Intelligent cruise ship passenger safety early warning system and method based on fusion of large model and visual identification

Abstract

The embodiment of the invention discloses a cruise ship passenger safety intelligent early warning system and method based on fusion of a large model and visual identification. According to the technical scheme provided by the embodiment of the invention, the dynamic risk field fused with environmental factors is introduced, fusion analysis of passenger behaviors, time-space and social multi-dimensional risk parameters is combined, and behavior chain deduction based on a large model is carried out, so that a safety early warning system identifies passenger risk context information, quantifies passenger risk entropies, and improves passenger risk performance. And a risk evolution path is predicted, and conversion from passive response to isolated events to active prediction and intervention of composite risks is realized. Compared with an existing early warning scheme which depends on a fixed rule and has data unrelated, the method improves the timeliness and accuracy of safety early warning, remarkably reduces the false alarm rate, and meets the requirements of timely early warning for passengers in a complex and closed scene of a cruise ship.

Description

Technical Field

[0001] This application relates to the field of computer vision technology, and in particular to a cruise ship passenger safety intelligent early warning system and method based on the fusion of large models and visual recognition. Background Technology

[0002] As enclosed maritime communities, cruise ships present passengers with various safety risks, such as falling overboard, conflicts, and accidental falls. Therefore, effective early warning systems are essential to promptly alert passengers to these risks and ensure their safety. Currently, cruise ship safety early warning systems primarily rely on the integrated use of video surveillance, personnel positioning, and sensor networks. These systems employ rule-based judgment logic to provide early warnings. For example, when a camera detects a passenger's dangerous actions (such as climbing railings) or when a passenger enters a danger zone designated by an electronic fence, the system triggers a safety alarm, thus providing a safety warning.

[0003] However, existing cruise ship safety warning systems operate with independent sensing systems and uncorrelated data, resulting in significant delays and high false alarm rates. For example, the system only triggers an alarm when it detects a passenger climbing over a railing, leading to delayed warnings, passive intervention, and an inability to effectively prevent various safety incidents from occurring. Summary of the Invention

[0004] This application provides a cruise passenger safety intelligent early warning system and method based on the fusion of large model and visual recognition. It can identify passenger risk context information, quantify passenger risk entropy value, and predict risk evolution path, thereby improving the timeliness and accuracy of cruise safety early warning and solving the problems of early warning lag and high false alarm rate in cruise safety early warning systems.

[0005] In a first aspect, embodiments of this application provide a cruise passenger safety intelligent early warning method based on the fusion of large models and visual recognition, including: Multimodal perception data of the cruise ship is collected, including passenger visual data, passenger positioning data and cruise ship environmental data. Based on the cruise ship environmental data, a dynamic risk field in the digital twin space of the cruise ship is constructed. The dynamic risk field is used to characterize the dynamic risk intensity of different locations in the cruise ship space as time and environment change. Based on passenger visual data and passenger location data, the target passenger's behavior is identified, trajectory is tracked and social relationship is analyzed to generate corresponding behavioral dimension risk parameters, spatiotemporal dimension risk parameters and social dimension risk parameters. Based on dynamic risk field, behavioral risk parameters, spatiotemporal risk parameters, and social risk parameters, the dynamic risk entropy value of the target passenger is calculated. When the dynamic risk entropy value reaches the set risk threshold, the behavior chain of the target passenger is deduced based on a specified large model to predict the risk behavior chain sequence of the target passenger. Based on the risk behavior chain sequence and the dynamic risk entropy value, a safety warning result is output.

[0006] Furthermore, a dynamic risk field is constructed within the cruise ship digital twin space based on cruise ship environmental data, including: Based on cruise ship environmental data, the basic risk intensity of predefined static risk areas in the cruise ship digital twin space is adjusted to generate a corresponding dynamic risk field. Cruise ship environmental data includes at least one of weather, wind speed, ship roll, visibility and time information.

[0007] Furthermore, social relationship analysis includes: Based on passenger visual data and passenger location data, the relative distance, direction of movement, and interaction posture of the target passenger and other passengers in the surrounding area are determined. The social status of the target passenger is determined based on the relative distance, direction of movement, and interaction posture. Based on the social status determination result, corresponding social dimension risk parameters are generated for the target passenger. The social status determination result includes whether the target passenger is in an isolated state, a companion state, or a conflict state.

[0008] Furthermore, based on the dynamic risk field, behavioral dimension risk parameters, spatiotemporal dimension risk parameters, and social dimension risk parameters, the dynamic risk entropy value of the target passenger is calculated, including: The basic risk value is obtained by weighting and fusing the risk parameters of the behavioral dimension, the spatiotemporal dimension, and the social dimension. The dynamic risk field is queried based on the target passenger's location to determine the real-time target risk field intensity of the target passenger. Based on the target risk field intensity and the dynamic risk field, the dynamic risk entropy value of the target passenger is calculated.

[0009] Furthermore, based on a specified large model, behavioral chain extrapolation is performed on the target passengers to predict the risk behavioral chain sequence of the target passengers, including: Based on passenger visual data and passenger location data, real-time behavior, spatiotemporal location, social status and historical trajectory information of target passengers are collected to obtain contextual prompts. Input contextual hints into a specified large model, and output at least one risk behavior chain sequence with probability information based on the specified large model.

[0010] Furthermore, based on the risk behavior chain sequence and dynamic risk entropy value, a safety warning result is output, including: The warning level base is determined based on the dynamic risk entropy value, and then the warning level base is corrected based on the danger level and probability of occurrence of the predicted behavior in the risk behavior chain sequence, generating a graded warning instruction as a safety warning result.

[0011] Furthermore, after generating tiered early warning instructions as a safety warning result, it also includes: Based on the graded warning instructions from the safety warning results, the corresponding proactive intervention measures will be implemented.

[0012] In a second aspect, embodiments of this application provide a cruise ship passenger safety intelligent early warning system based on the fusion of large models and visual recognition, comprising: The risk field construction module is used to collect multimodal perception data of the cruise ship, including passenger visual data, passenger positioning data and cruise ship environmental data. Based on the cruise ship environmental data, a dynamic risk field is constructed in the digital twin space of the cruise ship. The dynamic risk field is used to characterize the dynamic risk intensity of different locations in the cruise ship space as time and environment change. The parameter analysis module is used to perform behavior recognition, trajectory tracking and social relationship analysis on target passengers based on passenger visual data and passenger location data, and generate corresponding behavioral dimension risk parameters, spatiotemporal dimension risk parameters and social dimension risk parameters. The early warning module is used to calculate the dynamic risk entropy value of the target passenger based on the dynamic risk field, behavioral dimension risk parameters, spatiotemporal dimension risk parameters, and social dimension risk parameters. When the dynamic risk entropy value reaches the set risk threshold, it performs behavioral chain deduction on the target passenger based on a specified large model to predict the risk behavior chain sequence of the target passenger. Based on the risk behavior chain sequence and the dynamic risk entropy value, it outputs a safety early warning result.

[0013] In a third aspect, embodiments of this application provide an electronic device, including: Memory and one or more processors; The memory is used to store one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the intelligent early warning method for cruise passenger safety based on the fusion of large models and visual recognition as described in the first aspect.

[0014] In a fourth aspect, embodiments of this application provide a storage medium containing computer-executable instructions, which, when executed by a computer processor, are used to perform the intelligent early warning method for cruise passenger safety based on the fusion of large models and visual recognition as described in the first aspect.

[0015] This application embodiment collects multimodal perception data from the cruise ship, including passenger visual data, passenger positioning data, and cruise ship environmental data. Based on the cruise ship environmental data, a dynamic risk field is constructed in the cruise ship's digital twin space. The dynamic risk field is used to characterize the dynamic risk intensity of different locations within the cruise ship space as time and the environment change. Based on passenger visual data and passenger positioning data, behavior recognition, trajectory tracking, and social relationship analysis are performed on the target passenger to generate corresponding behavioral dimension risk parameters, spatiotemporal dimension risk parameters, and social dimension risk parameters. Based on the dynamic risk field, behavioral dimension risk parameters, spatiotemporal dimension risk parameters, and social dimension risk parameters, the dynamic risk entropy value of the target passenger is calculated. When the dynamic risk entropy value reaches a set risk threshold, a behavior chain deduction is performed on the target passenger based on a specified large model to predict the risk behavior chain sequence of the target passenger. Based on the risk behavior chain sequence and the dynamic risk entropy value, a safety warning result is output. By employing the aforementioned technical means, and introducing a dynamic risk field that integrates environmental factors, combined with the fusion analysis of multi-dimensional risk parameters related to passenger behavior, spatiotemporal factors, and social interactions, as well as behavioral chain extrapolation based on a large model, the safety early warning system can identify passenger risk context information, quantify passenger risk entropy values, and predict risk evolution paths. This achieves a shift from passively responding to isolated events to proactively predicting and intervening in complex risks. Compared to existing early warning schemes that rely on fixed rules and have uncorrelated data, this application improves the timeliness and accuracy of safety early warnings and significantly reduces the false alarm rate, meeting the need for timely early warnings for passengers in the complex and enclosed environment of cruise ships. Attached Figure Description

[0016] Figure 1 This is a flowchart of a cruise passenger safety intelligent early warning method based on the fusion of large model and visual recognition, provided in Embodiment 1 of this application; Figure 2 This is a flowchart of social relationship analysis in Embodiment 1 of this application; Figure 3 This is a flowchart of the dynamic risk entropy value calculation in Embodiment 1 of this application; Figure 4 This is a flowchart illustrating the generation of the risk behavior chain sequence in Embodiment 1 of this application; Figure 5 This is a schematic diagram of the structure of a cruise ship passenger safety intelligent early warning system based on the fusion of large model and visual recognition, provided in Embodiment 2 of this application; Figure 6 This is a schematic diagram of the structure of an electronic device provided in Embodiment 3 of this application. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this application clearer, specific embodiments of this application will be described in further detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are merely for explaining this application and not for limiting it. It should also be noted that, for ease of description, only the parts relevant to this application are shown in the drawings, not all of them. Before discussing exemplary embodiments in more detail, it should be mentioned that some exemplary embodiments are described as processes or methods depicted as flowcharts. Although the flowcharts describe operations (or steps) as sequential processes, many of these operations can be performed in parallel, concurrently, or simultaneously. Furthermore, the order of the operations can be rearranged. The process can be terminated when its operation is completed, but may also have additional steps not included in the drawings. The process can correspond to a method, function, procedure, subroutine, subprogram, etc.

[0018] Example 1: Figure 1 A flowchart of a cruise passenger safety intelligent early warning method based on the fusion of large-scale model and visual recognition, provided in Embodiment 1 of this application, is given. This method can be executed by a cruise passenger safety intelligent early warning device based on the fusion of large-scale model and visual recognition. This device can be implemented through software and / or hardware. It can consist of two or more physical entities, or it can consist of a single physical entity. Generally, this device can be a processing device such as a cruise safety early warning server.

[0019] The following description uses this intelligent safety early warning device as the main body for implementing the cruise passenger safety intelligent early warning method based on the fusion of large models and visual recognition. (Refer to...) Figure 1 The intelligent early warning method for cruise passenger safety based on the fusion of large models and visual recognition specifically includes: S110. Collect multimodal perception data of the cruise ship, including passenger visual data, passenger positioning data and cruise ship environmental data. Based on the cruise ship environmental data, construct a dynamic risk field in the digital twin space of the cruise ship. The dynamic risk field is used to characterize the dynamic risk intensity of different locations in the cruise ship space as time and environment change.

[0020] This application provides cruise ship passenger safety early warning by collecting multi-dimensional perception data through a sensor network integrated into the intelligent safety early warning device. This sensor network includes visual cameras, infrared thermal imagers, millimeter-wave radar, Wi-Fi / Bluetooth positioning beacons, anemometers, ship attitude sensors, and microphone arrays distributed across various collection areas, thereby continuously collecting multi-modal perception data covering the entire ship. The multi-modal perception data includes real-time video data and positioning coordinates collected from corresponding cruise passengers; the real-time video data is defined as passenger visual data, and the positioning coordinates are defined as passenger positioning data. In addition, information such as ship rolling amplitude, deck slippage, and environmental visibility can be collected through triaxial sensors, temperature and humidity sensors, and visual sensors as cruise ship environmental data. Furthermore, depending on actual needs, real-time weather data can be obtained from a weather platform based on the current cruise ship location as cruise ship environmental data.

[0021] Subsequently, based on this environmental data, the device dynamically calculates and generates a dynamic risk field within its internal digital twin space of the cruise ship. The cruise ship digital twin space is a pre-built virtual 3D spatial model of the cruise ship, mapping its actual physical layout, including various cabins, corridors, decks, ship's sides, and observation decks. The dynamic risk field is a virtual risk intensity distribution map, dynamically adjusting the risk intensity of each area based on different multimodal sensing data. For example, when the device detects that the ship's rolling amplitude exceeds a safety threshold, it increases the risk field intensity of all open ship's sides and observation deck areas; when the system time enters late at night and weather data reports rainfall, it increases the risk intensity of relevant areas based on the time anomaly and slippery ground conditions. Thus, by dynamically adjusting the risk intensity of each area of ​​the cruise ship in conjunction with cruise ship environmental data, a dynamic risk field characterizing the risk intensity of different locations within the cruise ship space is constructed, providing a foundational data containing contextual information for subsequent early warning analysis.

[0022] Specifically, a dynamic risk field is constructed in the digital twin space of a cruise ship based on cruise ship environmental data, including: Based on cruise ship environmental data, the basic risk intensity of predefined static risk areas in the cruise ship digital twin space is adjusted to generate a corresponding dynamic risk field. Cruise ship environmental data includes at least one of weather, wind speed, ship roll, visibility and time information.

[0023] When constructing a dynamic risk field, the initial value of the intelligent safety early warning equipment is the basic risk intensity of the static risk areas predefined in the cruise ship's digital twin space. By pre-configuring the basic risk intensity based on the inherent hazardous attributes of the cruise ship's physical space, the basic risk intensity can be set based on the ship's structural design and historical safety data. For example, areas such as the ship's side, observation decks, pool edges, engine room entrances, and children's play areas are marked as basic risk areas of different levels, and each area is assigned an initial, static risk intensity value. For instance, the basic risk value for a relatively safe indoor corridor is set to 1, while the basic risk value for the open top deck is preset to 5. The cruise ship environmental data collected in real time by the sensor network is used as a parameter for dynamically adjusting the static basic risk intensity. By pre-setting a mapping function, changes in each environmental parameter are converted into a modulation factor for the risk intensity of the corresponding static area.

[0024] For example, a dynamic risk field is represented as: DRF(x,y,t) = SBR(x,y) × [1 + α1·W(t) + α2·R(t) + α3·V(t) + α4·T(t)] × E(x,y,t) Wherein, DRF(x,y,t) is the dynamic risk field intensity, representing the real-time risk intensity value at time t and spatial location (x,y). The dynamic risk field is a continuously changing scalar field that can be defined in the digital twin space; SBR(x,y) is the static basic risk intensity, which is a fixed basic risk intensity predefined based on the structural characteristics of the cruise ship, for example: ordinary corridor = 1.0, indoor public area = 1.5, open deck = 3.0, ship side area = 5.0, high-altitude observation deck = 6.0; W(t) is the wind speed modulation factor, representing the normalized ratio of real-time wind speed to the safety threshold, W(t) = min(current wind speed / safe wind speed threshold, 2.0), which mainly affects the dynamic risk of open areas; R(t) is the hull roll modulation factor, representing the comprehensive influence coefficient based on the ship's roll angle. V(t) is the visibility modulation factor, representing the inverse influence coefficient of environmental visibility. V(t) = 1 - min(current visibility / standard visibility, 1.0). It can be configured with a higher value under low visibility conditions such as fog, rain, and nighttime. T(t) is the time modulation factor, which is adjusted based on the periodic risk of system time. For example: daytime (6:00-18:00) = 0.2, evening (18:00-22:00) = 0.5, late night (22:00-6:00) = 1.0; α1, α2, α3, α 4 represents the environmental parameter weighting coefficient, indicating the relative importance weight of each environmental factor. Typical values ​​are α1=0.3 (wind speed), α2=0.4 (sway), α3=0.2 (visibility), and α4=0.3 (time). These values ​​can be adjusted adaptively based on the specific characteristics of the cruise ship and historical data. E(x,y,t) is the event amplification factor, representing the amplification of local risks based on temporary safety events. For example, after a conflict occurs in a certain area, the E value of that area temporarily increases to 1.5-2.0 and then decays exponentially over time.

[0025] Ultimately, by running the aforementioned modulation algorithm in real time, the equipment transforms the basic risk intensity of the static risk area into a dynamic risk field that is continuously distributed across the entire ship's digital twin space, with its intensity fluctuating over time and depending on the cruise ship environment.

[0026] S120. Based on passenger visual data and passenger location data, perform behavior recognition, trajectory tracking and social relationship analysis on target passengers to generate corresponding behavioral dimension risk parameters, spatiotemporal dimension risk parameters and social dimension risk parameters.

[0027] Furthermore, after constructing the dynamic risk field of the cruise ship, the device analyzes each passenger based on passenger visual data and passenger location data, analyzing the passenger's risk status from three dimensions: behavior, spatiotemporal, and social, defining the currently analyzed passenger as the target passenger. In the behavior dimension, deep learning models (such as CNNs) can be used to analyze video frame sequences of passenger visual data, identifying various predefined behavioral patterns (such as running, climbing, waving, falling, pushing and shoving, and climbing over railings), and assigning a quantified risk coefficient to each identified behavioral pattern. Prior to this, video frame sequences of different behavioral patterns are collected for model training, enabling the device to recognize these behavioral patterns. Furthermore, by pre-constructing the mapping relationship between different behavioral patterns and corresponding behavioral dimension risk parameters, the model can determine the corresponding behavioral dimension risk parameters based on the identified behavioral patterns.

[0028] In the spatiotemporal dimension, the device uses a data fusion algorithm to calibrate the continuous positions tracked by visual tracking with the absolute coordinates of wireless signals, forming a passenger movement trajectory. Specifically, by fusing the continuous position trajectories from visual recognition with the positioning signals obtained from wireless positioning, a continuous sequence of passenger positions is constructed. Using multi-target tracking algorithms from computer vision, target passengers are continuously locked onto in the video stream, while area positioning calibration is performed via Wi-Fi / Bluetooth beacons, thus forming the passenger movement trajectory. This trajectory data is mapped in real-time into the cruise ship's digital twin space to analyze passenger movement speed, direction, and path anomalies, and to spatially correlate them with predefined danger zones in real time, thereby providing passengers' location context information for spatiotemporal risk analysis.

[0029] Specifically, by comparing this trajectory in real time with the marked hazardous areas (such as machine rooms and non-open ship rails) and time policies (such as access restrictions for children's areas at night), it can be determined whether the trajectory contains predefined risk situations (such as abnormal intrusion, prolonged loitering, or trajectory abrupt changes). For example, if a passenger is detected repeatedly going back and forth in a non-open ship rail area and staying for more than a preset time, it is determined to be a prolonged loitering risk. Similarly, by pre-setting the mapping relationship between different types of risks and spatiotemporal risk parameters, the spatiotemporal risk parameters of the target passenger can be determined based on the identified corresponding types of risks.

[0030] In the social dimension, computer vision recognition is used to continuously calculate the social status information of the target passenger and the surrounding people, thereby dynamically constructing and updating the passenger's real-time social status (such as predefined statuses such as companionship, conflict, or isolation), and outputting corresponding social risk parameters.

[0031] Among them, reference Figure 2 Social relationship analysis includes: S1201. Based on passenger visual data and passenger positioning data, determine the relative distance, direction of movement and interaction posture of the target passenger and other passengers in the surrounding area; S1202. Determine the social status of the target passenger based on the relative distance, direction of movement and interaction posture, and generate corresponding social dimension risk parameters for the target passenger based on the social status determination results. The social status determination results include whether the target passenger is in an isolated state, a companion state, or a conflict state.

[0032] When performing social relationship analysis, the intelligent safety early warning device first uses previously collected passenger visual and location data to calculate, in parallel, characteristic parameters such as the relative distance, direction of movement, and interactive posture between the passenger and other passengers. Specifically, for relative distance, the device calculates the instantaneous Euclidean distance between the target passenger and all nearby individuals within their field of vision using triangulation and visual depth information, based on passenger visual and location data. It then determines the average distance quantile between the target passenger and the surrounding population using statistical methods, thus quantifying whether the passenger is spatially isolated or clustered. For direction of movement, the device extracts the movement vectors of all tracked individuals from consecutive frames of visual data and calculates the cosine of the angle between the target passenger's direction of movement and the average direction of movement of the surrounding population. This cosine serves as a consistency coefficient representing the target passenger's direction of movement with other passengers, determining whether they are in a cooperative, companion-like state or in an unrelated, random state. For interactive posture, the device uses a pre-trained deep learning model to analyze the skeletal keypoint sequences in the visual data, identifying specific body posture patterns that represent positive or negative social interactions (such as standing face-to-face, waving, hugging, pushing, grabbing, etc.). After extracting the aforementioned features, the device determines the social status based on these three types of parameters according to predefined rules, generating a final social status assessment result. For example, if the relative distance between the target passenger and any other person is consistently greater than a predefined social distance threshold, the consistency coefficient of movement direction is lower than a predefined value, and no positive interactive postures are detected, the passenger is determined to be in an isolated state. If the passenger consistently maintains a distance less than the travel distance threshold with one or more specific passengers, the consistency coefficient of movement direction is consistently higher than a predefined value, and interactive postures such as conversation or travel are intermittently detected, the passenger is determined to be in a travel-to-travel state. If the distance between the target passenger and others suddenly and drastically shrinks to within a predefined distance, the movement vectors show a relative or chasing posture, and visual analysis clearly identifies conflict postures such as pushing or punching, the passenger is determined to be in a conflict state. Finally, based on this social status assessment result, the device assigns higher social dimension risk parameters to passengers in isolated and conflict states from a predefined mapping table, while assigning lower risk parameters to passengers in travel-to-travel states.

[0033] S130. Based on the dynamic risk field, behavioral dimension risk parameters, spatiotemporal dimension risk parameters, and social dimension risk parameters, calculate the dynamic risk entropy value of the target passenger. When the dynamic risk entropy value reaches the set risk threshold, perform behavioral chain deduction on the target passenger based on the specified large model to predict the risk behavior chain sequence of the target passenger. Based on the risk behavior chain sequence and the dynamic risk entropy value, output the safety warning result.

[0034] After determining the real-time behavioral, spatiotemporal, and social risk parameters of the target passenger, a weighted fusion of these three dimensions is performed to obtain a preliminary baseline risk value. This baseline risk value is then combined with the risk intensity value of the dynamic risk field at the target passenger's precise location to derive the final dynamic risk entropy value. In other words, the same behavior, after being superimposed with the dynamic risk field, results in different final dynamic risk entropy values ​​for passengers in relatively safe and dangerous areas. For example, the dynamic risk entropy value of the same behavior differs in a stable corridor area and in a swaying ship's side area, thus accurately reflecting the high-risk entropy value in dangerous environments.

[0035] Furthermore, when this entropy value exceeds the set risk threshold, the device will trigger a designated large model to initiate the behavior chain inference process. The designated large model can be a large model trained on a large language model for behavior chain inference, or a regular large model can be used directly, with corresponding prompt word templates configured to instruct the large language model to output behavior chain inference results that meet the requirements.

[0036] The device combines the target passenger's current state (e.g., "a passenger who just had an argument in a bar is walking alone and quickly towards the upper deck"), fragments of their historical trajectory, and their current dynamic risk entropy value into a structured contextual cue, which is then input into a large model. Based on its reasoning ability and common sense trained on a large amount of human behavior data, the large model outputs multiple possible future behavior chains and their probabilities, such as: "80% probability: encountering the follower again in a corner of the deck, resulting in a physical conflict; 20% probability: exhibiting dangerous tendencies due to emotional instability."

[0037] Ultimately, based on the aforementioned in-depth analysis, the intelligent safety early warning equipment executes a precise, tiered early warning and intervention process. By comprehensively considering the urgency of the current risk as represented by the dynamic risk entropy value, and the predicted future risk evolution path and severity by the risk behavior chain sequence, it generates the final early warning instruction. For example, in extreme cases where the entropy value is extremely high (reaching the corresponding threshold) and the behavior chain clearly points to dangerous behavior (such as jumping into the sea), the equipment will immediately trigger the highest level alarm. It will then automatically activate the alarm devices in the incident area to issue audible and visual alarms, and simultaneously push complete early warning data, including the target passenger's real-time location, dynamic video, risk entropy value, and the conclusions of the large-scale model, to the security center's large screen and the mobile terminals of the nearest patrol personnel with the highest priority, thereby achieving a timely and predictive safety early warning effect.

[0038] Optionally, refer to Figure 3 Based on dynamic risk field, behavioral risk parameters, spatiotemporal risk parameters, and social risk parameters, the dynamic risk entropy value of the target passenger is calculated, including: S1301. Weighted fusion of behavioral dimension risk parameters, spatiotemporal dimension risk parameters and social dimension risk parameters is performed to obtain the basic risk value; S1302. Query the dynamic risk field based on the location of the target passenger, determine the real-time target risk field intensity of the target passenger, and calculate the dynamic risk entropy value of the target passenger based on the target risk field intensity and the dynamic risk field.

[0039] When calculating dynamic risk entropy, the intelligent safety early warning device first normalizes the risk parameters across three dimensions: behavior, spatiotemporal, and social. Then, it calculates the base risk value using a pre-defined weighted fusion function. Specifically, this function can be expressed as: BRS=w b ×P b +w s ×P s +w t ×P t Where P b P s P t These represent the standardized behavioral risk parameters, spatiotemporal risk parameters, and social risk parameters, respectively. b w s w t These are configurable weight coefficients for the corresponding dimensions, satisfying the constraint that the sum of the weights is 1. The BRS (Basic Risk Ratio) reflects a fixed, quantified level of risk corresponding to the target passenger's own behavior, ignoring the environmental context.

[0040] Furthermore, based on the real-time location coordinates of the target passenger, the device queries the dynamic risk field in the digital twin space of the cruise ship to obtain the DRF (Target Risk Field Intensity) at that location at the current moment. The final dynamic risk entropy value E risk The calculation formula using the multiplication model can be expressed as follows: E risk =BRS×(1+β×DRF) Where β is a gain coefficient used to control the degree of environmental risk impact. This formula represents the final risk entropy value E when a passenger exhibits high-risk behavior (i.e., high BRS value) and is simultaneously in a severe high-risk environment (i.e., high DRF value). risk The corresponding values ​​are also relatively high. For example, a passenger (isolated, medium-high BRS) lingering on the ship's side during calm seas (low DRF) has a risk entropy value E. risk The risk entropy is only moderate; however, when the ship rolls violently (DRF increases), assuming passenger behavior patterns remain unchanged (i.e., BRS remains unchanged), the final risk entropy value E... riskIt will also increase as the target risk field strength (DRF) increases.

[0041] The dynamic risk entropy value calculation based on weighted fusion can improve the system's quantification accuracy of risks in complex environments, enabling early warning equipment to make decisions that combine passenger behavior and risk environment to accurately quantify the real-time risk entropy value of passengers, thereby achieving more accurate early warning capabilities.

[0042] Optionally, refer to Figure 4 Based on a specified large model, behavioral chain deduction is performed on the target passengers to predict the risk behavioral chain sequence of the target passengers, including: S1303. Based on passenger visual data and passenger positioning data, collect real-time behavior, spatiotemporal location, social status and historical trajectory information of the target passenger to obtain contextual prompt information. S1304. Input the contextual hints into the specified large model, and output at least one risk behavior chain sequence with occurrence probability information based on the specified large model.

[0043] When performing behavioral chain analysis, the intelligent safety early warning device first extracts the latest visual behavior recognition results of the target passenger (such as behaviors identified based on passenger visual data), obtains the passenger's coordinates through location data, and analyzes the current social status (such as isolation) based on social relationship analysis. It then retrieves the passenger's recent (e.g., the past 30 minutes) historical trajectory sequence and behavioral records. This data is integrated and structured into a unified contextual prompt, which can be expressed using a combination of natural language and structured data. For example: The target passenger is an adult male who has been wandering alone in the deck area for the past 15 minutes, recently had a brief verbal argument with another passenger, and is currently moving at a relatively fast pace towards the non-open area at the stern.

[0044] The device then inputs this contextual information via an application programming interface (API) into a designated large model (such as a large language model) trained on a large amount of human behavior data and fine-tuned for cruise ship safety scenarios. Based on the received contextual information, this large model utilizes its internally stored knowledge and causal logic to perform multi-step reasoning and prediction. This results in an output sequence of risky behaviors with probability information. The behavior sequence describes the possible future event progression path; for example, output: Sequence A (probability: 65%): Encounter with a tailgating person at the stern, escalating to physical conflict; Sequence B (probability: 25%): Climbing over railings, attempting dangerous actions; Sequence C (probability: 10%): Entering the equipment area, triggering other alarms. Each predicted sequence is accompanied by a quantified probability of occurrence, calculated by the large model based on the matching degree and logical coherence of the input context.

[0045] The behavior chain extrapolation method based on large models enables the safety early warning system to have the ability to extrapolate passenger behavior and make predictive judgments, thereby realizing the transformation of the safety early warning system from passive response to active intervention and improving the level of safety management.

[0046] Optionally, based on the risk behavior chain sequence and dynamic risk entropy value, a safety warning result is output, including: The warning level base is determined based on the dynamic risk entropy value, and then the warning level base is corrected based on the danger level and probability of occurrence of the predicted behavior in the risk behavior chain sequence, generating a graded warning instruction as a safety warning result.

[0047] Based on the calculated dynamic risk entropy value, the device determines the warning level baseline through a preset mapping function. This baseline reflects the immediate risk level based on currently observable facts, such as being divided into different levels like "Attention Level," "Warning Level," and "Alarm Level," with each level corresponding to a different initial response strategy.

[0048] Subsequently, the device incorporates a risk behavior chain sequence derived from the large model's behavior chain to correct the baseline. The correction process comprehensively considers two dimensions of the risk behavior chain sequence: first, the danger level of the predicted behavior itself (e.g., different levels of conflict, climbing, jumping into the sea, etc.); and second, the probability of the behavior chain occurring. The device's built-in decision algorithm calculates a correction coefficient based on these two parameters. For predicted behaviors with high danger levels and high probability of occurrence, a larger positive correction weight is assigned, thereby increasing the final warning level; conversely, for predictions with low probability or low danger levels, a smaller adjustment is made or the original level is maintained.

[0049] Specifically, the correction process can be implemented using a weighted function, for example: Final warning level = Warning level base + γ × Σ (Danger level weight × Probability of occurrence) Here, γ is a configurable gain factor for the system. After this correction, the generated tiered warning instructions include both an assessment based on the passenger's current behavior and a prediction of future risk evolution, making the output safety warning results more accurate and forward-looking. For example, if a passenger's dynamic risk entropy value only reaches the warning level, but their behavior chain deduction shows a high probability of engaging in the high-risk behavior of climbing railings, the device will raise the warning level to the highest alarm level through a correction mechanism, thereby triggering more proactive intervention measures.

[0050] By combining real-time risk entropy values ​​with future behavior chain predictions and introducing time-dimensional risk assessment, the early warning mechanism enables safety early warning devices to not only respond to apparent risks but also provide early warnings of potential high-risk evolution paths, significantly improving the predictability and accuracy of safety early warnings.

[0051] Optionally, after generating the tiered warning instruction as a safety warning result, the following may also be included: Based on the graded warning instructions from the safety warning results, the corresponding proactive intervention measures will be implemented.

[0052] After generating tiered early warning commands as security alert results, the intelligent security early warning device immediately initiates the corresponding proactive intervention execution process. Different levels of early warning commands are mapped to preset, differentiated combinations of response measures, forming a complete closed loop from early warning to action. Specifically, when the system determines the early warning level to L1 (attention level), the device executes minimal intervention measures, mainly including prominently marking the target passenger in the background monitoring interface and pushing its dynamic risk entropy curve and real-time video stream to the security personnel's monitoring workstation, reminding personnel to remain attentive. At this time, no on-site intervention is performed to avoid unnecessary interference to the passenger. When the early warning escalates to L2 (early warning level), the device's response measures are upgraded accordingly. At this time, the device automatically pushes the early warning information, the target passenger's real-time location, physical characteristics, and a brief risk overview to the mobile terminals of one or more nearest security personnel through the dispatch system, and suggests that they go to the site for verification and inquiry, achieving precise intervention.

[0053] When the dynamic risk entropy value or behavioral chain deduction result triggers the highest level L3 (alarm level) warning, the device will initiate comprehensive proactive intervention. First, the device immediately activates intelligent audio-visual equipment in the target area, such as triggering flashing lights at a specific frequency to deter the behavior, and playing pre-recorded warning messages via directional acoustic devices, such as "Please be careful and stay away from dangerous areas." Simultaneously, it sends a highest-priority alarm to the security command center and all relevant personnel's terminals. This alarm not only includes the target passenger's real-time status but also includes behavioral chain deduction conclusions generated by a large model, such as "High risk: Predicted target intends to climb the railing within 60 seconds," providing crucial decision support for on-site handling. This timely curbs high-risk behaviors, ensuring the accuracy and effectiveness of the safety warning system's intervention.

[0054] The above-mentioned process involves collecting multimodal perception data from the cruise ship, including passenger visual data, passenger positioning data, and cruise ship environmental data. A dynamic risk field is constructed within the cruise ship's digital twin space based on the cruise ship environmental data. This dynamic risk field characterizes the dynamic risk intensity at different locations within the cruise ship space as time and the environment change. Based on passenger visual and positioning data, behavior recognition, trajectory tracking, and social relationship analysis are performed on target passengers to generate corresponding behavioral, spatiotemporal, and social dimension risk parameters. Based on the dynamic risk field, behavioral, spatiotemporal, and social dimension risk parameters, the dynamic risk entropy value of the target passenger is calculated. When the dynamic risk entropy value reaches a set risk threshold, a behavioral chain deduction is performed on the target passenger based on a specified large-scale model to predict the risk behavior chain sequence of the target passenger. Based on the risk behavior chain sequence and the dynamic risk entropy value, a safety warning result is output. By employing the aforementioned technical means, and introducing a dynamic risk field that integrates environmental factors, combined with the fusion analysis of multi-dimensional risk parameters related to passenger behavior, spatiotemporal factors, and social interactions, as well as behavioral chain extrapolation based on a large model, the safety early warning system can identify passenger risk context information, quantify passenger risk entropy values, and predict risk evolution paths. This achieves a shift from passively responding to isolated events to proactively predicting and intervening in complex risks. Compared to existing early warning schemes that rely on fixed rules and have uncorrelated data, this application improves the timeliness and accuracy of safety early warnings and significantly reduces the false alarm rate, meeting the need for timely early warnings for passengers in the complex and enclosed environment of cruise ships.

[0055] Example 2: Based on the above embodiments, Figure 5 This is a schematic diagram of a cruise ship passenger safety intelligent early warning system based on the fusion of large model and visual recognition, provided in Embodiment 2 of this application. (Reference) Figure 5 The intelligent early warning system for cruise passenger safety based on the fusion of large models and visual recognition provided in this embodiment specifically includes: The risk field construction module 21 is used to collect multimodal perception data of the cruise ship. The multimodal perception data includes passenger visual data, passenger positioning data and cruise ship environmental data. Based on the cruise ship environmental data, a dynamic risk field in the digital twin space of the cruise ship is constructed. The dynamic risk field is used to characterize the dynamic risk intensity of different locations in the cruise ship space as time and environment change. The parameter analysis module 22 is used to perform behavior recognition, trajectory tracking and social relationship analysis on target passengers based on passenger visual data and passenger location data, and generate corresponding behavioral dimension risk parameters, spatiotemporal dimension risk parameters and social dimension risk parameters. The early warning module 23 is used to calculate the dynamic risk entropy value of the target passenger based on the dynamic risk field, behavioral dimension risk parameters, spatiotemporal dimension risk parameters and social dimension risk parameters. When the dynamic risk entropy value reaches the set risk threshold, it performs behavioral chain deduction on the target passenger based on the specified large model, predicts the risk behavior chain sequence of the target passenger, and outputs the safety early warning result based on the risk behavior chain sequence and the dynamic risk entropy value.

[0056] Specifically, a dynamic risk field is constructed in the digital twin space of a cruise ship based on cruise ship environmental data, including: Based on cruise ship environmental data, the basic risk intensity of predefined static risk areas in the cruise ship digital twin space is adjusted to generate a corresponding dynamic risk field. Cruise ship environmental data includes at least one of weather, wind speed, ship roll, visibility and time information.

[0057] Specifically, social relationship analysis includes: Based on passenger visual data and passenger location data, the relative distance, direction of movement, and interaction posture of the target passenger and other passengers in the surrounding area are determined. The social status of the target passenger is determined based on the relative distance, direction of movement, and interaction posture. Based on the social status determination result, corresponding social dimension risk parameters are generated for the target passenger. The social status determination result includes whether the target passenger is in an isolated state, a companion state, or a conflict state.

[0058] Specifically, based on the dynamic risk field, behavioral risk parameters, spatiotemporal risk parameters, and social risk parameters, the dynamic risk entropy value of the target passenger is calculated, including: The basic risk value is obtained by weighting and fusing the risk parameters of the behavioral dimension, the spatiotemporal dimension, and the social dimension. The dynamic risk field is queried based on the target passenger's location to determine the real-time target risk field intensity of the target passenger. Based on the target risk field intensity and the dynamic risk field, the dynamic risk entropy value of the target passenger is calculated.

[0059] Specifically, based on a specified large model, behavioral chain extrapolation is performed on the target passengers to predict the risky behavioral chain sequence of the target passengers, including: Based on passenger visual data and passenger location data, real-time behavior, spatiotemporal location, social status and historical trajectory information of target passengers are collected to obtain contextual prompts. Input contextual hints into a specified large model, and output at least one risk behavior chain sequence with probability information based on the specified large model.

[0060] Specifically, based on the risk behavior chain sequence and dynamic risk entropy value, the system outputs safety warning results, including: The warning level base is determined based on the dynamic risk entropy value, and then the warning level base is corrected based on the danger level and probability of occurrence of the predicted behavior in the risk behavior chain sequence, generating a graded warning instruction as a safety warning result.

[0061] Specifically, after generating tiered early warning instructions as a security warning result, the process also includes: Based on the graded warning instructions from the safety warning results, the corresponding proactive intervention measures will be implemented.

[0062] The above-mentioned process involves collecting multimodal perception data from the cruise ship, including passenger visual data, passenger positioning data, and cruise ship environmental data. A dynamic risk field is constructed within the cruise ship's digital twin space based on the cruise ship environmental data. This dynamic risk field characterizes the dynamic risk intensity at different locations within the cruise ship space as time and the environment change. Based on passenger visual and positioning data, behavior recognition, trajectory tracking, and social relationship analysis are performed on target passengers to generate corresponding behavioral, spatiotemporal, and social dimension risk parameters. Based on the dynamic risk field, behavioral, spatiotemporal, and social dimension risk parameters, the dynamic risk entropy value of the target passenger is calculated. When the dynamic risk entropy value reaches a set risk threshold, a behavioral chain deduction is performed on the target passenger based on a specified large-scale model to predict the risk behavior chain sequence of the target passenger. Based on the risk behavior chain sequence and the dynamic risk entropy value, a safety warning result is output. By employing the aforementioned technical means, and introducing a dynamic risk field that integrates environmental factors, combined with the fusion analysis of multi-dimensional risk parameters related to passenger behavior, spatiotemporal factors, and social interactions, as well as behavioral chain extrapolation based on a large model, the safety early warning system can identify passenger risk context information, quantify passenger risk entropy values, and predict risk evolution paths. This achieves a shift from passively responding to isolated events to proactively predicting and intervening in complex risks. Compared to existing early warning schemes that rely on fixed rules and have uncorrelated data, this application improves the timeliness and accuracy of safety early warnings and significantly reduces the false alarm rate, meeting the need for timely early warnings for passengers in the complex and enclosed environment of cruise ships.

[0063] The intelligent early warning system for cruise passenger safety based on the fusion of large model and visual recognition provided in Embodiment 2 of this application can be used to execute the intelligent early warning method for cruise passenger safety based on the fusion of large model and visual recognition provided in Embodiment 1 above, and has the corresponding functions and beneficial effects.

[0064] Example 3: Embodiment 3 of this application provides an electronic device, which is described below. Figure 6The electronic device includes a processor 31, a memory 32, a communication module 33, an input device 34, and an output device 35. The electronic device may have one or more processors and one or more memories. The processor, memory, communication module, input device, and output device of the electronic device can be connected via a bus or other means.

[0065] Memory, as a computer-readable storage medium, can be used to store software programs, computer-executable programs, and modules, such as the program instructions / modules corresponding to the cruise passenger safety intelligent early warning method based on large model and visual recognition fusion described in any embodiment of this application (e.g., various modules in the cruise passenger safety intelligent early warning system based on large model and visual recognition fusion). Memory may primarily include a program storage area and a data storage area, wherein the program storage area may store the operating system and at least one application program required for a function; the data storage area may store data created based on device usage, etc. Furthermore, memory may include high-speed random access memory and non-volatile memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some instances, memory may further include memory remotely located relative to the processor, and these remote memories can be connected to the device via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0066] The communication module is used for data transmission.

[0067] The processor executes various functional applications and data processing of the device by running software programs, instructions, and modules stored in memory, thereby realizing the above-mentioned intelligent early warning method for cruise passenger safety based on the fusion of large models and visual recognition.

[0068] Input devices can be used to receive input numerical or character information, and to generate key signal inputs related to user settings and function control of the device. Output devices may include display devices such as displays.

[0069] The electronic device provided above can be used to execute the intelligent early warning method for cruise passenger safety based on the fusion of large model and visual recognition provided in Embodiment 1 above, and has corresponding functions and beneficial effects.

[0070] Example 4: This application embodiment also provides a storage medium containing computer-executable instructions. When executed by a computer processor, these instructions are used to execute a cruise ship passenger safety intelligent early warning method based on the fusion of large-scale modeling and visual recognition. This method includes: collecting multimodal perception data from the cruise ship, including passenger visual data, passenger positioning data, and cruise ship environmental data; constructing a dynamic risk field in the cruise ship's digital twin space based on the cruise ship environmental data; and characterizing the dynamic risk field as it represents the changes in different locations within the cruise ship space over time and with environmental changes. The system assesses the intensity of dynamic risk; based on passenger visual data and passenger location data, it performs behavior recognition, trajectory tracking, and social relationship analysis on target passengers to generate corresponding behavioral dimension risk parameters, spatiotemporal dimension risk parameters, and social dimension risk parameters; based on the dynamic risk field, behavioral dimension risk parameters, spatiotemporal dimension risk parameters, and social dimension risk parameters, it calculates the dynamic risk entropy value of the target passenger; when the dynamic risk entropy value reaches the set risk threshold, it performs behavior chain deduction on the target passenger based on a specified large model to predict the risk behavior chain sequence of the target passenger; based on the risk behavior chain sequence and the dynamic risk entropy value, it outputs a safety warning result.

[0071] Storage medium – any type of memory device or storage device. The term “storage medium” is intended to include: mounting media, such as CD-ROM, floppy disk, or magnetic tape devices; computer system memory or random access memory, such as DRAM, DDR RAM, SRAM, EDO RAM, Rambus RAM, etc.; non-volatile memory, such as flash memory, magnetic media (e.g., hard disk or optical storage); registers or other similar types of memory elements, etc. Storage medium may also include other types of memory or combinations thereof. Furthermore, storage medium may reside in a first computer system in which the program is executed, or it may reside in a different second computer system connected to the first computer system via a network (such as the Internet). The second computer system can provide program instructions to the first computer for execution. The term “storage medium” can include two or more storage media residing in different locations (e.g., in different computer systems connected via a network). Storage medium may store program instructions (e.g., specifically implemented as a computer program) executable by one or more processors.

[0072] Of course, the computer-executable instructions provided in the embodiments of this application are not limited to the intelligent early warning method for cruise passenger safety based on the fusion of large model and visual recognition as described above, but can also perform related operations in the intelligent early warning method for cruise passenger safety based on the fusion of large model and visual recognition provided in any embodiment of this application.

[0073] The cruise passenger safety intelligent early warning system, storage medium, and electronic device based on the fusion of large model and visual recognition provided in the above embodiments can execute the cruise passenger safety intelligent early warning method based on the fusion of large model and visual recognition provided in any embodiment of this application. For technical details not described in detail in the above embodiments, please refer to the cruise passenger safety intelligent early warning method based on the fusion of large model and visual recognition provided in any embodiment of this application.

[0074] The above description is merely a preferred embodiment and the technical principles employed in this application. This application is not limited to the specific embodiments described herein, and various obvious changes, readjustments, and substitutions that can be made by those skilled in the art will not depart from the scope of protection of this application. Therefore, although this application has been described in detail through the above embodiments, this application is not limited to the above embodiments, and may include more other equivalent embodiments without departing from the concept of this application, the scope of which is determined by the scope of the claims.

Claims

1. A smart early warning method for cruise passenger safety based on the fusion of large models and visual recognition, characterized in that, include: Multimodal perception data of the cruise ship is collected, including passenger visual data, passenger positioning data and cruise ship environmental data. Based on the cruise ship environmental data, a dynamic risk field in the digital twin space of the cruise ship is constructed. The dynamic risk field is used to characterize the dynamic risk intensity of different locations in the cruise ship space as time and environment change. Based on the passenger visual data and the passenger location data, the target passenger's behavior is identified, trajectory is tracked and social relationship is analyzed to generate corresponding behavioral dimension risk parameters, spatiotemporal dimension risk parameters and social dimension risk parameters. Based on the dynamic risk field, the behavioral dimension risk parameters, the spatiotemporal dimension risk parameters, and the social dimension risk parameters, the dynamic risk entropy value of the target passenger is calculated. When the dynamic risk entropy value reaches the set risk threshold, the behavioral chain of the target passenger is extrapolated based on a specified large model to predict the risk behavior chain sequence of the target passenger. Based on the risk behavior chain sequence and the dynamic risk entropy value, a safety warning result is output.

2. The intelligent early warning method for cruise passenger safety based on the fusion of large model and visual recognition as described in claim 1, characterized in that, The construction of a dynamic risk field in the cruise ship digital twin space based on the cruise ship environmental data includes: Based on the cruise ship environmental data, the basic risk intensity of the predefined static risk area in the cruise ship digital twin space is adjusted to generate a corresponding dynamic risk field. The cruise ship environmental data includes at least one of weather, wind speed, ship roll, visibility and time information.

3. The intelligent early warning method for cruise passenger safety based on the fusion of large model and visual recognition as described in claim 1, characterized in that, The social relationship analysis includes: Based on the passenger visual data and the passenger positioning data, the relative distance, direction of movement, and interaction posture between the target passenger and other passengers in the surrounding area are determined. The social status of the target passenger is determined based on the relative distance, the direction of movement, and the interactive posture. Corresponding social dimension risk parameters are generated for the target passenger based on the social status determination result. The social status determination result includes whether the target passenger is in an isolated state, a companion state, or a conflict state.

4. The intelligent early warning method for cruise passenger safety based on the fusion of large model and visual recognition as described in claim 1, characterized in that, The calculation of the target passenger's dynamic risk entropy value based on the dynamic risk field, the behavioral dimension risk parameters, the spatiotemporal dimension risk parameters, and the social dimension risk parameters includes: The basic risk value is obtained by weighting and fusing the behavioral dimension risk parameter, the spatiotemporal dimension risk parameter, and the social dimension risk parameter. The dynamic risk field is queried based on the location of the target passenger to determine the real-time target risk field intensity of the target passenger, and the dynamic risk entropy value of the target passenger is calculated based on the target risk field intensity and the dynamic risk field.

5. The intelligent early warning method for cruise passenger safety based on the fusion of large model and visual recognition as described in claim 1, characterized in that, The step of performing behavioral chain extrapolation on the target passenger based on a specified large model to predict the risk behavior chain sequence of the target passenger includes: Based on the passenger visual data and the passenger location data, real-time behavior, spatiotemporal location, social status and historical trajectory information of the target passenger are collected to obtain contextual prompt information; The contextual hints are input into a specified large model, and at least one risk behavior chain sequence with occurrence probability information is output based on the specified large model.

6. The intelligent early warning method for cruise passenger safety based on the fusion of large model and visual recognition as described in claim 1, characterized in that, The step of outputting a safety warning result based on the risk behavior chain sequence and the dynamic risk entropy value includes: The warning level base is determined based on the dynamic risk entropy value, and the warning level base is corrected based on the danger level and probability of occurrence of the predicted behavior in the risk behavior chain sequence, generating a graded warning instruction as a safety warning result.

7. The intelligent early warning method for cruise passenger safety based on the fusion of large model and visual recognition as described in claim 6, characterized in that, After generating the graded early warning instruction as a safety early warning result, the following is also included: Based on the graded warning instructions of the aforementioned safety warning results, corresponding proactive intervention measures will be implemented.

8. A cruise ship passenger safety intelligent early warning system based on the fusion of large model and visual recognition, characterized in that, include: The risk field construction module is used to collect multimodal perception data of the cruise ship, including passenger visual data, passenger positioning data and cruise ship environmental data. Based on the cruise ship environmental data, a dynamic risk field in the digital twin space of the cruise ship is constructed. The dynamic risk field is used to characterize the dynamic risk intensity of different locations in the cruise ship space as time and environment change. The parameter analysis module is used to perform behavior recognition, trajectory tracking and social relationship analysis on the target passenger based on the passenger visual data and the passenger positioning data, and generate corresponding behavioral dimension risk parameters, spatiotemporal dimension risk parameters and social dimension risk parameters. The early warning module is used to calculate the dynamic risk entropy value of the target passenger based on the dynamic risk field, the behavioral dimension risk parameters, the spatiotemporal dimension risk parameters, and the social dimension risk parameters. When the dynamic risk entropy value reaches a set risk threshold, the module performs behavioral chain deduction on the target passenger based on a specified large model to predict the risk behavior chain sequence of the target passenger. Based on the risk behavior chain sequence and the dynamic risk entropy value, the module outputs a safety early warning result.

9. An electronic device, characterized in that, include: Memory and one or more processors; The memory is used to store one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the intelligent early warning method for cruise passenger safety based on the fusion of large models and visual recognition as described in any one of claims 1-7.

10. A storage medium containing computer-executable instructions, characterized in that, The computer-executable instructions, when executed by a computer processor, are used to perform the intelligent early warning method for cruise passenger safety based on the fusion of large models and visual recognition as described in any one of claims 1-7.

Images