-loop intervention of risks through interface highlighting and feedback optimization mechanisms. In addition, the system has the ability to incrementally learn and adaptively adjust models, continuously optimize decision-making accuracy, and realize intelligent recognition and response management of the entire process of hazardous chemical transportation.

[0082] The above intervention operations can be based on, for example, Figure 3 The following is an explanation of a decision tree model with a three-level feedback mechanism. Starting from the result analysis node, it first detects whether the driver is in a high-risk area. If so, the warning system is activated and an audio warning is issued, and the first-level response process is initiated to further assess whether the driver has entered a dangerous area. If the driver has indeed entered a dangerous area, the second-level intervention is triggered, including automatically controlling the driver's operation and updating the system status.

[0083] If the driver has not entered a high-risk area, continue to check whether he is fatigued or shows signs of fatigue. Based on the analysis results of eye movement data or other physiological data, decide whether to trigger a level 2 or level 3 intervention. Level 3 intervention includes, but is not limited to, limiting the vehicle's operating speed to ensure safety during transportation.

[0084] The aforementioned three-level feedback mechanism decision tree model monitors the driver's status and external environment in real time, and automatically judges and executes different levels of intervention operations based on the risk level, thereby achieving safety control and dynamic management of the transportation process.

[0085] The interaction process between the decision tree model of the above three-level feedback mechanism and the vehicle can be based on, for example... Figure 4 The CAN bus protocol interaction flowchart shown illustrates this process. The gaze analysis module checks if the driver's attention is more than 85% distracted. If high distraction is detected, the system prioritizes Level 3 intervention, adjusting the vehicle's speed. If the distraction threshold is not met, the next step is to check for a failed CRC check. If the check fails, the system uses the ESC module to limit the vehicle's speed and sends an error feedback signal. If the CRC check succeeds, the system sends a CAN message (ID: 0x1A0) via the CAN bus module to perform data parsing and control the vehicle.

[0086] The above decision tree model can also be used as follows: Figure 5The optimization process is illustrated in the flowchart of the decision tree model. Specifically, the system starts by initializing data such as the driver's eye movement data and location information. Based on the ε-greedy strategy, appropriate intervention actions are selected, such as adjusting the interface prompt box or activating voice reminders. After execution, the system calculates the instant reward value by monitoring the driver's feedback behavior, and then updates the Q value in the decision model. As the feedback is continuously updated, the system can gradually optimize the intervention strategy. Through continuous iterative learning, the system can improve safety and intelligent response capabilities in complex transportation environments.

[0087] The above methods and steps can quantify personnel status through eye-tracking data, enabling "human risk prediction," reducing the accident rate caused by human error, and can be adapted to complex scenarios (nighttime, high temperature): the thermal imaging camera (FLIR A700) enhances low-light monitoring and links with eye-tracking data to analyze blind spots, thereby improving regulatory efficiency.

[0088] like Figure 6 As shown, Figure 6 This is a flowchart of a method for detecting and managing hazardous chemical transport personnel based on eye tracking, provided by an embodiment of the present invention. The method includes the following steps:

[0089] 601. Obtain eye movement data of the current transportation personnel and environmental status data of the current transportation scenario.

[0090] In this embodiment of the invention, the above-mentioned eye-tracking-based hazardous chemical transport personnel detection and management method can be applied to an eye-tracking-based hazardous chemical transport personnel detection and management platform. The above-mentioned eye-tracking-based hazardous chemical transport personnel detection and management system has functions such as tracking data processing, tracking data transmission and reception, and tracking data memory storage, and can be built based on a server or server cluster. The server or server cluster can be an electronic device with tracking data capability.

[0091] The aforementioned eye-tracking behavior data can refer to eye movement data of transportation personnel collected and processed by eye-tracking devices, which are used to reflect their attention distribution and gaze behavior characteristics, including but not limited to fixation point coordinates, fixation duration, saccade path, and pupil diameter change rate. Among these, the fixation point coordinates can be the position of the gaze on the screen or scene, the fixation duration can be the duration for which the gaze stays on the same point, and the saccade path can be the trajectory of the gaze moving between different targets.

[0092] More specifically, the aforementioned eye-tracking data can be collected in real time at high frequency using a non-invasive head-mounted eye tracker, and after preprocessing, sent to a dynamic data processing unit to assess whether the driver's attention status is normal. For example, a sudden increase in pupil diameter or an abnormally shortened fixation time may indicate driver fatigue or distraction.

[0093] The aforementioned environmental status data can be a set of information data reflecting the environmental conditions and working environment under the current transportation scenario. It is used to help determine the complexity of the external environment in which the driver is located and the degree of impact on human eye attention. Specifically, it may include, but is not limited to, light intensity, such as daytime, nighttime or tunnel light brightness; temperature and humidity, such as the ambient temperature and humidity inside and outside the monitoring room or driver's cab, which will affect personnel comfort and fatigue level; and noise level, such as environmental background noise, wind and rain noise or mechanical noise, measured in dB.

[0094] More specifically, when dealing with hazardous chemical transportation, the aforementioned environmental status data can also collect specific hazardous environmental elements, such as tank status information of the transport tank, including sensor data such as internal pressure, liquid level, and external leak detection signals. In this embodiment, comprehensive environmental status data can also be collected, and the environmental interference in the eye-tracking behavior data can be calibrated through the aforementioned eye-tracking-based hazardous chemical transportation personnel detection and management system. For example, when the light is too dim and the aperture is magnified, it is not determined to be pupil dilation caused by fatigue. Furthermore, environmental factors are assigned certain weights according to the aforementioned dynamic fusion algorithm, thereby eliminating the risk impact of environmental status data on driver attention assessment.

[0095] 602. Through dynamic fusion algorithm, eye-tracking behavior data and environmental state data are fused to determine the current attention state data of transportation personnel and the corresponding high-risk target area data.

[0096] In this embodiment of the invention, the dynamic fusion algorithm described above can be any multi-source data fusion algorithm used to perform weighted fusion, feature extraction, and comprehensive analysis of eye-tracking behavior data and environmental state data according to a certain strategy. Specifically, the algorithm can optimize and adjust the fusion strategy according to different scenarios or historical data, so that the fusion process is adaptive to environmental changes and individual differences. In this embodiment, features from eye-tracking devices, such as the average gaze duration and pupil change rate, can be linearly or nonlinearly combined with data from environmental sensors, such as temperature and illumination, after normalization processing by preset weight parameters or fusion formulas to obtain fused feature data.

[0097] The corresponding fusion formula and weight parameters can be obtained through the following equations:

[0098]

[0099] in, and The mean and standard deviation of the eye-tracking data are used; α=0.7 and β=0.3 are weighting coefficients, which can be optimized based on historical accident datasets.

[0100] For example, when transport personnel are in a rainstorm environment, the weight of environmental data can be increased and weighted fusion processing can be performed through a dynamic fusion algorithm: for example, using the weight coefficients obtained by optimizing historical data, eye-tracking data is given a weight of α=0.7 and environmental data is given a weight of β=0.3, and fusion features are calculated in real time. Through the above methods and steps, the influence of environmental factors on attention in rainstorm scenarios can be dynamically adapted.

[0101] In one possible embodiment, the aforementioned eye-tracking-based hazardous chemical transport personnel detection and management system performs time-series synchronization, scale normalization, and feature-level fusion calculations on eye-tracking behavior data and environmental status data to generate comprehensive features that can be used for risk assessment. This includes, but is not limited to, weighting data synthesis based on the importance of each data source to risk prediction, or extracting key fusion features through data dimensionality reduction and feature extraction algorithms (such as principal component analysis, PCA).

[0102] For example, when two indicators from different sources—the duration of a driver's continuous gaze at the road ahead and the current degree of road slipperiness—are combined into a composite indicator that better represents the danger, and then input into the attention state recognition model and the risk area recognition model, it can ensure that multidimensional information from the human eye and the environment is comprehensively utilized within the same decision-making framework, making risk analysis more accurate and improving its robustness.

[0103] The aforementioned attention state data can refer to the quantitative representation or classification results of the current attention distribution and concentration level of transportation personnel obtained after fusion analysis. Generally, the aforementioned attention state data can be output by a preset attention state recognition model. The preset attention state recognition model can be an AI model that uses a practical random forest classifier to classify the aforementioned eye-tracking behavior data and environmental state data.

[0104] The aforementioned high-risk target area data can be a set of information on target areas in the current scenario that have high safety risks and require special attention, identified by the aforementioned eye-tracking-based hazardous chemical transport personnel detection and management system. These areas are crucial to transport safety in the monitoring screen or on-site environment, such as valve interfaces of hazardous chemical tanks, warning labels for hazardous chemicals, road areas where the vehicle's trajectory may deviate, obstacles ahead, or potential leak sources.

[0105] Specifically, the aforementioned high-risk target area data can be extracted by a risk area identification model (such as an image recognition algorithm based on CNN). This data can be represented as image coordinates or spatial location and a description of its risk attributes, such as "leakage marker location (x,y), vehicle trajectory deviation area range, abnormal tank pressure area," and other information.

[0106] The data on the high-risk target areas will be matched and analyzed with the driver's attention status data. For example, it will check whether the driver's gaze point and scanning trajectory cover these high-risk areas. If a certain high-risk area is not gazed at continuously, it will be identified as a monitoring blind spot, and corresponding interventions will be taken, such as highlighting the area on the interface or using other sensor methods for inspection.

[0107] 603. Based on attention state data and corresponding high-risk target area data, determine transportation risk score data.

[0108] In this embodiment of the invention, the aforementioned transportation risk score data can refer to the quantitative risk assessment result calculated after comprehensively considering driver attention state data and high-risk target area data. The aforementioned transportation risk score can be generated by a preset transportation risk assessment model. For example, by analyzing features such as attention distribution density, area coverage, and changes in saccade paths, and calculating the area overlap with features of dangerous areas (hazardous goods signs, trajectory deviation areas, etc.), "gaze matching degree" data is generated. Then, based on these data, the final risk score is calculated. The aforementioned transportation risk score data can be a score of 0-100 (the larger the value, the higher the risk) or a risk probability in the form of a percentage. For example, when "attention distraction risk score 85%" means that the risk assessment of an accident occurring in the current state is 85%, which is already at a high risk level, the aforementioned hazardous chemical transportation personnel detection and management system based on eye tracking will compare the score with a preset threshold or level classification standard to determine the corresponding risk management level.

[0109] 604. Based on transportation risk scoring data, determine the corresponding risk management level and implement corresponding risk level intervention operations for transportation personnel.

[0110] In this embodiment of the invention, the aforementioned risk management level can be a severity level of risk classified based on transportation risk scoring data, used to determine the intensity of regulatory and intervention measures. Generally, it can be divided into multiple risk levels, each corresponding to a predetermined set of handling strategies. Intervention operations can be adopted from soft to hard based on different levels of risk management. This embodiment can be implemented through a three-level response mechanism, triggered step by step, for example:

[0111] 1. Level 1 Response: The risk area is highlighted on the interface (flashing red box), and a voice prompt of ≥70dB is given simultaneously;

[0112] 2. Level 2 Response: Mandate personnel rotation and transmit fatigue reports to the monitoring terminal;

[0113] 3. Level 3 response: The vehicle's ESC module is linked via the CAN bus protocol to limit the vehicle speed to a safe threshold (e.g., 40 km / h).

[0114] Among these features, dynamic strategy adjustments are possible:

[0115] For example, if the average viewing time of an operator for a hazardous chemical label (such as a corrosive label) is less than 1 second (the experimental verification threshold), the label will be automatically enlarged to 150% and warning text will be superimposed.

[0116] If the same area is not monitored three times in a row, a drone inspection will be initiated and the footage will be simultaneously transmitted to the monitoring terminal.

[0117] Understandably, the determination of different risk management levels can be made and modified according to the specific implementation plan in order to obtain transportation risks that are more in line with the current environmental conditions.

[0118] The aforementioned interventions can be specific control and guidance measures implemented based on the determined risk management level, aimed at reducing risk, correcting driver condition, or strengthening external protection. Generally, these interventions can be proactive response actions to detected anomalies or hazards, encompassing various forms such as human-machine interface prompts, alarm notifications, and direct control of people or vehicles.

[0119] For example, when the risk is low, the intervention can be a prominent visual prompt on the interface (such as a bright flashing box to mark the risk area) or an auditory warning (a voice prompt to the driver to pay attention); when severe driver fatigue and distraction are detected (medium to high risk), the intervention is upgraded to management measures: such as sending a shift rest instruction, generating a fatigue alarm report to notify the supervisor to intervene, or triggering a contingency plan (such as activating a backup driver to take over).

[0120] In extremely high-risk emergencies, intervention measures also include automated control measures: for example, by integrating with the vehicle control system to forcibly reduce vehicle speed or adjust driving parameters; or by activating external equipment such as drones or robots for on-site investigation. It is important to emphasize that intervention measures are characterized by tiered triggering and dynamic adjustment: the system will select the appropriate level of intervention in real time based on the development of the risk, and can evaluate the effectiveness of the intervention through continuous monitoring, escalating or reversing measures as necessary.

[0121] In one possible embodiment, when the transportation environment is at night, the above-mentioned hazardous chemical transportation personnel detection and management system based on eye tracking detects insufficient light intensity through the environmental perception module, automatically activates the thermal imaging camera and overlays the image onto the monitoring interface, and at the same time integrates eye tracking data to identify situations where the driver does not pay enough attention to key parts (such as tank valves), thereby triggering high-brightness warnings and voice reminders, and dispatching drones for close-range inspections when necessary.

[0122] The above methods and steps can quantify personnel status through eye-tracking data, enabling "human risk prediction," reducing the accident rate caused by human error, and can be adapted to complex scenarios (nighttime, high temperature): the thermal imaging camera (FLIR A700) enhances low-light monitoring and links with eye-tracking data to analyze blind spots, thereby improving regulatory efficiency.

[0123] In this embodiment of the invention, eye-tracking behavior data of the current transport personnel and environmental status data of the current transport scenario are acquired. A dynamic fusion algorithm is used to fuse the eye-tracking behavior data and environmental status data to determine the current transport personnel's attention status data and corresponding high-risk target area data. Based on the attention status data and the corresponding high-risk target area data, transport risk score data is determined. Based on the transport risk score data, the corresponding risk management level is determined, and intervention operations corresponding to the risk level are performed on the transport personnel. Through the above method steps, operator fatigue, distraction, and other states can be identified, and proactive intervention can be implemented through audible and visual alarms, interface optimization, and strategy adjustments to reduce accident risks and ensure the safety and controllability of the entire hazardous chemical transport process.

[0124] Optionally, in the steps of acquiring the eye movement data of the current transport personnel and the environmental status data of the current transport scenario, a non-invasive eye-tracking device can be used to identify the gaze point data, gaze duration data, saccade path data, and pupil diameter change data of the current transport personnel to determine the eye movement data of the current transport personnel; and an environmental sensing device can be used to collect and identify the light, humidity, temperature, noise, and tank status data of the current transport personnel to determine the environmental status data of the current transport scenario.

[0125] In this embodiment of the invention, the aforementioned gaze point data can be the coordinate information of a specific location on the interface or in the physical scene at a particular moment (such as a tank valve, a leak sign, etc. in the image). For example, in night mode, if the driver's gaze is focused on the dashboard rather than on a hazard sign in the forward monitoring screen, it will be determined that the driver has not paid attention to the critical target.

[0126] The aforementioned gaze duration data can refer to the duration for which a driver's gaze remains fixed on a certain location. Generally speaking, when it is detected that the driver's gaze duration on the "corrosive chemicals" warning label is less than the preset duration, or when the driver's gaze duration on multiple different road condition points is less than the preset duration, it indicates that the driver's attention is not focused and is looking around, which is considered to be a cognitive risk.

[0127] The aforementioned scanning path data can also refer to the eye trajectory formed when the driver switches between different gaze points, which is used to analyze their visual inspection habits. For example, if the recorded scanning path repeatedly skips high-risk areas (such as the area near the level gauge or pressure valve), it indicates that there are blind spots and intervention operations need to be triggered.

[0128] The pupil diameter change data mentioned above can be used to reflect the driver's physiological and psychological state. Since the pupils will dilate when a person encounters a sudden event or is in a daze, if the pupil diameter change rate ΔP>15% is detected in high temperature or sudden situations, it can be determined that the driver may be fatigued or in a state of stress, and prompts or rest suggestions can be triggered.

[0129] In this embodiment, the brightness values ​​of the cockpit and monitoring screen can be detected in real time using the MAX44009 light sensor. For example, when the ambient light is below 50 lux in tunnel or nighttime transportation conditions, the FLIR A700 thermal imaging camera can be automatically activated to supplement visual information and analyze blind spots in conjunction with eye-tracking data.

[0130] Humidity values ​​can be collected using the DHT22 temperature and humidity sensor. It is understood that if the humidity is higher than 85%, it may cause driver discomfort or blurred vision. This information can be recorded and used to correct the error of the attention state model, thereby improving the accuracy of fatigue recognition.

[0131] Temperature values ​​can be collected using a DHT22 sensor. It is understandable that if the room temperature is higher than 35°C, the high temperature may cause people to feel tired. At this time, it can be determined that the current environment has an impact on people's attention and physiological state. Combined with pupil change data, it can help identify fatigue state. When necessary, it can also be linked with the air conditioner for adjustment.

[0132] The MAX9814 microphone module can also detect background noise intensity. For example, if noise ≥80dB is detected in heavy rain, it may increase the driver's hearing burden and mask the warning sound. In this case, the feedback mode will be switched to visual warning as the main mode.

[0133] The aforementioned tank status data can be obtained by integrating pressure sensors, liquid level sensors, and gas leak detection modules (such as H2S concentration sensors). In this case, the aforementioned eye-tracking-based hazardous chemical transport personnel detection and management system will acquire the real-time status of the tank, including liquid level >90%, abnormal pressure rise, or leak signals. If the driver does not pay attention to these abnormal parts, the aforementioned eye-tracking-based hazardous chemical transport personnel detection and management system will treat them as high-risk target areas.

[0134] In one possible embodiment, the above-mentioned eye-tracking-based hazardous chemical transport personnel detection and management system collects real-time data on the driver's gaze point, gaze duration, saccade path, and pupil diameter changes using a non-invasive eye-tracking device. Combined with data on illumination, temperature and humidity, noise, and tank status collected by the environmental perception module, it completes the accurate perception and modeling of personnel attention status and the transport environment.

[0135] The above methods and steps not only achieve accurate modeling of the driver's attention state, but also effectively compensate for the blind spots in perception under different scenarios. Furthermore, with the help of sensor fusion and dynamic recognition models, efficient linkage and supervision of "people-vehicle-goods-environment" can be achieved throughout the transportation process.

[0136] Optionally, in the step of fusing eye-tracking behavior data and environmental state data using a dynamic fusion algorithm to determine the current attention state data of the transport personnel and the corresponding high-risk target area data, the method further includes fusing eye-tracking behavior data and environmental state data according to preset weighting parameters to obtain fused feature data; inputting the fused feature data into a preset attention state recognition model for recognition processing to obtain the current attention state data of the transport personnel; and inputting the fused feature data into a preset risk area recognition model for recognition processing to obtain the corresponding high-risk target area data.

[0137] In this embodiment of the invention, the aforementioned preset weighting parameters refer to the different weight ratios assigned to multidimensional data from different sources by the eye-tracking-based hazardous chemical transport personnel detection and management system during the fusion processing. It can be understood that these weight ratios are set to account for the impact of eye-tracking behavior data and environmental state data under different conditions. For example, when the weather is good, the environmental weighting parameter will be slightly reduced. Generally, a weighting parameter ratio of α=0.7 (eye-tracking behavior) and β=0.3 (environmental state) can be used to set the daily fusion processing. It should be noted that these daily preset weighting parameters can be derived from the analysis of 1,000 historical transport accident samples, emphasizing that the driver's own attention performance is more important than the environment itself when driving in rainy weather, but the physiological impact of weather still needs to be considered. Furthermore, the aforementioned preset weighting parameters can be pre-set during deployment and can be adaptively updated according to different transport environments.

[0138] In one possible embodiment, the aforementioned eye-tracking-based hazardous chemical transport personnel detection and management system performs time synchronization, normalization, and weighted calculation on the aforementioned multidimensional data to achieve the process of integrating the collected eye-tracking data and environmental data into multimodal data and obtaining fused feature data.

[0139] The aforementioned fused feature data can refer to the set of feature vectors formed after fusion processing, which contains comprehensive data information reflecting the driver's state and environmental risks. For example, it can include derived features such as "weighted average pupil change rate per unit time", "gaze stability coefficient under high noise environment", and "gaze heatmap offset when tank pressure is abnormal". These features are then input into the attention state recognition model (such as random forest classifier) ​​and the risk area recognition model (such as CNN neural network) respectively to generate corresponding attention state data and high-risk target area data.

[0140] Optionally, in the step of determining transportation risk score data based on attention state data and corresponding high-risk target area data, the method further includes inputting the attention state data and corresponding high-risk target area data into a preset transportation risk assessment model for matching processing to determine the distribution feature data and gaze matching data of the attention state data and corresponding high-risk target area data; and determining the transportation risk score data based on the distribution feature data and gaze matching data.

[0141] In this embodiment of the invention, the aforementioned preset transportation risk assessment model can be a machine learning evaluation model deployed in the aforementioned eye-tracking-based hazardous chemical transportation personnel detection and management system. This model is used to perform safety predictions on the input fused feature data, thereby deriving the transportation risk score data of the current transportation personnel during the transportation process. Specifically, the aforementioned preset transportation risk assessment model can be constructed using a random forest classification algorithm and trained using a training dataset containing 1,000 sets of historical transportation accident samples and normal records. It should be noted that the aforementioned training dataset originates from publicly available data and a self-built data warehouse of the China Chemical Safety Association.

[0142] In one possible embodiment, the aforementioned preset transportation risk assessment model obtains corresponding distribution feature data and gaze matching data by pairing the features and logic between the aforementioned attention state data and the corresponding high-risk target area data.

[0143] Specifically, by analyzing whether there are blind spots or mismatches in the gaze coverage between attention state data and corresponding high-risk target area data, for example, by matching and comparing the gaze heatmap with the coordinates of the area where the dangerous goods label is located, it can be determined whether the dangerous goods label has been effectively gazed upon by the transport personnel and whether the gaze intensity has reached the safety threshold.

[0144] That is, the aforementioned distribution feature data can be a dataset extracted from the aforementioned attention state data to describe the spatial distribution features in the scene, including but not limited to gaze heatmaps, gaze area coverage, saccade path concentration, gaze time distribution, and other data.

[0145] The aforementioned gaze matching data can be a measure of regional overlap and attention sufficiency generated by comparing gaze data with high-risk area data, used to determine whether the driver has paid sufficient attention to the risk area.

[0146] In another possible embodiment, the aforementioned preset risk area identification model uses the distribution feature data and gaze matching data as input features, and analyzes and calculates factors such as attention distraction level, high-risk area matching degree, and gaze time in key areas. The output risk score is 82 points (out of 100), which exceeds the system's set "Level 1 Risk" threshold (70 points). Based on this, it is determined that there is a significant safety hazard in the current transportation process, and the following intervention operation is immediately triggered:

[0147] Level 1 intervention: Zoom in on the sulfuric acid area image to 150% on the driver monitoring interface and add a red warning message: "Corrosive chemicals - please conduct key inspections";

[0148] Monitoring upgrade: The system records the blind spots in the current period and synchronizes risk reports to the remote dispatch center;

[0149] If the area remains undetected in the next cycle, the drone inspection program will be automatically activated to conduct high-definition inspections of the sulfuric acid storage tanks to fill in the gaps.

[0150] By using the above methods and steps, compared with traditional solutions that rely solely on cameras or fatigue algorithms, this embodiment can combine spatial gaze distribution with the effectiveness of visual inspection, making risk scoring more targeted and dynamically adjustable, thereby improving the accuracy of risk response and the timeliness of intervention in the scenario of mixed transportation of hazardous chemicals.

[0151] Optionally, the steps of inputting attention state data and corresponding high-risk target area data into a preset transportation risk assessment model for matching processing, and determining the distribution characteristic data and gaze matching data of attention state data and corresponding high-risk target area data, further include: determining the corresponding gaze heatmap based on the attention state data; determining the corresponding gaze distribution density data, area coverage data, and scanning path change amplitude based on the gaze heatmap; extracting hazardous area features from the high-risk target area data to determine hazardous material identification area data and trajectory deviation area data; and performing area overlap analysis between the gaze distribution density data, area coverage data, and scanning path change amplitude and the hazardous material identification area data and trajectory deviation area data to generate gaze matching data.

[0152] In this embodiment of the invention, the above-mentioned gaze distribution density data can be a statistical value of the cumulative number of gaze points in a unit area, used to measure the concentration of attention in a certain visual field area. Generally speaking, it can display the degree or duration of gaze in a certain area of ​​the driver's forward visual field.

[0153] The aforementioned area coverage data can be the degree to which a certain object or feature is covered by the line of sight of a transporter. For example, when a pothole appears on the roadside, it is detected that the transporter's line of sight coverage of the pothole is only 10%, meaning they can only see a corner of the pothole. If the pothole is large or poses other safety hazards, the transporter cannot accurately determine all the safety hazards of the pothole based on only a 10% line of sight coverage. Therefore, the calculation and analysis of area coverage data can improve the accuracy of identifying road safety hazards.

[0154] The aforementioned changes in the saccade path can be used to measure the degree of eye movement fluctuation or stability within the driver's field of vision, typically expressed as changes in angular velocity or trajectory steering angle. For example, on winding roads, if the system detects frequent directional changes in the saccade path exceeding 10°, it indicates frequent shifts in gaze but a lack of stable fixation. In this scenario, the average standard deviation of the saccade trajectory in the right-side monitoring area was 19.2%, far exceeding the normal range (<10%), further confirming that the driver was not focusing on the high-risk area.

[0155] In one possible embodiment, image recognition algorithms, including CNN models, can be used to identify structured risk areas in the monitoring footage. Specifically, the location and size information of key areas such as corrosive labels, flammable warning icons, tank valves, and pressure gauges can be extracted by analyzing the camera footage. In this example, the system accurately calibrates the "corrosive" label (center point X, Y coordinates) within the right-side sulfuric acid tank area and sets it as a primary priority marking area.

[0156] It can also detect, through vehicle trajectory sensing equipment and image registration technology, that transport vehicles briefly deviate from the set route trajectory when passing through construction areas in industrial zones, and that the area is close to the emission area of ​​chemical plants, and mark the corresponding forward road segment in the video footage as a high-risk area for trajectory deviation.

[0157] In another possible embodiment, the above-mentioned eye-tracking-based hazardous chemical transport personnel detection and management system performs spatial alignment and overlap matching analysis on the above-mentioned "gaze heatmap analysis results" and the identified "hazardous area layer", calculates the gaze matching degree, and then summarizes the above results to generate "gaze matching data", which is combined with other features and input into the transport risk assessment model. Finally, the model outputs a risk score of 89 points, reaching the level of level three risk.

[0158] By employing the methods and steps described above, risk identification becomes more accurate and responses are more timely, thereby reducing transportation risks caused by blind spots in monitoring or lack of attention.

[0159] Optionally, the steps of determining the corresponding risk management level based on transportation risk scoring data and performing corresponding risk level intervention operations on transportation personnel may further include: if the transportation risk scoring data is at the first risk level, then performing an intervention operation to highlight the identified high-risk target area on the interface and output a voice warning message; if the transportation risk scoring data is at the second risk level, then performing an intervention operation to send a shift rotation reminder instruction and generate a fatigue warning report; if the transportation risk scoring data is at the third risk level, then performing an intervention operation to limit the current transportation vehicle's operating speed to a preset safe speed threshold and contacting the transportation personnel.

[0160] In this embodiment of the invention, three scenarios are used to illustrate the three risk levels:

[0161] In the first risk level scenario, where the driver is entering a construction zone, if before entering the construction zone, it is identified that the driver briefly loses focus and fails to effectively look at the "Temporary Leakage Warning Sign in the Construction Area," with a gaze heatmap showing a coverage rate of only 42%, the assessment model calculates a current risk score of 75, classifying it as the first risk level. Immediately, a Level 1 intervention is required.

[0162] In the vehicle monitoring interface, the area where the indicator is located is automatically highlighted and marked with a red flashing box.

[0163] At the same time, a voice warning was issued through a loudspeaker: "Please note: There is a chemical leak warning sign in the construction area ahead. Please check."

[0164] The event will be recorded in the behavior log for this period for subsequent inspections and retrospective analysis.

[0165] At the second risk level, the scenario involves driving on a road in a high-temperature area at midday. Environmental sensors detect an indoor temperature of 37°C. Eye movement data shows a pupil diameter change rate ΔP of 19%, decreased fixation concentration, and drastic changes in saccade path. The aforementioned assessment model classifies this as a moderate fatigue state, with a current transportation risk score of 84, triggering the second risk level response and automatically executing a level-two intervention:

[0166] Send a "shift rotation reminder instruction" to the driver's terminal and remind the driver to hand over the shift to the backup driver at the next service area;

[0167] A "Fatigue Risk Warning Report" is generated, which includes: environmental parameters at the time, eye movement index data, risk score details and recommended handling measures, and is uploaded to the remote safety supervision platform in real time.

[0168] If the driver does not respond to the handover instruction within 10 minutes, a voice reminder will be issued again, prompting the dispatcher to intervene manually.

[0169] At the third risk level, the scenario is entering a tunnel at night. If the driver fails to look at the exit area for an extended period (blind spot lasting >8 seconds), and the heatmap overlap is extremely low; simultaneously, the monitoring footage detects the vehicle deviating from its lane. Combined with other feature models, the output score is 92, indicating a serious safety risk. Therefore, a level three intervention is immediately implemented.

[0170] The vehicle's electronic stability control module (ESC) is controlled via the CAN bus to force the driving speed to be limited to a safe threshold of 40 km / h.

[0171] The vehicle-mounted communication module initiates a real-time voice communication request with the duty dispatch center to report the current risk level, location, and driver status.

[0172] It also activates a full-screen red alert, locks high-risk areas, and continuously broadcasts voice messages.

[0173] It should be noted that different levels of intervention will be implemented based on different risk levels, and the three risk levels can be used to intervene in a step-by-step manner to address data fluctuations under different changes in a particular scenario.

[0174] like Figure 7 As shown, this embodiment of the invention also provides an eye-tracking-based hazardous chemical transport personnel detection and management device 700, which includes:

[0175] The first acquisition module 701 is used to acquire the eye movement behavior data of the current transportation personnel and the environmental status data of the current transportation scenario;

[0176] The first determining module 702 is used to fuse the eye-tracking behavior data and environmental state data through a dynamic fusion algorithm to determine the current attention state data of the transport personnel and the corresponding high-risk target area data.

[0177] The second determining module 703 is used to determine transportation risk score data based on the attention state data and the corresponding high-risk target area data;

[0178] The intervention module 704 is used to determine the corresponding risk management level based on the transportation risk score data, and to perform intervention operations on the transportation personnel according to the corresponding risk level.

[0179] Optionally, the first acquisition module 701 mentioned above further includes:

[0180] The first determining submodule is used to identify the current transporter's gaze point data, gaze duration data, saccade path data, and pupil diameter change data using a non-invasive eye-tracking device, and determine the current transporter's eye movement behavior data.

[0181] The second determination submodule is used to collect and identify the current light, humidity, temperature, noise, and tank status data of the transportation personnel through environmental sensing devices, and determine the environmental status data of the current transportation scenario.

[0182] Optionally, the first determining module 702 mentioned above further includes:

[0183] The fusion submodule is used to fuse the eye-tracking behavior data and environmental state data according to preset weighting parameters to obtain fused feature data;

[0184] The acquisition submodule is used to input the fused feature data into a preset attention state recognition model for recognition processing to obtain the attention state data of the current transport personnel.

[0185] The processing submodule is used to input the fused feature data into a preset risk area identification model for identification processing to obtain the corresponding high-risk target area data.

[0186] Optionally, the second determining module 703 mentioned above includes:

[0187] The matching submodule is used to input the attention state data and the corresponding high-risk target area data into a preset transportation risk assessment model for matching processing, and to determine the distribution feature data and gaze matching data of the attention state data and the corresponding high-risk target area data.

[0188] The third determining submodule is used to determine the transportation risk score data based on the distribution feature data and the gaze matching data.

[0189] Optionally, the above matching submodule includes:

[0190] The fourth determination submodule is used to determine the corresponding gaze heatmap based on the attention state data;

[0191] The fifth determination submodule is used to determine the corresponding gaze distribution density data, area coverage data, and saccade path change amplitude based on the gaze heatmap.

[0192] The sixth determination submodule is used to extract dangerous area features from the high-risk target area data and determine dangerous goods identification area data and trajectory offset area data;

[0193] The generation submodule is used to perform regional overlap analysis on the gaze distribution density data, regional coverage data, and saccade path change amplitude with the hazardous material identification area data and trajectory offset area data to generate gaze matching data.

[0194] Optionally, the intervention module 704 mentioned above includes:

[0195] The first intervention submodule is used to perform an intervention operation that highlights the identified high-risk target area on the interface and outputs a voice warning message when the transportation risk score data is at the first risk level.

[0196] The second intervention submodule is used to execute the intervention operation of sending a shift reminder instruction and generating a fatigue warning report if the transportation risk score data is at the second risk level.

[0197] The third intervention submodule is used to limit the operating speed of the current transport vehicle to a preset safe speed threshold and contact the transport personnel if the transport risk score data is at the third risk level.

[0198] like Figure 8 As shown, this embodiment of the invention also provides an electronic device 800, including a processor, which can execute any of the above-mentioned methods for the detection and management of hazardous chemical transport personnel based on eye tracking.

[0199] Specifically, it includes a processor 801 and a memory 802, as well as a computer program stored in the memory 802 and capable of running on the processor 801, which executes a method for detecting and managing hazardous chemical transport personnel based on eye tracking, wherein:

[0200] The processor 801 runs the calculator program stored in memory 802, which is a method for detecting and managing hazardous chemical transport personnel based on eye tracking, and executes the following steps:

[0201] Acquire eye-tracking data of current transport personnel and environmental status data of the current transport scenario;

[0202] The eye-tracking behavior data and environmental state data are fused together using a dynamic fusion algorithm to determine the current attention state data of the transport personnel and the corresponding high-risk target area data.

[0203] Based on the attention state data and the corresponding high-risk target area data, the transportation risk score data is determined;

[0204] Based on the transportation risk score data, the corresponding risk management level is determined, and intervention operations corresponding to the risk level are carried out on the transportation personnel.

[0205] Optionally, the processor 801 performs the process of acquiring the eye-tracking behavior data of the current transport personnel and the environmental state data of the current transport scenario, including:

[0206] Using a non-invasive eye-tracking device, the eye movement behavior data of the current transport personnel is identified by analyzing their gaze point data, gaze duration data, saccade path data, and pupil diameter change data.

[0207] By using environmental sensing devices, data on light intensity, humidity, temperature, noise, and tank status are collected and identified to determine the environmental status data of the current transportation scenario.

[0208] Optionally, the processor 801 executes the dynamic fusion algorithm to fuse the eye-tracking behavior data and environmental state data to determine the current transport personnel's attention state data and corresponding high-risk target area data, including:

[0209] The eye-tracking behavior data and environmental state data are fused according to preset weighting parameters to obtain fused feature data;

[0210] The fused feature data is input into a preset attention state recognition model for recognition processing to obtain the current attention state data of the transport personnel;

[0211] The fused feature data is input into a preset risk area identification model for identification processing to obtain the corresponding high-risk target area data.

[0212] Optionally, the processor 801 executes the process of determining transportation risk score data based on the attention state data and the corresponding high-risk target area data, including:

[0213] The attention state data and the corresponding high-risk target area data are input into a preset transportation risk assessment model for matching processing to determine the distribution characteristic data and gaze matching data of the attention state data and the corresponding high-risk target area data.

[0214] Based on the distribution feature data and gaze matching data, the transportation risk score data is determined.

[0215] Optionally, the processor 801 further performs the process of inputting the attention state data and the corresponding high-risk target area data into a preset transportation risk assessment model for matching, and determines the distribution feature data and gaze matching data of the attention state data and the corresponding high-risk target area data, including:

[0216] Based on the attention state data, the corresponding gaze heatmap is determined;

[0217] Based on the gaze heatmap, determine the corresponding gaze distribution density data, area coverage data, and saccade path variation range;

[0218] Hazardous area features are extracted from the high-risk target area data to determine hazardous material identification area data and trajectory offset area data;

[0219] The gaze distribution density data, regional coverage data, and saccade path change amplitude are compared with the hazardous material identification area data and trajectory offset area data to generate gaze matching data.

[0220] Optionally, the processor 801 further executes the process of determining the corresponding risk management level based on the transportation risk score data, and performing intervention operations on the transportation personnel according to the corresponding risk level, including:

[0221] If the transportation risk score data is at the first risk level, then the interface will highlight the identified high-risk target area and output a voice warning message as an intervention operation.

[0222] If the transportation risk score data is at the second risk level, then the intervention operation of sending a shift reminder instruction and generating a fatigue warning report will be executed.

[0223] If the transportation risk score data is at the third risk level, then the operating speed of the current transport vehicle will be limited to a preset safe speed threshold, and the transport personnel will be contacted.

[0224] This invention also provides a computer-readable storage medium storing a computer program. When executed by a processor, the computer program implements the various processes of the eye-tracking-based hazardous chemical transport personnel detection and management method or the application-side eye-tracking-based hazardous chemical transport personnel detection and management method provided in this invention, and achieves the same technical effect. To avoid repetition, it will not be described again here.

[0225] Those skilled in the art will understand that implementing all or part of the processes in the above embodiments can be done by a computer program instructing related hardware, and can be stored in a computer-readable storage medium. When executed, the program can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0226] The above description discloses only preferred embodiments of the present invention and should not be construed as limiting the scope of the present invention. Therefore, equivalent variations made in accordance with the claims of the present invention are still within the scope of the present invention.

Claims

1. A method for monitoring and managing personnel transporting hazardous chemicals based on eye-tracking, characterized in that, include: Acquire eye-tracking data of current transport personnel and environmental status data of the current transport scenario; The eye-tracking behavior data and environmental state data are fused together using a dynamic fusion algorithm to determine the current attention state data of the transport personnel and the corresponding high-risk target area data. Based on the attention state data and the corresponding high-risk target area data, the transportation risk score data is determined; Based on the transportation risk score data, the corresponding risk management level is determined, and intervention operations corresponding to the risk level are carried out on the transportation personnel. The process of determining transportation risk score data based on the attention state data and the corresponding high-risk target area data includes: The attention state data and the corresponding high-risk target area data are input into a preset transportation risk assessment model for matching processing to determine the distribution characteristic data and gaze matching data of the attention state data and the corresponding high-risk target area data. Based on the distribution feature data and gaze matching data, the transportation risk score data is determined; The step of inputting the attention state data and the corresponding high-risk target area data into a preset transportation risk assessment model for matching processing, and determining the distribution characteristic data and gaze matching data of the attention state data and the corresponding high-risk target area data, includes: Based on the attention state data, the corresponding gaze heatmap is determined; Based on the gaze heatmap, determine the corresponding gaze distribution density data, area coverage data, and saccade path variation range; Hazardous area features are extracted from the high-risk target area data to determine hazardous material identification area data and trajectory offset area data; The gaze distribution density data, area coverage data, and saccade path change amplitude are compared with the hazardous materials identification area data and trajectory offset area data to perform area overlap analysis, thereby generating gaze matching data. The process involves fusing the eye-tracking behavior data and environmental state data using a dynamic fusion algorithm to determine the current attention state data of the transport personnel and the corresponding high-risk target area data, including: According to preset weighting parameters, the eye-tracking behavior data and environmental state data are fused to obtain fused feature data. The preset weighting parameters assign different weight ratios to multidimensional data from different sources. The weight ratios are set to account for the influence of eye-tracking behavior data and environmental state data under different conditions. The environmental state data is collected for hazardous chemical transportation, including sensor data of pressure inside the transport tank, liquid level, and external leakage detection signals. The fused feature data is input into a preset attention state recognition model for recognition processing to obtain the current attention state data of the transport personnel; The fused feature data is input into a preset risk area identification model for identification processing to obtain the corresponding high-risk target area data.

2. The method for monitoring and managing hazardous chemical transport personnel based on eye tracking as described in claim 1, characterized in that, The acquisition of eye-tracking data of the current transport personnel and environmental status data of the current transport scenario includes: Using a non-invasive eye-tracking device, the eye movement behavior data of the current transport personnel is identified by analyzing their gaze point data, gaze duration data, saccade path data, and pupil diameter change data. By using environmental sensing devices, data on light intensity, humidity, temperature, noise, and tank status are collected and identified to determine the environmental status data of the current transportation scenario.

3. The method for monitoring and managing hazardous chemical transport personnel based on eye tracking as described in claim 1, characterized in that, The process of determining the corresponding risk management level based on the transportation risk scoring data and implementing corresponding risk level intervention operations for the transportation personnel includes: If the transportation risk score data is at the first risk level, then the intervention operation will be performed to highlight the identified high-risk target area on the interface and output a voice warning message. If the transportation risk score data is at the second risk level, then the intervention operation of sending a shift reminder instruction and generating a fatigue warning report will be executed. If the transportation risk score data is at the third risk level, then the operating speed of the current transport vehicle will be limited to a preset safe speed threshold, and the transport personnel will be contacted.

4. A device for detecting and managing personnel transporting hazardous chemicals based on eye tracking, characterized in that, include: The first acquisition module is used to acquire the eye movement behavior data of the current transportation personnel and the environmental status data of the current transportation scenario; The first determining module is used to fuse the eye-tracking behavior data and environmental state data through a dynamic fusion algorithm to determine the current attention state data of the transport personnel and the corresponding high-risk target area data. The second determining module is used to determine transportation risk score data based on the attention state data and the corresponding high-risk target area data; The intervention module is used to determine the corresponding risk management level based on the transportation risk score data, and to perform intervention operations on the transportation personnel according to the corresponding risk level. The second determining module is further configured to input the attention state data and the corresponding high-risk target area data into a preset transportation risk assessment model for matching processing, to determine the distribution feature data and gaze matching data of the attention state data and the corresponding high-risk target area data; and to determine the transportation risk score data based on the distribution feature data and gaze matching data. Based on the attention state data, a corresponding gaze heatmap is determined; according to the gaze heatmap, corresponding gaze distribution density data, region coverage data, and saccade path change amplitude are determined; dangerous area features are extracted from the high-risk target area data to determine dangerous goods identification area data and trajectory deviation area data; the gaze distribution density data, region coverage data, and saccade path change amplitude are compared with the dangerous goods identification area data and trajectory deviation area data to generate gaze matching data; The intervention module is also used to fuse the eye-tracking behavior data and environmental state data according to preset weighting parameters to obtain fused feature data. The preset weighting parameters are set by assigning different weight ratios to multidimensional data from different sources. The weight ratios are set to account for the influence of eye-tracking behavior data and environmental state data under different states. When transporting hazardous chemicals, the environmental state data collects specific hazardous environmental elements, including sensor data of pressure inside the transport tank, liquid level, and external leakage detection signals. The fused feature data is input into a preset attention state recognition model for recognition processing to obtain the current attention state data of the transport personnel; The fused feature data is input into a preset risk area identification model for identification processing to obtain the corresponding high-risk target area data.

5. A hazardous chemical transport personnel detection and management system based on eye tracking, characterized in that, The eye-tracking-based hazardous chemical transport personnel detection and management system includes: an eye-tracking-based hazardous chemical transport personnel detection and management device; The eye-tracking-based hazardous chemical transport personnel detection and management device implements the eye-tracking-based hazardous chemical transport personnel detection and management method described in claim 1.

6. An electronic device, characterized in that, include: The memory, the processor, and the computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the steps of the eye-tracking-based method for detecting and managing personnel transporting hazardous chemicals as described in any one of claims 1 to 3.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the eye-tracking-based method for detecting and managing hazardous chemical transport personnel as described in any one of claims 1 to 3.

Images