Intelligent control rescue method, system and equipment for personnel falling into water and medium

Abstract

The invention relates to an intelligent control rescue method, system and equipment for people falling into water and a medium. The method comprises the following steps: firstly, collecting parameters from a water area environment to construct a multi-source dynamic data set; processing the data uncertainty by using a particle filtering algorithm based on the data set, determining the probability distribution of the initial position of the drowning person, and deducing a real-time trajectory prediction path; classifying data exception modes through a support vector machine, and determining a boundary coordinate set of the potential dangerous area in combination with spatial clustering and boundary detection; carrying out overlay analysis on the track and the dangerous area boundary, screening a conflict area, dynamically correcting a path, and calculating an interception priority based on the corrected track to determine a rescue interception position sequence; and if the interception position contains the environment complex variable, iteratively executing a particle filter algorithm to adjust a trajectory deviation vector, and generating a safe trajectory model. By adopting the method, the data uncertainty can be reduced to improve the positioning precision of the drowning person, a high-risk area is avoided, the interception effectiveness is improved, and the precise rescue requirement of a complex water area is met.

Description

Technical Field

[0001] This invention belongs to the field of water rescue technology, and in particular relates to an intelligent control rescue method, system, equipment and medium for people who have fallen into the water. Background Technology

[0002] In the field of water rescue technology, drowning accidents have become a major threat to global public safety due to their high incidence and suddenness. World Health Organization data shows that approximately 360,000 people die from drowning globally each year. Traditional rescue methods (such as manual search and inflatable boat interception) have inherent drawbacks, including long response times, significant interference from water flow and weather conditions, and limited rescue range. Existing technologies, methods for locating drowning victims based on single sensor data struggle to handle data uncertainty, relying on human experience for dangerous area identification is inefficient, and trajectory prediction does not fully integrate terrain features and dynamic environmental parameters, resulting in a lack of real-time performance and safety in rescue path planning. While the development of the Internet of Things, big data, and intelligent algorithms has led to the emergence of sensor network-based rescue systems, the lack of multi-source dynamic data fusion models and adaptive trajectory correction mechanisms still fails to meet the precise rescue needs in complex aquatic environments. Summary of the Invention

[0003] Therefore, it is necessary to provide a method, system, equipment, and medium for intelligent control rescue of people who have fallen into the water, which can improve the positioning accuracy, response speed, and path safety of water rescue and meet the needs of precise rescue in complex environments, in order to address the above-mentioned technical problems.

[0004] Firstly, this application provides an intelligent control rescue method for people who have fallen into the water, including:

[0005] Data on water flow velocity, topographic features, and meteorological parameters in the aquatic environment are acquired to construct a multi-source dynamic dataset.

[0006] Based on a multi-source dynamic dataset, a particle filter algorithm is used to handle data uncertainty, determine the probability distribution of the initial position of the person who fell into the water, and obtain the real-time trajectory prediction path based on the position distribution.

[0007] Support vector machines are used to classify abnormal patterns in multi-source dynamic datasets, and the set of boundary coordinates of potential hazardous areas is determined based on the classification results.

[0008] The real-time trajectory prediction path is overlaid with the boundary coordinate set to obtain the risk intersection point set. Based on the intersection point set, the trajectory prediction path is generated and the interception priority is calculated to determine the rescue interception location sequence.

[0009] If the rescue interception location sequence includes complex environmental variables, then the particle filter algorithm is iteratively executed to adjust the trajectory deviation vector to obtain a safe trajectory model; the complex environmental variables include terrain complexity and dynamic obstacle density.

[0010] In one embodiment, a particle filter algorithm is used to handle data uncertainty based on a multi-source dynamic dataset, determine the probability distribution of the initial position of the person who fell into the water, and obtain a real-time trajectory prediction path based on the position distribution, including:

[0011] The initial location data of the person who fell into the water was extracted from the multi-source dynamic dataset. The initial location data includes the latitude and longitude information of the person who fell into the water, the data acquisition timestamp, and the noise parameters of the corresponding sensor.

[0012] The particle filter algorithm is used to process the uncertainty introduced by sensor noise in the initial position data to obtain the probability distribution of the initial position of the person who fell into the water.

[0013] The drift velocity vector of the person who fell into the water is calculated based on the probability distribution of the initial position of the person and the multi-source dynamic dataset, and the predicted trajectory of the person's real-time position is determined based on the drift velocity vector.

[0014] Dynamic weights are generated based on the deviation characteristics of the predicted trajectory. The predicted trajectory is then adjusted using these dynamic weights to obtain an optimized trajectory path distribution.

[0015] If the confidence level of the optimized trajectory path distribution is lower than the preset confidence threshold, supplementary environmental data is extracted from the multi-source dynamic dataset. The supplementary environmental data includes ocean current velocity data and wind speed data above the water. The preset confidence threshold is set according to the rescue accuracy requirements.

[0016] Supplementary environmental data is input into the particle filter algorithm to update the state transition parameters, the corrected probability distribution of the drowning person's location is recalculated, and the updated real-time trajectory prediction path of the drowning person is determined.

[0017] In one embodiment, a particle filtering algorithm is used to process the uncertainty introduced by sensor noise in the initial position data to obtain the probability distribution of the initial position of the person who fell into the water, including:

[0018] Initial latitude and longitude measurements of the drowning victims based on multi-source dynamic datasets and sensor noise covariance ,generate The initial particle set is obtained from the initial particles. .

[0019]

[0020] in, Indicates the first The state vectors of particles, and satisfying , This indicates a normal distribution.

[0021] Based on the obtained initial particle set Combining initial aquatic environment micro-disturbance factors Calculate the corrected sensor noise covariance matrix for fusion environmental micro-perturbations. and for each particle Calculate the corresponding weights using the following formula. The particle weight set is obtained. .

[0022] exp

[0023] in, express The determinant, express The inverse matrix.

[0024] Based on particle weight set The weights of all particles are normalized to obtain a normalized weight set. .

[0025] Combined with the initial particle set with normalized weight set Discrete probability distribution constituting the initial position of the person who fell into the water .

[0026]

[0027] in, This represents the Dirac function.

[0028] In one embodiment, a support vector machine is used to classify abnormal patterns in a multi-source dynamic dataset, and the set of boundary coordinates of potentially hazardous areas is determined based on the classification results, including:

[0029] Extract anomaly pattern data containing time series features and spatial features from multi-source dynamic datasets; the anomaly pattern data includes the timestamp sequence of the anomaly occurrence, the latitude and longitude coordinates of the corresponding spatial location, and the values ​​of the anomaly environmental parameters.

[0030] The abnormal pattern data is input into the support vector machine model to obtain the abnormal pattern category labels; the abnormal pattern category labels include the eddy current zone, the rapid current zone, and the underwater obstacle influence zone.

[0031] Spatial clustering analysis is performed on anomaly pattern data corresponding to the same category label based on anomaly pattern category labels to determine the spatial distribution characteristics of potential hazard areas for each category; spatial distribution characteristics include the core coordinates of the hazard area, anomaly point density, and spatial spread trend.

[0032] Based on the spatial distribution characteristics, a boundary detection algorithm is used to fit the edge anomalies of the potential danger area, and the boundary coordinate set of the potential danger area is calculated. The boundary coordinate set is a sequence of latitude and longitude coordinates that constitute the polygonal outline of the danger area.

[0033] In one embodiment, the real-time trajectory prediction path is overlaid with the boundary coordinate set to obtain a risk intersection point set. Based on the intersection point set, a trajectory prediction path is generated, and the interception priority ranking is calculated to determine the rescue interception location sequence, including:

[0034] Spatial analysis algorithms are used to overlay and analyze the real-time trajectory prediction path and boundary coordinate set, identify spatial intersection points, and statistically analyze the spatial distribution density of intersection points to obtain the risk point density distribution.

[0035] The spatial conflict detection algorithm filters out areas in the risk point density distribution whose density value is higher than the risk density threshold. Adjacent high-risk density areas are merged into continuous areas to determine the conflict area set. The risk density threshold is preset according to the water rescue safety standard. The conflict area set includes the outline coordinates of each conflict area and the corresponding conflict level.

[0036] Based on the correction rules, a dynamic path correction scheme is generated by correcting the set of conflict areas. The original real-time trajectory prediction path is adjusted according to the scheme to obtain the corrected trajectory prediction path. The correction rules include avoiding conflict areas, minimizing trajectory offset, and adapting to the direction of water flow in the water area.

[0037] Based on the corrected trajectory prediction path, an interception priority evaluation index is set. The priority score of each candidate interception point is calculated by weighting, and the rescue interception location sequence is determined by sorting the scores from high to low. The priority evaluation index includes the distance between the interception point and the trajectory prediction location, the adaptability of the water environment at the interception point, and the accessibility of rescue equipment. The rescue interception location sequence includes the latitude and longitude coordinates of each interception point, the recommended interception time, and the priority score.

[0038] In one embodiment, the priority score of each candidate interception point is calculated using the following formula:

[0039]

[0040] in, This represents the priority score of the candidate interception point, with a value range of [value missing]. , This represents the straight-line distance between the candidate interception point and the corresponding predicted position at the time step in the corrected trajectory prediction path. 2R R represents the Earth's average radius. Indicates the latitude of the candidate interception point. , Indicates the latitude coordinates of the candidate interception point. The latitude represents the predicted location of the trajectory. , This indicates the latitude coordinates of the predicted trajectory location. This represents the difference in latitude between two points. , This represents the difference in longitude between two points. , Indicates the longitude of the candidate interception point. , Indicates the longitude coordinates of the candidate interception point. The longitude of the predicted trajectory location. , Represents the longitude coordinates of the predicted trajectory location. This indicates the maximum allowable distance between the preset interception point and the predicted trajectory location. This represents the environmental adaptability coefficient of the water body at the interception point, calculated from the water flow velocity at the interception point, and its value ranges from [value missing]. , This represents the accessibility score for rescue equipment, with a value range of [value missing]. , , , These represent the weighting coefficients for the distance between the interception point and the predicted trajectory location, the adaptability of the water environment at the interception point, and the accessibility of rescue equipment, respectively.

[0041] In one embodiment, if the rescue interception location sequence includes complex environmental variables, a particle filter algorithm is iteratively executed to adjust the trajectory deviation vector to obtain a safe trajectory model, including:

[0042] If the rescue interception location sequence includes environmental complexity variables, then extract the environmental complexity variables from the rescue interception location sequence; the environmental complexity variables include terrain complexity and dynamic obstacle density.

[0043] The particle filter algorithm is used to treat the complex environmental variables as constraints for particle state transitions, and the trajectory deviation vector is calculated. The trajectory deviation vector is the position offset between the corrected trajectory prediction path and the original prediction path.

[0044] The trajectory deviation vector is adjusted based on a preset safety offset threshold to generate an initial safe trajectory model; the safety offset threshold is set according to the mobility of the rescue equipment; the initial safe trajectory model is a predicted trajectory of the person who has fallen into the water to avoid the influence of complex environmental variables.

[0045] Based on the initial safe trajectory model, dynamic obstacle influence factors are extracted to determine trajectory optimization parameters. The dynamic obstacle influence factors include the comprehensive quantitative value of obstacle movement speed, volume, and trajectory intersection probability. The trajectory optimization parameters include the state update coefficient and weight adjustment threshold in the particle filter algorithm.

[0046] The trajectory optimization parameters are input into the particle filter algorithm to update the particle weight distribution in the algorithm, and the optimized trajectory deviation vector is obtained by recalculation.

[0047] The optimized trajectory deviation vector is judged based on the preset trajectory stability threshold. If the judgment result is that the trajectory stability threshold is met, a safe trajectory model is generated based on the vector. The trajectory stability threshold is the maximum allowable fluctuation value of the trajectory deviation between adjacent time steps. The safe trajectory model includes the safe drift path of the person who fell into the water with the timestamp.

[0048] Secondly, this application also provides an intelligent control and rescue system for people who have fallen into the water, the system comprising:

[0049] The dataset construction module is used to acquire water flow velocity data, terrain feature data, and meteorological parameters in the aquatic environment to construct a multi-source dynamic dataset.

[0050] The danger zone determination module is used to process data uncertainty based on a multi-source dynamic dataset using a particle filter algorithm, determine the probability distribution of the initial position of the person who fell into the water, and obtain the real-time trajectory prediction path based on the position distribution; it is also used to classify abnormal patterns in the multi-source dynamic dataset using a support vector machine, and determine the set of boundary coordinates of potential danger zones based on the classification results.

[0051] The rescue location determination module is used to overlay and analyze the real-time trajectory prediction path with the boundary coordinate set to obtain the risk intersection point set, generate the trajectory prediction path based on the intersection point set, and calculate the interception priority ranking to determine the rescue interception location sequence.

[0052] The trajectory model generation module is used to iteratively execute a particle filter algorithm to adjust the trajectory deviation vector and obtain a safe trajectory model if the rescue interception location sequence includes complex environmental variables. The complex environmental variables include terrain complexity and dynamic obstacle density.

[0053] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method described above.

[0054] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the aforementioned method.

[0055] The aforementioned intelligent control and rescue method, system, computer equipment, and storage medium for people falling into water first collect water flow velocity, terrain features, and meteorological parameters from the aquatic environment to construct a multi-source dynamic dataset containing spatiotemporal multidimensional information. Then, based on this dataset, a particle filtering algorithm is used to handle data uncertainty. By generating an initial particle set, calculating particle weights for fused environmental micro-perturbations, and performing normalization, the probability distribution of the initial position of the person falling into water is determined, and a real-time trajectory prediction path is derived by combining the water flow velocity vector. Subsequently, a support vector machine is used to classify abnormal patterns (such as eddies and rapids) in the dataset, and spatial clustering and boundary detection algorithms are used to determine the polygonal boundary coordinate set of potential danger zones. Next, the real-time trajectory prediction path and the danger zone boundaries are spatially overlaid for analysis, the density distribution of risk points is statistically analyzed, and conflict areas are screened. High-risk areas are avoided through dynamic path correction, and interception priorities are calculated based on the corrected trajectory to determine the rescue interception position sequence. If there are complex environmental variables such as terrain complexity or dynamic obstacle density in the interception position sequence, the particle filtering algorithm is iteratively executed, and a final safe trajectory model is generated by adjusting the trajectory deviation vector. This method integrates multi-dimensional data such as water flow, topography, and meteorology through the construction of a multi-source dynamic dataset. It effectively addresses uncertainties introduced by sensor noise and environmental disturbances using a particle filter algorithm, improving the initial location accuracy of the person in the water. Support vector machine-based anomaly pattern classification and spatial clustering analysis shorten the response time for identifying dangerous areas and control boundary coordinate calculation errors. The combination of trajectory prediction and dangerous area overlay analysis with a dynamic path correction mechanism allows for real-time avoidance of high-risk conflict areas. An interception priority algorithm considers distance, environmental adaptability, and equipment accessibility, enhancing the effectiveness of rescue interception locations. Iterative optimization of the particle filter for complex environmental variables controls trajectory deviation fluctuations, ultimately forming a safe trajectory model that meets the precise rescue needs in complex aquatic environments, providing a systematic technical solution for intelligent control and rescue. Attached Figure Description

[0056] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0057] Figure 1 A flowchart illustrating an intelligent control and rescue method for a person who has fallen into water, provided as an embodiment of the present invention;

[0058] Figure 2 This is a structural block diagram of an intelligent control and rescue system for people who have fallen into the water, provided as an embodiment of the present invention. Detailed Implementation

[0059] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0060] In one embodiment, such as Figure 1 As shown, this application provides an intelligent control rescue method for people who have fallen into the water, which may include the following steps:

[0061] Step S101: Obtain water flow velocity data, terrain feature data, and meteorological parameters in the aquatic environment to construct a multi-source dynamic dataset.

[0062] Specifically, by deploying a sensor network in water bodies (such as flow velocity sensors, underwater terrain radar, and weather stations), the system collects water flow velocity (including velocity magnitude and direction), terrain features (such as underwater elevation, slope, and obstacle distribution), and meteorological parameters (such as wind speed, wind direction, and visibility) in real time. The system also performs timestamp alignment, format standardization, and outlier preprocessing (such as removing noisy data that exceeds the physical reasonable range) on the collected multi-type data, and finally constructs a multi-source dynamic dataset containing spatiotemporal correlations.

[0063] Step S102: Based on the multi-source dynamic dataset, the particle filter algorithm is used to process the data uncertainty, determine the probability distribution of the initial position of the person who fell into the water, and obtain the real-time trajectory prediction path based on the position distribution.

[0064] Specifically, measurement data (including latitude and longitude, acquisition time, and sensor noise parameters) of the initial position of the person who fell into the water are extracted from a multi-source dynamic dataset. A particle filtering algorithm is used to generate a large number of initial particles to simulate possible position states. The sensor noise covariance matrix is ​​corrected by combining micro-disturbance factors of the aquatic environment (such as local water flow fluctuations), and the weight of each particle is calculated and normalized to form a discrete probability distribution of the initial position of the person who fell into the water. Based on this probability distribution, the water flow velocity vector and drift model (such as physical motion equations considering buoyancy and resistance) are fused to derive the position change of the person who fell into the water step by step, and a real-time trajectory prediction path including timestamps is obtained.

[0065] Step S103: Use a support vector machine to classify the abnormal patterns in the multi-source dynamic dataset, and determine the set of boundary coordinates of potential dangerous areas based on the classification results.

[0066] Furthermore, abnormal pattern data with time-series characteristics (such as the duration of sudden changes in flow velocity) and spatial characteristics (such as the latitude and longitude range of abnormal areas) are extracted from multi-source dynamic datasets. These data are then input into a pre-trained support vector machine model for classification, outputting abnormal category labels such as eddy current area, rapid current area, and underwater obstacle influence area. Spatial clustering (such as density-based clustering algorithm) is performed on abnormal data with the same label to determine the core coordinates and distribution range of each dangerous area. A boundary detection algorithm (such as contour fitting) is used to perform polygon fitting on the edge points of the clustering results, generating a set of boundary coordinates of potential dangerous areas composed of latitude and longitude coordinate sequences.

[0067] Step S104: Overlay the real-time trajectory prediction path with the boundary coordinate set to obtain the risk intersection point set, generate the trajectory prediction path based on the intersection point set, and calculate the interception priority ranking to determine the rescue interception location sequence.

[0068] Specifically, a spatial overlay analysis algorithm (such as intersection analysis in GIS) is used to perform spatial calculations on the real-time trajectory prediction path (a line feature composed of continuous latitude and longitude points) and the boundary coordinate set of the potential danger area (polygonal surface features), and the intersection point of the two is identified as the risk intersection point; the spatial distribution density of the intersection point is statistically analyzed, and high-risk sections with a density higher than a preset threshold are screened out; the original trajectory is corrected based on the rules of "avoiding high-risk sections, minimizing trajectory offset, and adapting to the direction of water flow" to obtain an optimized trajectory prediction path; for the candidate interception points on the corrected trajectory, the comprehensive scores of distance, environmental adaptability, and equipment accessibility are calculated by weighting, and the rescue interception location sequence is determined by sorting the scores.

[0069] Step S105: If the rescue interception location sequence includes environmental complexity variables, then the particle filter algorithm is iteratively executed to adjust the trajectory deviation vector to obtain a safe trajectory model; the environmental complexity variables include terrain complexity and dynamic obstacle density.

[0070] For each location in the rescue interception location sequence, terrain complexity (such as the slope standard deviation calculated based on terrain elevation data) and dynamic obstacle density (such as the number of ships and floating objects per unit area) are extracted. If any environmental complexity variable at any location exceeds the preset safety range, it is used as a constraint condition and input into the particle filter algorithm to recalculate the trajectory deviation vector (correcting the positional offset between the trajectory and the original trajectory). A safety offset threshold is set based on the mobility of the rescue equipment, the deviation vector is adjusted, and an initial safe trajectory model is generated. The influence factors of dynamic obstacles (such as moving speed and the probability of intersection with the trajectory) are further extracted, the optimization parameters of the particle filter are determined to update the particle weight distribution, and the optimized trajectory deviation vector is obtained through iterative calculation. When the vector meets the preset trajectory stability threshold (the deviation fluctuation between adjacent time steps is within the allowable range), the final safe trajectory model is output.

[0071] The aforementioned intelligent control and rescue method for people who have fallen into water first collects data on water flow velocity, terrain features, and meteorological parameters from the aquatic environment to construct a multi-source dynamic dataset. Based on this dataset, a particle filtering algorithm is used to handle data uncertainty, determine the probability distribution of the initial position of the person in the water, and derive a real-time trajectory prediction path. Then, a support vector machine is used to classify abnormal data patterns, and spatial clustering and boundary detection are combined to determine the set of boundary coordinates of potential dangerous areas. The trajectory is overlaid and analyzed with the dangerous area boundaries to filter conflict areas and dynamically correct the path. Based on the corrected trajectory, the interception priority is calculated to determine the sequence of rescue interception positions. If the interception position contains complex environmental variables such as terrain complexity and dynamic obstacle density, the particle filtering algorithm is iteratively executed to adjust the trajectory deviation vector, generating a final safe trajectory model. This method integrates multi-dimensional data, uses particle filtering to reduce data uncertainty to improve the positioning accuracy of people in the water, uses support vector machines to shorten the response time for dangerous area identification and control boundary errors, avoids high-risk areas and improves interception effectiveness through trajectory correction and priority algorithms, and iteratively optimizes and controls trajectory deviation to ultimately form a safe trajectory model that meets the needs of precise rescue in complex waters, providing a systematic intelligent rescue solution.

[0072] In one embodiment, a particle filter algorithm is used to process data uncertainty based on a multi-source dynamic dataset to determine the probability distribution of the initial position of the person who fell into the water. The real-time trajectory prediction path is then obtained based on the position distribution, which may include the following steps:

[0073] Step S201: Extract the initial location data of the person who fell into the water from the multi-source dynamic dataset; the initial location data includes the latitude and longitude information of the person who fell into the water, the data acquisition timestamp, and the noise parameters of the corresponding sensor.

[0074] Step S202: The particle filter algorithm is used to process the uncertainty introduced by sensor noise in the initial position data to obtain the probability distribution of the initial position of the person who fell into the water.

[0075] Preferably, sensor noise (such as GPS drift caused by electromagnetic interference, or the impact of water flow disturbance on underwater sensors) can introduce uncertainty into the initial position data. The particle filtering algorithm addresses this issue through a process of "particle simulation - weight calculation - state update": First, a large number of initial particles (typically 1000-5000) are generated, each representing a possible initial position state, and the particle distribution follows a probability distribution (such as a normal distribution) that matches the sensor noise parameters. Then, combined with environmental constraints in the multi-source dynamic dataset (such as the restriction of the drowning person's position by the direction of water flow), the weight of each particle is calculated (a higher weight indicates a higher degree of matching between the particle and the true position). Finally, all particle weights are normalized to form a discrete probability distribution of the drowning person's initial position—this distribution quantifies the probability that different positions are the true initial positions through the spatial distribution of particles and their corresponding weights, effectively reducing the interference of sensor noise on positioning.

[0076] Step S203: Calculate the drift velocity vector of the person who fell into the water based on the probability distribution of the initial position of the person and the multi-source dynamic dataset, and determine the predicted trajectory of the person's real-time position based on the drift velocity vector.

[0077] Step S204: Generate dynamic weights based on the deviation characteristics of the predicted trajectory, and adjust the predicted trajectory using the dynamic weights to obtain the optimized trajectory path distribution.

[0078] Step S205: If the confidence level of the optimized trajectory path distribution is lower than the preset confidence threshold, supplementary environmental data is extracted from the multi-source dynamic dataset; the supplementary environmental data includes ocean current velocity data and wind speed data above the water area; the preset confidence threshold is set according to the rescue accuracy requirements.

[0079] The preset confidence threshold is a trajectory reliability evaluation standard set according to the accuracy requirements of the rescue scenario (e.g., positioning accuracy within 10 meters for near-shore rescue, and relaxed to within 50 meters for offshore rescue). It is typically set between 0.7 and 0.9, with higher confidence indicating more reliable trajectory prediction. When the confidence of the optimized trajectory path distribution is lower than this threshold, it indicates that the current model input environmental data (e.g., only surface water flow velocity) is insufficient to support high-precision trajectory prediction. Key environmental variables need to be supplemented—ocean current velocity data (including the velocity difference between deep and surface currents, and ocean current vortex distribution) can correct the longitudinal drift error of the person in the water, and wind speed data above the water (combined with the air resistance formula) can adjust the horizontal drift velocity calculation results. The extraction of supplementary environmental data must be aligned with the timestamp of the initial location data to ensure temporal consistency.

[0080] Step S206: Input the supplementary environmental data into the particle filter algorithm to update the state transition parameters, recalculate the corrected probability distribution of the drowning person's location, and determine the updated real-time trajectory prediction path of the drowning person.

[0081] Specifically, the initial position data of the person who fell into the water is first extracted from a multi-source dynamic dataset. This data includes latitude and longitude information, collection timestamps, and noise parameters of the corresponding sensors. A particle filter algorithm is used to process the uncertainty introduced by sensor noise in the initial position data. By generating an initial particle set, calculating particle weights, and normalizing, the probability distribution of the initial position of the person who fell into the water is obtained. Based on this probability distribution, the drift velocity vector of the person who fell into the water (a comprehensive motion vector that integrates environmental factors such as water flow and terrain) is calculated in combination with the multi-source dynamic dataset. The predicted trajectory of the person's real-time position is then derived step by step based on the drift velocity vector. Dynamic weights are generated based on the deviation characteristics of the predicted trajectory (such as the dispersion of position deviations between adjacent time steps). The predicted trajectory is adjusted using dynamic weights to obtain an optimized trajectory path distribution. If the confidence level of the optimized trajectory path distribution is lower than a preset confidence threshold (this threshold is set according to the rescue accuracy requirements), supplementary environmental data containing ocean current velocity data and wind speed data above the water area is extracted from the multi-source dynamic dataset. This supplementary data is then input into the particle filter algorithm to update the state transition parameters, and the corrected probability distribution of the person's position is recalculated. Finally, the updated predicted real-time trajectory path of the person who fell into the water is determined.

[0082] This embodiment uses a particle filtering algorithm to quantify the initial position uncertainty caused by sensor noise, thereby reducing the initial positioning error of the person in the water. Based on probability distribution and drift velocity vector, trajectory prediction combined with a dynamic weight adjustment mechanism effectively improves the stability of the trajectory path distribution. By triggering the introduction of supplementary environmental data through confidence level judgment, the state transition parameters of the particle filtering model are dynamically updated, improving the real-time performance and accuracy of trajectory prediction in complex aquatic environments (such as sudden changes in ocean currents or sudden increases in wind speed). This ensures that the real-time trajectory prediction path of the person in the water can adapt to environmental changes, providing a high-precision location reference for subsequent rescue decisions.

[0083] In one embodiment, a particle filtering algorithm is used to process the uncertainty introduced by sensor noise in the initial position data to obtain the probability distribution of the initial position of the person who fell into the water. This may include the following steps:

[0084] Step S301: Initially measure the latitude and longitude of the person who fell into the water based on the multi-source dynamic dataset. and sensor noise covariance ,generate The initial particle set is obtained from the initial particles. .

[0085]

[0086] in, Indicates the first The state vectors of particles, and satisfying , This indicates a normal distribution.

[0087] Step S302, based on the obtained initial particle set Combining initial aquatic environment micro-disturbance factors Calculate the corrected sensor noise covariance matrix for fusion environmental micro-perturbations. and for each particle Calculate the corresponding weights using the following formula. The particle weight set is obtained. .

[0088] exp

[0089] in, express The determinant, express The inverse matrix.

[0090] Step S303, based on particle weight set The weights of all particles are normalized to obtain a normalized weight set. .

[0091] Step S304, combining the initial particle set with normalized weight set Discrete probability distribution constituting the initial position of the person who fell into the water .

[0092]

[0093] in, This represents the Dirac function.

[0094] This embodiment integrates sensor noise covariance and aquatic environmental perturbation factors to quantify and process sensor measurement errors and dynamic environmental disturbances, thereby reducing the root mean square error (RMSE) of the initial location of the person in the water. The particle weighting calculation model fully considers the coupled effects of measurement data uncertainty and environmental disturbances, improving positioning accuracy in complex water flow environments compared to traditional particle filtering algorithms based solely on noise covariance. The construction of a discrete probability distribution provides a probabilistic description of the initial position, offering a quantitative basis for subsequent trajectory prediction that includes uncertainty. This makes the positioning results more consistent with the objective scenario of "random errors in multi-source data" in actual water rescue operations, effectively solving the problem of insufficient reliability of single measurement data and providing a scientific probability distribution basis for subsequent trajectory prediction and dynamic correction.

[0095] In one embodiment, using a support vector machine to classify abnormal patterns in a multi-source dynamic dataset and determining the set of boundary coordinates of potentially hazardous areas based on the classification results may include the following steps:

[0096] Step S401: Extract anomaly pattern data containing time series features and spatial features from the multi-source dynamic dataset; the anomaly pattern data includes the timestamp sequence of the anomaly occurrence, the latitude and longitude coordinates of the corresponding spatial location, and the values ​​of the anomaly environmental parameters.

[0097] Step S402: Input the abnormal pattern data into the support vector machine model to obtain the abnormal pattern category labels; the abnormal pattern category labels include eddy current zone, rapid current zone and underwater obstacle influence zone.

[0098] Step S403: Based on the anomaly pattern category label, perform spatial clustering analysis on the anomaly pattern data corresponding to the same category label to determine the spatial distribution characteristics of potential hazardous areas for each category; the spatial distribution characteristics include the core coordinates of the hazardous area, the density of anomaly points, and the spatial spread trend.

[0099] Preferably, for data with the same anomaly pattern category label (such as eddy current zones), a density clustering algorithm (such as DBSCAN) is used for spatial clustering. By calculating the spatial distance and density threshold of data points, clustered anomalies are divided into the same region. This determines the core coordinates (cluster center), anomaly density (number of anomalies per unit area), and spatial spread trend (direction and extent of region expansion) of potential hazard areas. For example, the clustering results of rapid flow zones can reflect the concentrated location and diffusion direction of water flow anomalies.

[0100] Step S404: Based on the spatial distribution characteristics, a boundary detection algorithm is used to fit the edge anomalies of the potential danger area and calculate the boundary coordinate set of the potential danger area; the boundary coordinate set is the latitude and longitude coordinate sequence that constitutes the polygonal outline of the danger area.

[0101] Furthermore, based on spatial distribution characteristics, boundary detection algorithms (such as the convex hull algorithm or the Ramer-Douglas-Peucker algorithm) are used to fit polygons to the edge anomalies of the clustered regions. By calculating the spatial topological relationships of the edge points, internal points are eliminated while key contour points are retained, generating a polygonal boundary composed of a sequence of latitude and longitude coordinates. This set of boundary coordinates accurately describes the spatial extent of the hazardous area. For example, the polygonal contour of the eddy current zone can be directly used for subsequent trajectory and hazardous area overlay analysis, ensuring the accuracy of collision detection.

[0102] Specifically, firstly, abnormal pattern data is extracted from a multi-source dynamic dataset. This data includes time-series features (time stamp sequences of abnormal occurrences), spatial features (latitude and longitude coordinates of corresponding spatial locations), and abnormal environmental parameter values. The extracted abnormal pattern data is then input into a pre-trained support vector machine model. Through the model's mapping and classification of data features, corresponding abnormal pattern category labels are output, including eddy current zones, rapid current zones, and underwater obstacle influence zones. Based on the category labels of the same type of abnormal pattern, spatial clustering analysis (such as using density clustering algorithms) is performed on the corresponding abnormal pattern data. By calculating the spatial clustering degree and correlation strength of data points, the spatial distribution characteristics of potential hazard areas of each category are determined, including the core coordinates of the hazard area (the center position with the highest density), the density of abnormal points (the number of abnormal data points per unit area), and the spatial spread trend (the direction and range of data points spreading to the surrounding area). Based on the above spatial distribution characteristics, a boundary detection algorithm (such as a contour fitting algorithm) is used to perform polygon fitting on the abnormal points at the edge of the potential hazard area. By calculating the spatial topological relationship and geometric boundary of the edge points, a set of boundary coordinates of the potential hazard area composed of latitude and longitude coordinate sequences is finally obtained. This set accurately describes the polygonal contour range of the hazard area.

[0103] This embodiment achieves a multi-dimensional characterization of aquatic environmental anomalies by extracting time-series and spatial feature anomaly pattern data. The classification results of the support vector machine model improve the accuracy of anomaly pattern identification. Spatial clustering analysis based on category labels transforms scattered anomaly data points into dangerous areas with clear spatial relationships, which is more efficient than traditional manual division methods. The boundary detection algorithm fits edge anomaly points, keeping the calculation error of dangerous area boundary coordinates within a preset range and ensuring the consistency between the boundary outline and the actual dangerous area. This effectively reduces the risk of decision-making errors caused by ambiguous identification of dangerous areas during rescue operations.

[0104] In one embodiment, the process of overlaying the real-time trajectory prediction path with the boundary coordinate set to obtain a risk intersection point set, generating a trajectory prediction path based on the intersection point set, and calculating the interception priority ranking to determine the rescue interception location sequence may include the following steps:

[0105] Step S501: Use a spatial analysis algorithm to overlay and analyze the real-time trajectory prediction path and the boundary coordinate set, identify spatial intersection points and count the spatial distribution density of the intersection points to obtain the risk point density distribution.

[0106] Step S502: Using a spatial conflict detection algorithm, regions with density values ​​higher than the risk density threshold in the risk point density distribution are selected, and adjacent high-risk density regions are merged into continuous regions to determine the conflict region set; the risk density threshold is preset according to the water rescue safety standard; the conflict region set includes the outline coordinates of each conflict region and the corresponding conflict level.

[0107] Step S503: Based on the correction rules, the set of conflict areas is corrected to generate a dynamic path correction scheme. The original real-time trajectory prediction path is adjusted according to the scheme to obtain the corrected trajectory prediction path. The correction rules include avoiding conflict areas, minimizing trajectory offset, and adapting to the direction of water flow in the water area.

[0108] Step S504: Based on the corrected trajectory prediction path, set the interception priority evaluation index, calculate the priority score of each candidate interception point by weighting, and determine the rescue interception location sequence by sorting the scores from high to low; the priority evaluation index includes the distance between the interception point and the trajectory prediction location, the adaptability of the water environment at the interception point and the accessibility of rescue equipment; the rescue interception location sequence includes the latitude and longitude coordinates of each interception point, the recommended interception time and the priority score.

[0109] Specifically, firstly, a spatial analysis algorithm is used to overlay and analyze the real-time trajectory prediction path (a line element composed of continuous latitude and longitude points) and the boundary coordinate set of potential danger areas (polygonal surface elements). By calculating the spatial topological relationship between the two, the intersection points are identified as spatial intersection points. The number of intersection points per unit area is counted to obtain the risk point density distribution. Then, a spatial conflict detection algorithm is used to filter out areas in the risk point density distribution whose density values ​​are higher than a preset risk density threshold (this threshold is determined according to the water rescue safety standard). A region merging algorithm is used to integrate adjacent high-risk density areas into a continuous region, forming a set of conflict areas that includes the outline coordinates of each conflict area and the corresponding conflict level (based on density value division). Based on the correction rules of "avoiding conflict areas, minimizing trajectory deviation, and adapting to the direction of water flow", path avoidance calculations are performed on the set of conflict areas to generate a dynamic path correction scheme. The original real-time trajectory prediction path is adjusted segment by segment according to the scheme to obtain the corrected trajectory prediction path. Finally, based on the corrected trajectory prediction path, interception priority evaluation indicators are set (including the distance between the interception point and the trajectory prediction location, the adaptability of the water environment at the interception point, and the accessibility of rescue equipment). By assigning preset weights to each indicator and performing weighted calculations, the priority score of each candidate interception point is obtained. The rescue interception position sequence is determined by sorting the scores from high to low. This sequence includes the latitude and longitude coordinates of each interception point, the recommended interception time, and the priority score.

[0110] This embodiment achieves quantitative characterization of the conflict between trajectory and dangerous area through spatial overlay analysis and density statistics, improving the accuracy of risk point density distribution compared to traditional spatial matching methods. The screening and merging of conflict areas improves the identification efficiency of high-risk areas, and the classification of conflict levels can provide differentiated basis for subsequent path correction. The application of dynamic path correction rules ensures that conflict areas are avoided while keeping the trajectory deviation within a minimum range, and the adaptation to the adjustment of water flow direction further reduces the difficulty of path execution. The interception priority evaluation system improves the effectiveness of recommended interception locations through multi-index weighted calculation, and the generation of rescue interception location sequences directly provides accurate spatiotemporal reference for rescue resource scheduling.

[0111] In one embodiment, the priority score of each candidate interception point is calculated using the following formula:

[0112]

[0113] in, This represents the priority score of the candidate interception point, with a value range of [value missing]. , This represents the straight-line distance between the candidate interception point and the corresponding predicted position at the time step in the corrected trajectory prediction path. 2R R represents the average radius of the Earth. Indicates the latitude of the candidate interception point. , Indicates the latitude coordinates of the candidate interception point. The latitude represents the predicted location of the trajectory. , This indicates the latitude coordinates of the predicted trajectory location. This represents the difference in latitude between two points. , This represents the difference in longitude between two points. , Indicates the longitude of the candidate interception point. , Indicates the longitude coordinates of the candidate interception point. The longitude of the predicted trajectory location. , Represents the longitude coordinates of the predicted trajectory location. This indicates the maximum allowable distance between the preset interception point and the predicted trajectory location. This represents the environmental adaptability coefficient of the water body at the interception point, calculated from the water flow velocity at the interception point, and its value ranges from [value missing]. , This represents the accessibility score for rescue equipment, with a value range of [value missing]. , , , These represent the weighting coefficients for the distance between the interception point and the predicted trajectory location, the adaptability of the water environment at the interception point, and the accessibility of rescue equipment, respectively.

[0114] Preferably, The maximum permissible distance between the preset interception point and the predicted trajectory location is set based on the effective operating radius of the rescue equipment (such as the remote control range of a rescue robot or the maneuverability of an inflatable boat), typically between 50 and 200 meters; when At that time, the candidate interception point was directly determined to be outside the effective rescue range. (No priority).

[0115] The water environment adaptability coefficient at the interception point is determined by the water flow velocity at the interception point. Calculations show that if This represents the preset safety water flow velocity threshold for rescue, typically taken as 1.5 meters per second. ;like This indicates the preset rescue warning water flow velocity threshold, typically taken as 2.5 meters per second. ,like ,but (The environment is not suitable, so this interception point is excluded).

[0116] Accessibility rating for rescue equipment is based on the suitability assessment of the type of rescue equipment (such as rescue robots, inflatable boats) and the water terrain at the interception point (such as the presence of shallows or reefs): if the equipment can reach the target directly and there are no terrain obstacles, If a detour is required but the destination is still attainable, If there are insurmountable terrain obstacles (such as large reefs or shallow waters that pose a risk of grounding). (The device cannot reach this point, so this interception point is excluded).

[0117] This embodiment employs a spherical distance formula based on latitude and longitude (considering the Earth's curvature) to accurately quantify the spatial deviation between candidate interception points and predicted trajectory locations. It also standardizes distance indicators by combining this with a maximum permissible distance parameter. Furthermore, it integrates aquatic environment adaptability coefficients (correlated with water flow velocity) and rescue equipment accessibility scores. By dynamically balancing the influence of distance, environment, and equipment through weighted coefficients, the priority score objectively reflects the comprehensive adaptability of the interception point within the 0-100 range. Compared to traditional assessment methods based solely on planar distance, this method reduces distance calculation errors in large-scale water areas. The weighted fusion of multi-dimensional indicators improves the actual rescue success rate of high-priority interception points, providing a quantitative decision-making basis for the precise allocation of rescue resources and effectively avoiding interception location selection biases caused by single-indicator evaluations.

[0118] In one embodiment, if the rescue interception location sequence includes complex environmental variables, the particle filter algorithm is iteratively executed to adjust the trajectory deviation vector to obtain a safe trajectory model, which may include the following steps:

[0119] Step S601: If the rescue interception location sequence includes environmental complexity variables, then extract the environmental complexity variables from the rescue interception location sequence; the environmental complexity variables include terrain complexity and dynamic obstacle density.

[0120] Step S602: The particle filter algorithm is used to use the complex environmental variables as constraints for particle state transitions to calculate the trajectory deviation vector; the trajectory deviation vector is the position offset between the corrected trajectory prediction path and the original prediction path.

[0121] Step S603: Adjust the trajectory deviation vector based on the preset safety offset threshold to generate an initial safe trajectory model; the safety offset threshold is set according to the mobility of the rescue equipment; the initial safe trajectory model is a predicted trajectory of the person who has fallen into the water to avoid the influence of complex environmental variables.

[0122] Preferably, a safe deviation threshold is pre-set based on the mobility of the rescue equipment. This threshold represents the maximum deviation range within which the equipment can safely perform path adjustments in complex environments. The trajectory deviation vector is limited and adjusted based on this threshold. If the deviation exceeds the threshold, normalization is performed to ensure that the path adjustment remains within the equipment's capabilities. Finally, an initial safe trajectory model is generated. This model actively avoids complex terrain areas and dynamic obstacles, forming a preliminary safe drift path for the person who has fallen into the water.

[0123] Step S604: Extract dynamic obstacle influence factors based on the initial safe trajectory model and determine trajectory optimization parameters; dynamic obstacle influence factors include the comprehensive quantitative value of obstacle movement speed, volume and trajectory intersection probability; trajectory optimization parameters include the state update coefficient and weight adjustment threshold in the particle filter algorithm.

[0124] Step S605: Input the trajectory optimization parameters into the particle filter algorithm to update the particle weight distribution in the algorithm, and recalculate to obtain the optimized trajectory deviation vector.

[0125] Step S606: Based on the preset trajectory stability threshold, the obtained optimized trajectory deviation vector is judged. If the judgment result is that the trajectory stability threshold is met, a safe trajectory model is generated based on the vector. The trajectory stability threshold is the maximum allowable fluctuation value of the trajectory deviation between adjacent time steps. The safe trajectory model includes the safe drift path of the person who fell into the water with the timestamp.

[0126] Furthermore, a trajectory stability threshold is set, which defines the maximum allowable fluctuation in trajectory deviation between adjacent time steps, used to measure the smoothness and reliability of the trajectory. The optimized trajectory deviation vector is compared with this threshold. If the deviation fluctuations of multiple consecutive time steps are within the threshold range, it indicates that the trajectory has stabilized. At this point, a final safe trajectory model is generated based on this vector. This model contains accurate timestamp information, fully recording the safe drift path of the person who fell into the water in a complex environment, providing a reliable path planning reference for rescue operations.

[0127] Specifically, if there are complex environmental variables (including terrain complexity and dynamic obstacle density) in the rescue interception location sequence, these variables are first extracted from the sequence. The complex environmental variables are then used as constraints for particle state transitions and input into the particle filtering algorithm to calculate the positional offset between the corrected trajectory prediction path and the original path, resulting in a trajectory deviation vector. Based on a safety offset threshold set according to the mobility of the rescue equipment, the trajectory deviation vector is adjusted to generate an initial safe trajectory model that avoids the influence of complex environmental variables. From the initial safe trajectory model, a comprehensive quantitative value (i.e., dynamic obstacle influence factor) is extracted, representing the obstacle's moving speed, volume, and trajectory intersection probability. Based on this, the state update coefficient and weight adjustment threshold (i.e., trajectory optimization parameters) of the particle filtering algorithm are determined. The trajectory optimization parameters are input into the particle filtering algorithm to update the particle weight distribution, and the optimized trajectory deviation vector is recalculated. Finally, the optimized vector is judged based on a preset maximum allowable fluctuation value for trajectory deviation between adjacent time steps (i.e., trajectory stability threshold). If the threshold is met, a safe drift path for the person in the water containing a timestamp (i.e., a safe trajectory model) is generated.

[0128] This embodiment improves the accuracy of trajectory offset calculation by using environmental complexity variables as particle filtering constraints in water areas with a terrain complexity index ≥ 0.7 or a dynamic obstacle density ≥ 0.3 per 100㎡. A safety offset threshold set based on equipment maneuverability ensures the initial safe trajectory model meets the safety distance requirements of dangerous areas. An iterative calculation mechanism for dynamic obstacle influence factors and trajectory optimization parameters enables adaptive updating of particle weight distribution, improving the stability of the trajectory deviation vector and controlling the fluctuation amplitude between adjacent time steps to within 15 meters. This effectively solves the problem of low reliability in trajectory prediction under complex environments, and the final generated safe trajectory model has a lower positioning error in complex water areas compared to traditional methods.

[0129] In one embodiment, such as Figure 2 As shown, this application also provides an intelligent control and rescue system for people who have fallen into the water, the system may include:

[0130] The dataset construction module 701 is used to acquire water flow velocity data, terrain feature data and meteorological parameters in the aquatic environment to construct a multi-source dynamic dataset.

[0131] The danger zone determination module 702 is used to process data uncertainty based on a multi-source dynamic dataset using a particle filter algorithm, determine the probability distribution of the initial position of the person who fell into the water, and obtain the real-time trajectory prediction path based on the position distribution; it is also used to classify abnormal patterns in the multi-source dynamic dataset using a support vector machine, and determine the set of boundary coordinates of potential danger zones based on the classification results.

[0132] The rescue location determination module 703 is used to overlay and analyze the real-time trajectory prediction path with the boundary coordinate set to obtain the risk intersection point set, generate the trajectory prediction path based on the intersection point set, and calculate the interception priority ranking to determine the rescue interception location sequence.

[0133] The trajectory model generation module 704 is used to iteratively execute a particle filter algorithm to adjust the trajectory deviation vector and obtain a safe trajectory model if the rescue interception location sequence includes complex environmental variables; the complex environmental variables include terrain complexity and dynamic obstacle density.

[0134] The aforementioned intelligent control and rescue system for people who have fallen into the water comprises two modules. The dataset construction module acquires water flow velocity data, terrain feature data, and meteorological parameters from the aquatic environment through a sensor network. After standardization, a multi-source dynamic dataset is constructed to provide basic data support for subsequent analysis. The danger zone determination module, based on this dataset, uses a particle filter algorithm to handle data uncertainty, determine the probability distribution of the initial position of the person in the water, and derive a real-time trajectory prediction path. It also uses a support vector machine to classify abnormal patterns in the dataset and determines the set of boundary coordinates for potential danger zones based on the classification results. The rescue location determination module overlays the real-time trajectory prediction path with the set of boundary coordinates of the danger zones, extracts a risk intersection point set, generates an optimized trajectory prediction path, and determines the rescue interception location sequence by calculating interception priority ranking. When the trajectory model generation module detects that the rescue interception location sequence contains complex environmental variables such as terrain complexity and dynamic obstacle density, it iteratively executes a particle filter algorithm to adjust the trajectory deviation vector, ultimately generating a safe trajectory model.

[0135] In this embodiment, the modules form a complete technical link through the orderly transmission of data flow. The dataset construction module provides a multi-source dynamic dataset for the danger zone determination module. The real-time trajectory prediction path and the set of boundary coordinates of potential danger zones output by the danger zone determination module serve as input data for the rescue location determination module. The rescue interception location sequence generated by the rescue location determination module is passed to the trajectory model generation module. If the interception location sequence contains complex environmental variables, the trajectory model generation module iteratively optimizes and outputs a safe trajectory model based on this. In addition, the dataset construction module achieves unified integration of multi-source environmental data, providing a high-quality data foundation for subsequent analysis. The danger zone determination module integrates particle filtering and support vector machine algorithms to simultaneously improve the accuracy of trajectory prediction for those falling into the water and the efficiency of danger zone identification, resulting in a comprehensive performance improvement compared to traditional single-module processing methods. The rescue location determination module improves the effectiveness of rescue interception locations through spatial overlay analysis and priority ranking. The trajectory model generation module's iterative optimization mechanism for complex environments ensures the stability of the safe trajectory model in high-interference scenarios, with trajectory deviation fluctuations controlled within 15 meters.

[0136] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0137] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps of the intelligent control rescue method, system, device, and medium for a person who has fallen into the water as described above.

[0138] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0139] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0140] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A method for intelligent control of rescue of a falling person, characterized in that, The method includes: Acquire water flow velocity data, terrain feature data, and meteorological parameters in the aquatic environment to construct a multi-source dynamic dataset; Based on the multi-source dynamic dataset, a particle filter algorithm is used to process data uncertainty, determine the probability distribution of the initial position of the person who fell into the water, and obtain the real-time trajectory prediction path based on the position distribution. Support vector machine is used to classify the abnormal patterns in the multi-source dynamic dataset, and the set of boundary coordinates of potential dangerous areas is determined based on the classification results; The real-time trajectory prediction path is overlaid with the boundary coordinate set to obtain a risk intersection point set. Based on the intersection point set, a trajectory prediction path is generated and the interception priority ranking is calculated to determine the rescue interception location sequence. If the rescue interception location sequence includes environmental complexity variables, the particle filtering algorithm is iteratively executed to adjust the trajectory deviation vector and obtain a safe trajectory model; the environmental complexity variables include terrain complexity and dynamic obstacle density.

2. The method of claim 1, wherein, The process of using a particle filter algorithm to process data uncertainty based on the multi-source dynamic dataset, determining the probability distribution of the initial position of the person who fell into the water, and obtaining a real-time trajectory prediction path based on the position distribution includes: The initial location data of the person who fell into the water is extracted from the multi-source dynamic dataset; the initial location data includes the latitude and longitude information of the person who fell into the water, the data acquisition timestamp, and the noise parameters of the corresponding sensor; The particle filter algorithm is used to process the uncertainty introduced by sensor noise in the initial position data to obtain the probability distribution of the initial position of the person who fell into the water. The drift velocity vector of the person who fell into the water is calculated based on the probability distribution of the initial position of the person who fell into the water and combined with the multi-source dynamic dataset, and the predicted trajectory of the person who fell into the water is determined based on the drift velocity vector. Dynamic weights are generated based on the deviation characteristics of the predicted trajectory, and the predicted trajectory is adjusted using the dynamic weights to obtain an optimized trajectory path distribution. If the confidence level of the optimized trajectory path distribution is lower than the preset confidence threshold, supplementary environmental data is extracted from the multi-source dynamic dataset; the supplementary environmental data includes ocean current velocity data and wind speed data above the water area; the preset confidence threshold is set according to the rescue accuracy requirements. The supplementary environmental data is input into the particle filter algorithm to update the state transition parameters, the corrected probability distribution of the drowning person's location is recalculated, and the updated real-time trajectory prediction path of the drowning person is determined.

3. The method of claim 2, wherein, The step of using a particle filtering algorithm to process the uncertainty introduced by sensor noise in the initial position data to obtain the probability distribution of the initial position of the person who fell into the water includes: based on an initial measurement of latitude and longitude of the fallen person in the multi-source dynamic data set and sensor noise covariance , generating an initial particle set from the initial particle ; wherein, represents the state vector of the th particle, and satisfies , represents a normal distribution; Based on the obtained initial particle set Combining the initial water area environment micro-perturbation factor , the modified sensor noise covariance matrix fused with the environment micro-perturbation is calculated , and for each particle The corresponding weight is calculated according to the following formula The particle weight set is obtained : exp wherein denotes the determinant of denotes the inverse matrix of Based on the particle weight set The weights of all particles are normalized to obtain a normalized weight set. ; Combined with the initial particle set With the normalized weight set Discrete probability distribution constituting the initial position of the person who fell into the water : in, This represents the Dirac function.

4. The method according to claim 1, characterized in that, The step of classifying abnormal patterns in the multi-source dynamic dataset using a support vector machine and determining the set of boundary coordinates of potentially hazardous areas based on the classification results includes: Extract anomaly pattern data containing time series features and spatial features from the multi-source dynamic dataset; the anomaly pattern data includes the timestamp sequence of the anomaly occurrence, the latitude and longitude coordinates of the corresponding spatial location, and the values ​​of the anomaly environmental parameters; The abnormal pattern data is input into a support vector machine model to obtain abnormal pattern category labels; the abnormal pattern category labels include eddy current region, rapid current region and underwater obstacle influence region. Based on the anomaly pattern category labels, spatial clustering analysis is performed on the anomaly pattern data corresponding to the same category labels to determine the spatial distribution characteristics of potential hazardous areas for each category; the spatial distribution characteristics include the core coordinates of the hazardous area, the density of anomaly points, and the spatial spread trend; Based on the spatial distribution characteristics, a boundary detection algorithm is used to fit the edge anomalies of the potential danger area, and the boundary coordinate set of the potential danger area is calculated; the boundary coordinate set is a sequence of latitude and longitude coordinates that constitute the polygonal outline of the danger area.

5. The method according to claim 1, characterized in that, The step of overlaying the real-time trajectory prediction path with the boundary coordinate set to obtain a risk intersection point set, generating a trajectory prediction path based on the intersection point set, and calculating the interception priority ranking to determine the rescue interception location sequence includes: A spatial analysis algorithm is used to overlay the real-time trajectory prediction path with the boundary coordinate set to identify spatial intersection points and statistically analyze the spatial distribution density of the intersection points, thereby obtaining the risk point density distribution. The spatial conflict detection algorithm filters out areas in the risk point density distribution whose density values ​​are higher than the risk density threshold, and merges adjacent high-risk density areas into a continuous area to determine the conflict area set; the risk density threshold is preset according to the water rescue safety standard; the conflict area set includes the outline coordinates of each conflict area and the corresponding conflict level. Based on the correction rules, a dynamic path correction scheme is generated by correcting the set of conflict areas. The original real-time trajectory prediction path is adjusted according to the scheme to obtain the corrected trajectory prediction path. The correction rules include avoiding conflict areas, minimizing trajectory offset, and adapting to the direction of water flow in the water area. Based on the corrected trajectory prediction path, an interception priority evaluation index is set. The priority score of each candidate interception point is calculated by weighting, and the rescue interception location sequence is determined by sorting the scores from high to low. The priority evaluation index includes the distance between the interception point and the trajectory prediction location, the adaptability of the water environment at the interception point, and the accessibility of rescue equipment. The rescue interception location sequence includes the latitude and longitude coordinates of each interception point, the recommended interception time, and the priority score.

6. The method according to claim 5, characterized in that, The priority score of each candidate interception point is calculated using the following formula: in, This represents the priority score of the candidate interception point, with a value range of [value missing]. , This represents the straight-line distance between the candidate interception point and the corresponding predicted position at the time step in the corrected trajectory prediction path. 2R R represents the Earth's average radius. Indicates the latitude of the candidate interception point. , Indicates the latitude coordinates of the candidate interception point. The latitude represents the predicted location of the trajectory. , This indicates the latitude coordinates of the predicted trajectory location. This represents the difference in latitude between two points. , This represents the difference in longitude between two points. , Indicates the longitude of the candidate interception point. , Indicates the longitude coordinates of the candidate interception point. The longitude of the predicted trajectory location. , Represents the longitude coordinates of the predicted trajectory location. This indicates the maximum allowable distance between the preset interception point and the predicted trajectory location. This represents the environmental adaptability coefficient of the water body at the interception point, calculated from the water flow velocity at the interception point, and its value ranges from [value missing]. , This represents the accessibility score for rescue equipment, with a value range of [value missing]. , , , These represent the weighting coefficients for the distance between the interception point and the predicted trajectory location, the adaptability of the water environment at the interception point, and the accessibility of rescue equipment, respectively.

7. The method according to claim 1, characterized in that, If the rescue interception location sequence includes complex environmental variables, then the particle filter algorithm is iteratively executed to adjust the trajectory deviation vector to obtain a safe trajectory model, including: If the rescue interception location sequence includes environmental complexity variables, then the environmental complexity variables are extracted from the rescue interception location sequence; the environmental complexity variables include terrain complexity and dynamic obstacle density. The particle filtering algorithm is used to treat the complex environmental variables as constraints for particle state transitions, and the trajectory deviation vector is calculated. The trajectory deviation vector is the position offset between the corrected trajectory prediction path and the original prediction path. The trajectory deviation vector is adjusted based on a preset safety offset threshold to generate an initial safe trajectory model; the safety offset threshold is set according to the mobility of the rescue equipment; the initial safe trajectory model is a predicted trajectory of the person who has fallen into the water to avoid the influence of the complex environmental variables. Based on the initial safe trajectory model, dynamic obstacle influence factors are extracted to determine trajectory optimization parameters; the dynamic obstacle influence factors include a comprehensive quantitative value of obstacle movement speed, volume, and trajectory intersection probability; the trajectory optimization parameters include the state update coefficient and weight adjustment threshold in the particle filter algorithm. The trajectory optimization parameters are input into the particle filter algorithm to update the particle weight distribution in the algorithm, and the optimized trajectory deviation vector is recalculated. The optimized trajectory deviation vector is judged based on a preset trajectory stability threshold. If the judgment result is that the trajectory stability threshold is met, a safe trajectory model is generated based on the vector. The trajectory stability threshold is the maximum allowable fluctuation value of the trajectory deviation between adjacent time steps. The safe trajectory model includes the safe drift path of the person who fell into the water with the timestamp.

8. A smart control and rescue system for people who have fallen into the water, characterized in that, The system includes: The dataset construction module is used to acquire water flow velocity data, terrain feature data and meteorological parameters in the aquatic environment to construct a multi-source dynamic dataset; The danger zone determination module is used to process data uncertainty based on the multi-source dynamic dataset using a particle filter algorithm, determine the probability distribution of the initial position of the person who fell into the water, and obtain a real-time trajectory prediction path based on the position distribution; it is also used to classify abnormal patterns in the multi-source dynamic dataset using a support vector machine, and determine the set of boundary coordinates of potential danger zones based on the classification results. The rescue location determination module is used to overlay and analyze the real-time trajectory prediction path with the boundary coordinate set to obtain a risk intersection point set, generate a trajectory prediction path based on the intersection point set, and calculate the interception priority ranking to determine the rescue interception location sequence. The trajectory model generation module is used to iteratively execute the particle filtering algorithm to adjust the trajectory deviation vector and obtain a safe trajectory model if the rescue interception location sequence includes environmental complex variables; the environmental complex variables include terrain complexity and dynamic obstacle density.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images