Personnel detection method and system applied to fire scene

Abstract

The present invention provides a personnel detection method and system applied to a fire scene. First, a multi-source signal data set of the fire scene is obtained. This multi-source signal data set includes communication signal units with time tags collected at different positions. Then, signal feature extraction processing is performed on the multi-source signal data set to obtain the spatial distribution characteristics and time series change characteristics of the communication signal units. Then, a pre-trained personnel positioning model is called to perform joint positioning analysis on the spatial distribution characteristics and time series change characteristics to generate a potential personnel position candidate set corresponding to the communication signal unit. The target position information of the personnel to be rescued in the fire scene is determined according to the potential personnel position candidate set. Finally, a rescue guidance instruction including position coordinates is generated based on the target position information, and the rescue guidance instruction is sent to a rescue terminal to trigger a personnel search and rescue operation, thereby utilizing communication signals for indirect positioning, overcoming the limitations of traditional methods, and improving rescue efficiency.

Description

Technical Field

[0001] The present invention relates to the field of emergency rescue technology, and in particular to a personnel detection method and system applied to a fire scene. Background Art

[0002] In the field of emergency rescue, search and rescue at fire scenes has always been a particularly challenging task. Fires often present extremely harsh conditions, characterized by high temperatures, thick smoke, and toxic gases. These conditions can also be accompanied by dangerous conditions such as toxic gas explosions and building collapses. These factors not only pose a serious threat to the lives of those trapped but also create significant obstacles for firefighters in their search and rescue efforts.

[0003] In traditional disaster relief, life detection technology primarily focuses on detecting biosignals, such as weak signals from human breathing and heartbeats, or detecting bio-heat sources to locate survivors. However, these traditional methods have significant limitations in real-world fire scenarios. Smoke absorbs and scatters infrared signals, making it difficult for infrared thermal imaging-based detection technologies to accurately identify occupants. Obstacles such as walls reflect and attenuate radar signals, affecting detection accuracy. Extremely high temperatures can interfere with the ability of life detectors to accurately capture biosignals, resulting in inaccurate detection results or even the inability to detect trapped individuals.

[0004] At the same time, with the widespread adoption of mobile devices, the per capita mobile phone ownership rate in modern society is extremely high. During disasters such as fires, most trapped individuals typically carry their phones, which may be in standby mode or have a weak signal. Even if a trapped individual is unconscious and their body temperature is low, making them difficult to detect by infrared or radar, as long as their phone is powered, it will periodically scan the network or transmit wireless signals, providing a stable and observable signal source for locating individuals. Therefore, traditional biometric signal detection methods are no longer sufficient for search and rescue operations in the complex environments of fire scenes, and a new, more effective human location technology is urgently needed. Summary of the Invention

[0005] In view of the above-mentioned problems, in combination with the first aspect of the present invention, an embodiment of the present invention provides a method for detecting people at a fire scene, the method comprising:

[0006] Acquire a multi-source signal data set at a fire scene, wherein the multi-source signal data set includes communication signal units with time stamps collected at different locations;

[0007] Performing signal feature extraction processing on the multi-source signal data set to obtain spatial distribution features and temporal variation features of the communication signal unit;

[0008] Calling a pre-trained personnel positioning model to perform joint positioning analysis on the spatial distribution characteristics and the temporal variation characteristics to generate a set of potential personnel position candidates corresponding to the communication signal unit;

[0009] Determine target location information of persons to be rescued at the fire scene based on the potential personnel location candidate set;

[0010] A rescue guidance instruction including location coordinates is generated based on the target location information, and the rescue guidance instruction is sent to a rescue terminal to trigger a personnel search and rescue operation.

[0011] On the other hand, an embodiment of the present invention also provides a personnel detection system for use at a fire scene, comprising a processor and a machine-readable storage medium, wherein the machine-readable storage medium is connected to the processor, the machine-readable storage medium is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the machine-readable storage medium to implement the above method.

[0012] Based on the above, the embodiments of the present invention are fundamentally different from traditional biosignal detection technologies. Their innovation lies in the use of single or multiple mobile robots (such as robot dogs) to measure mobile phone signals and estimate the approximate location of trapped individuals. By acquiring a multi-source signal data set containing time-stamped communication signal units at different locations at the fire scene, comprehensive and accurate information on mobile phone signals in both spatial and temporal dimensions is collected.

[0013] Signal feature extraction and processing of multi-source signal data sets reveal the spatial distribution and temporal variation characteristics of communication signal units. This allows for in-depth exploration of the propagation patterns and changing trends of mobile phone signals within the complex environment of a fire scene, effectively overcoming the signal inaccuracy inherent in traditional methods due to environmental interference. A pre-trained personnel location model is used to perform a joint location analysis of the spatial distribution and temporal variation characteristics. This model leverages its powerful data analysis and processing capabilities, comprehensively considering multiple factors, and generates a set of potential personnel location candidates corresponding to communication signal units, significantly improving the accuracy and reliability of personnel location detection.

[0014] The target location information of the person to be rescued is determined based on the set of potential personnel position candidates, and a rescue guidance instruction containing the location coordinates is generated based on the information and sent to the rescue terminal to trigger the personnel search and rescue operation, realizing the full automation and intelligence from signal acquisition to personnel search and rescue. The technical solution of this embodiment is not affected by environmental factors such as smoke, walls, and high temperature. It can quickly and accurately locate trapped people at complex and changeable fire scenes, significantly improving the rescue efficiency and gaining precious time to ensure the safety of the lives of trapped people. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a schematic diagram of the execution flow of the personnel detection method applied to a fire scene provided by an embodiment of the present invention.

[0016] Figure 2 FIG. 4 is a schematic diagram of exemplary hardware and software components of a personnel detection system applied to a fire scene provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0017] The present invention will be described in detail below with reference to the accompanying drawings. Figure 1 The figure is a flow chart of a personnel detection method applied to a fire scene provided by an embodiment of the present invention. The personnel detection method applied to a fire scene is introduced in detail below.

[0018] Step S110: Acquire a multi-source signal data set at the fire scene, where the multi-source signal data set includes communication signal units with time stamps collected at different locations.

[0019] For example, a fire scene at a large commercial complex presents numerous complex factors. High temperatures, dense smoke, and toxic gases pose not only dangers to rescue workers but also interfere with signal transmission and acquisition. A multi-source signal dataset contains time-stamped communication signal units collected from different locations. These units may originate from mobile devices such as mobile phones of trapped individuals. By analyzing these units, the location of the trapped individuals can be inferred.

[0020] Step S111: The mobile detection equipment deployed at the fire scene performs signal scanning operations at different spatial points, and records the physical coordinate information of each scanning point.

[0021] In detail, at the fire scene of a large commercial complex, multiple mobile detection equipment can be deployed, and a device similar to a robot dog can be used here. The robot dog can be specially designed to adapt to the harsh environment of the fire scene.

[0022] The robot dog is equipped with a core control unit and several important sensors. For example, the core control unit is the robot dog's control center, responsible for coordinating the work of various sensors and performing preliminary processing and analysis of collected data. It can rationally allocate sensor work based on different task requirements, ensuring that the robot dog can efficiently complete signal scanning and positioning tasks.

[0023] LiDAR is one of the robot dog's key sensors. It accurately calculates the distance and position of surrounding objects by emitting a laser beam and measuring the time it takes for the laser to reflect back. At a fire scene, LiDAR can construct a three-dimensional map of the surrounding environment in real time, helping the robot dog understand its own position and the distribution of surrounding obstacles, enabling autonomous navigation and obstacle avoidance.

[0024] The camera can provide visual information on the scene, capturing images and videos of the fire scene. This is very helpful for determining the intensity of the fire, smoke levels, and the presence of trapped people. By analyzing the camera images, the robot dog's movement path and key scanning areas can be further determined.

[0025] The inertial measurement unit (IMU) measures the robot's posture and motion. It monitors its acceleration, angular velocity, and other information in real time, accurately calculating its position and orientation. As the robot moves, the IMU helps it maintain a stable posture and ensures accurate signal scanning.

[0026] The robot dog can operate in two modes: autonomous movement and remote control. In autonomous movement mode, the robot dog can automatically move between different locations in the commercial complex according to a pre-set path or a real-time generated map. During movement, it can continuously update its location information and adjust its movement strategy according to changes in the surrounding environment. In remote control mode, the operator can control the robot dog in real time through a remote control device, directing it to a designated location for signal scanning.

[0027] The robot's waterproof, fireproof, and dustproof design allows it to continue operating in the harsh environment of a fire scene. While high temperatures and dense smoke can damage standard equipment, the robot's special housing and protective measures effectively protect its internal electronic components, ensuring normal operation.

[0028] The robot dog will follow a pre-planned route or adjust its path in real time based on actual conditions, moving through different spatial locations within the commercial complex. These locations include public areas on each floor, such as corridors and lobbies; the interiors of various stores, where trapped people may seek shelter; and stairwells and elevator vestibules, where trapped people may escape or await rescue.

[0029] When the robot dog reaches each scanning point, the LiDAR and IMU work together to record the physical coordinates of that scanning point. The LiDAR creates a three-dimensional map that provides precise location information of the surrounding environment, while the IMU provides information on the robot dog's own posture and position changes. By fusing and processing these two types of information, the physical coordinates of the robot dog's scanning point can be accurately determined. This physical coordinate information is based on a pre-established coordinate system within the commercial complex. This coordinate system uses a fixed corner of the commercial complex as its origin and represents positions in three-dimensional space using the X, Y, and Z axes.

[0030] Step S112: During the signal scanning process at each scanning point, the communication signal in the environment is sampled and processed according to a preset time interval to generate a signal sampling segment with a continuous time series.

[0031] At each scanning point, the robot dog activates its wireless signal detection and positioning module to scan for signals. This module is equipped with a multi-standard antenna and RF front-end. The multi-standard antenna can receive communication signals in multiple frequency bands, including common 2G, 3G, 4G, and 5G cellular network signals, as well as short-range wireless communication signals such as Wi-Fi and Bluetooth.

[0032] The RF front end performs preliminary processing on the signal received by the antenna. It can amplify the signal and enhance its strength for more accurate subsequent analysis and processing. It also filters the signal to remove noise and interference, improving signal quality.

[0033] The preset time interval is determined based on the characteristics of the signal and the requirements for subsequent processing. If the interval is set too long, some important signal changes may be missed; if the interval is set too short, a large amount of data will be generated, increasing the burden of subsequent processing. During each interval, the robot dog can sample the communication signal in the surrounding environment.

[0034] Each sampling operation generates a signal sample containing information such as the specific frequency band and intensity. Over time, a series of continuous time-series signal sampling segments are generated at the same scanning point. These signal sampling segments record the communication signal conditions at that scanning point at different times.

[0035] Step S113: adding corresponding scanning point physical coordinate information and sampling timestamp information to each signal sampling segment to form a basic signal unit with a temporal and spatial correlation relationship.

[0036] Each generated signal sampling segment needs to be associated with the corresponding scanning point physical coordinate information and sampling timestamp information. The scanning point physical coordinate information has been accurately recorded in step S111, while the sampling timestamp information is accurately recorded by the robot dog's clock system when the signal is sampled.

[0037] Through a specific data processing program, the physical coordinates of the scanning point and the sampling timestamp are added to the data structure of each signal sampling segment. As a result, each signal sampling segment is associated with a specific spatial location and time point, forming a basic signal unit with a temporal and spatial correlation.

[0038] For example, if a signal sampling segment is collected in a store on a certain floor of a commercial complex, and the collection time is a specific moment after a fire occurs, then the signal sampling segment will be marked with the precise physical coordinates of the store and the corresponding timestamp. These basic signal units combine time and space information with communication signals.

[0039] Step S114: performing communication protocol parsing processing on the basic signal units to screen out valid signal units that meet preset communication standards. The preset communication standards include cellular network communication protocols and short-range wireless communication protocols.

[0040] The signal recognition and processing unit in the robot dog's wireless signal detection and positioning module performs communication protocol analysis on all basic signal units. At the commercial complex fire scene, various types of communication signals were present in the environment, including interference signals or signals that did not meet the requirements.

[0041] The signal identification and processing unit performs a detailed analysis of the protocol format and signature codes of the basic signal units. For cellular network communication protocols, it can check whether the signal's frame structure and signaling format conform to the specifications of 2G, 3G, 4G, and 5G networks. Different cellular network protocols have different signal characteristics and communication rules. By identifying these characteristics, it can determine whether the signal originates from a legitimate cellular network device.

[0042] For short-range wireless communication protocols such as Wi-Fi and Bluetooth, you can check whether the signal's frequency band, modulation method, and protocol header information comply with the corresponding standards. Wi-Fi signals have specific frequency bands and protocol formats, and Bluetooth signals also have unique characteristics. By analyzing these characteristics, you can screen out short-range wireless communication signals that meet the standards.

[0043] Only signals that meet pre-set communication standards are considered valid signal units. These signals are more likely to be from the trapped person's mobile phone or other portable electronic device. After screening, signals that do not meet the standards are eliminated, resulting in a series of valid signal units.

[0044] Step S115: reorganize the valid signal units in sequence according to the sampling timestamp order to generate a multi-source signal data set containing multiple spatiotemporally correlated signal sub-units, each signal sub-unit in the multi-source signal data set carries the corresponding scanning point physical coordinate information and sampling timestamp information.

[0045] After filtering out the valid signal units, they need to be reordered and integrated. The data processing program will sort the valid signal units in ascending order based on their sampling timestamp information.

[0046] After the sorting is completed, these ordered valid signal units are combined to form a multi-source signal data set containing multiple spatiotemporally correlated signal sub-units. In this multi-source signal data set, each signal sub-unit still retains its corresponding scanning point physical coordinate information and sampling timestamp information.

[0047] This sequence reorganization allows for temporal ordering and spatial correlation of multi-source signal data sets, facilitating subsequent holistic signal analysis and processing. For example, it allows for more effective observation of signal trends across time and space.

[0048] Step S120: performing signal feature extraction processing on the multi-source signal data set to obtain spatial distribution features and temporal variation features of communication signal units.

[0049] Multi-source signal data sets contain a wealth of information, but to more accurately determine the location of trapped individuals, feature extraction is necessary. Spatial distribution features reflect the intensity distribution of signals at different spatial locations, while temporal variation features reveal how signals change over time. Analyzing these two types of features provides a more comprehensive understanding of the characteristics of communication signals.

[0050] Step S121: performing signal strength value extraction processing on each signal sub-unit in the multi-source signal data set to obtain a signal strength parameter of each signal sub-unit.

[0051] In a multi-source signal data set, each signal subunit contains specific communication signal information. Signal strength is an important parameter that can reflect factors such as the distance between the signal source and the scanning point, the impact of obstacles, etc.

[0052] The robot dog's signal recognition and processing unit will extract and process the signal strength value of each signal sub-unit. It can calculate the signal strength parameters by measuring and analyzing the signal's voltage, power and other physical quantities.

[0053] At a fire scene, signal propagation is affected by a variety of factors. High temperatures, dense smoke, and building structures can all cause signal attenuation and scattering. Therefore, accurately extracting signal strength parameters is crucial for subsequent analysis. By analyzing signal strength parameters, we can initially determine the approximate location and distance of the signal source.

[0054] Step S122: Based on the physical coordinate information of the scanning points corresponding to the signal subunits and the signal strength parameters, a corresponding relationship matrix between signal strength and spatial position is constructed as a spatial distribution feature. The spatial distribution feature includes a mapping relationship between different scanning points and corresponding signal strengths.

[0055] Step S1221: extracting the physical coordinate information of all scanning points in the signal subunit to form a spatial sampling point set containing multiple coordinate points.

[0056] The physical coordinate information of the scanning points corresponding to each signal subunit is extracted from the multi-source signal data set. These physical coordinate information has been accurately recorded in step S111 and represents the various locations where the robot dog performs signal scanning in the commercial complex.

[0057] The physical coordinate information of all these scanning points is aggregated to form a spatial sampling point set containing multiple coordinate points. This spatial sampling point set covers different floors and areas of the commercial complex, reflecting the spatial range of the signal scan.

[0058] Step S1222: Associating each coordinate point in the spatial sampling point set with the corresponding signal strength parameter one by one to generate a target association data pair.

[0059] For each coordinate point in the set of spatial sampling points, it is associated with the signal strength parameter of the corresponding signal subunit. Through the data processing program, the coordinate point and the signal strength parameter are combined into a data pair, which represents the signal strength at the coordinate point.

[0060] For example, a coordinate point corresponds to a specific location on a certain floor of a commercial complex. This coordinate point is associated with the signal strength parameters of the signal subunit collected at that location to form a target association data pair. In this way, spatial location and signal strength information are integrated.

[0061] Step S1223: Perform spatial grid division processing on the target associated data pair, and divide the fire scene area into multiple evenly distributed spatial grid units.

[0062] In order to analyze the spatial distribution of the signal in more detail, it is necessary to perform spatial grid division on the target correlation data. According to the layout and scale of the commercial complex, the entire fire scene area is divided into multiple spatial grid cells of equal size.

[0063] Each spatial grid cell can be considered a small area. This division simplifies complex spatial regions, facilitating signal strength statistics and analysis. The division process takes into account factors such as the commercial complex's architectural structure and channel distribution, ensuring that each spatial grid cell is representative and independent.

[0064] Step S1224: Calculate the average signal strength of all target association data pairs in each spatial grid unit to generate an average signal strength value for each spatial grid unit.

[0065] For each spatial grid cell, the signal strength parameters of all target-associated data pairs contained in it are collected. Then, the average signal strength within the spatial grid cell is calculated by summing these signal strength parameters and dividing by the number of data pairs.

[0066] The average signal strength value represents the overall signal strength level within the spatial grid cell. By calculating the average signal strength value, the influence of individual abnormal signals can be eliminated, and the signal distribution in the area can be more accurately reflected.

[0067] Step S1225: Matrixing the coordinate range information of the spatial grid cells and the corresponding average signal strength values ​​to generate a corresponding relationship matrix between signal strength and spatial position that reflects the law of signal strength changing with spatial position as a spatial distribution feature.

[0068] The coordinate range information and corresponding average signal strength values ​​of each spatial grid cell are sorted and arranged to form a correspondence matrix. The rows and columns of the correspondence matrix represent the coordinate range of the spatial grid cell, and the elements of the correspondence matrix are the corresponding average signal strength values.

[0069] The signal strength and spatial position correspondence matrix reflects the variation of signal strength with spatial position. By analyzing this correspondence matrix, the strength distribution of the signal at different spatial positions can be intuitively monitored.

[0070] Step S123: performing time dimension analysis processing on the continuous time series signal subunits in the multi-source signal data set, and calculating the signal strength change rate parameter between adjacent time stamp signal subunits.

[0071] Step S1231: extracting the timestamp information and corresponding signal strength parameters of the continuous time series signal subunits to generate time-intensity series data.

[0072] Extracting continuous time series signal subunits from a multi-source signal data set. These signal subunits are obtained by sampling at the same scanning point according to a preset time interval and have continuous time stamp information.

[0073] The timestamp information of each signal subunit and the corresponding signal strength parameter are combined to generate a time-intensity series data, which reflects the change of signal strength over time at the scanning point.

[0074] Step S1232: performing time interval normalization processing on the time-intensity series data to make the time intervals of adjacent time stamp signal subunits consistent.

[0075] Since there may be inconsistent time intervals in the actual sampling process, in order to facilitate subsequent analysis, it is necessary to perform time interval standardization on the time-intensity series data.

[0076] Through data processing algorithms, the timestamp information in the time-intensity series data is adjusted so that the time intervals of adjacent timestamp signal subunits are consistent, thereby ensuring the accuracy and comparability in the subsequent calculation of the signal intensity change rate.

[0077] Step S1233: traverse each pair of adjacent signal subunits in the time-intensity sequence data, and calculate the difference between the signal strength parameter of the subsequent timestamp signal subunit and the signal strength parameter of the previous timestamp signal subunit.

[0078] In the time-intensity series data that has been normalized for time intervals, each pair of adjacent signal subunits is traversed in sequence. For each pair of adjacent signal subunits, the difference between the signal intensity parameter of the latter timestamp signal subunit and the signal intensity parameter of the former timestamp signal subunit is calculated.

[0079] The difference reflects the change in signal strength between adjacent time points. By calculating the difference, we can initially understand the trend of signal strength change, whether it is increasing or decreasing.

[0080] Step S1234: Divide the difference by the time interval between adjacent timestamps to obtain a signal strength change rate parameter between adjacent timestamp signal subunits.

[0081] The calculated signal strength difference is divided by the time interval between adjacent timestamps to obtain a signal strength change rate parameter between adjacent timestamp signal subunits. The signal strength change rate parameter reflects the change in signal strength per unit time.

[0082] By calculating the signal strength change rate parameter, you can more accurately analyze the dynamic characteristics of the signal. For example, if the signal strength change rate is positive, it means the signal strength is increasing; if it is negative, it means the signal strength is decreasing.

[0083] Step S1235: performing smoothing filtering on the signal strength change rate parameter to generate a stable signal strength change rate parameter sequence.

[0084] Since there may be noise and interference in the actual sampling process, which causes the signal strength change rate parameter to fluctuate, in order to obtain more stable and reliable signal strength change rate information, it is necessary to perform smoothing filtering on the signal strength change rate parameter.

[0085] The signal strength rate of change parameter is processed by using an appropriate filtering algorithm, such as a moving average filter or a median filter. The filtering algorithm can smooth the signal strength rate of change parameter, eliminate noise and outliers, and generate a stable signal strength rate of change parameter sequence, thereby more accurately reflecting the trend of signal strength changes over time.

[0086] Step S130: calling a pre-trained personnel positioning model to perform joint positioning analysis on spatial distribution features and temporal variation features, and generating a set of potential personnel position candidates corresponding to the communication signal unit.

[0087] The pre-trained personnel positioning model is trained based on a large amount of historical data and simulation scenarios. It can comprehensively analyze the spatial distribution characteristics and temporal change characteristics of the input, thereby inferring the potential personnel location corresponding to the communication signal unit.

[0088] Step S131: inputting the spatial distribution features into the spatial feature processing module of the personnel positioning model, and generating a signal intensity distribution prediction map covering the fire scene area through a spatial interpolation algorithm.

[0089] The spatial feature processing module of the personnel positioning model receives spatial distribution features, namely the matrix of correspondences between signal strength and spatial position. Using a spatial interpolation algorithm, this module predicts the signal strength of other unsampled locations within the fire scene based on the known signal strength values ​​of the spatial grid cells.

[0090] The spatial interpolation algorithm takes into account the correlation of spatial locations and the characteristics of signal propagation. By applying weighted averages and other calculation methods to known signal strength values, it generates a signal strength distribution prediction map covering the entire fire scene. This signal strength distribution prediction map provides a more comprehensive picture of the spatial distribution of signals at the fire scene.

[0091] Step S132: input the time series variation characteristics into the time feature processing module of the personnel positioning model, and generate an evolution trend curve of the signal strength in the time dimension through a time series prediction algorithm.

[0092] The temporal feature processing module is an important component of the personnel location model. It is used to process the input temporal variation features. The input temporal variation features are a stable sequence of signal strength change rate parameters obtained after a series of processing. This sequence reflects the temporal variation trend of signal strength.

[0093] The time series prediction algorithm is the core algorithm of the temporal feature processing module. It predicts future changes in signal strength based on historical signal strength rate of change parameters. In real-world fire scenarios, signal strength fluctuates over time and may be affected by various factors, such as the movement of trapped individuals and changing environmental conditions. The time series prediction algorithm takes these factors into account and identifies patterns in signal strength changes by learning and analyzing historical data.

[0094] During the execution of a time series forecasting algorithm, it may first perform data preprocessing on the input time series variation characteristics. This may include checking data integrity to ensure there are no missing values ​​or outliers. If missing values ​​exist, they may be supplemented using interpolation; if outliers exist, they may be filtered or removed. The algorithm then selects an appropriate model structure based on the data characteristics, such as the Autoregressive Integrated Moving Average (ARIMA) model or the Long Short-Term Memory (LSTM) network. Different model structures are suitable for different types of time series data, and selecting the appropriate model can improve forecast accuracy.

[0095] The Autoregressive Integrated Moving Average (ARIMA) model primarily considers the autocorrelation and trend of the data. The ARIMA model determines the model order based on historical data, namely the order of the autoregressive term, the differencing order, and the moving average term. Adjusting these orders allows the model to better fit the historical data. During the fitting process, methods such as least squares can be used to estimate the model parameters to minimize the error between the model's predicted values ​​and the actual values.

[0096] The Long Short-Term Memory (LSTM) network is a deep learning model capable of processing time series data with long-term dependencies. Signal intensity variations at a fire scene may exhibit long-term trends and cyclical changes, which the LSTM network can capture through its unique gating mechanism. When training an LSTM network, the input temporal variation features can be divided into a training set and a validation set. The training set is used to update the network weights so that the network can learn the characteristics of the data; the validation set is used to evaluate the network's performance and prevent overfitting. Through multiple iterations of training, the network weights are continuously adjusted until satisfactory performance is achieved.

[0097] A time series prediction algorithm generates a temporal trend curve for signal strength. This curve reflects the changing trend of signal strength over time, such as whether it is gradually increasing, decreasing, or remaining stable. Analysis of this trend curve provides further insights into the activities of trapped individuals and the status of the signal source.

[0098] Step S133: input the signal strength distribution prediction graph and the evolution trend curve into the feature fusion module of the personnel positioning model, perform feature association analysis based on preset spatiotemporal association rules, and generate a fusion feature set with spatiotemporal consistency constraints.

[0099] The feature fusion module is a key component of the personnel location model, integrating spatial distribution features with temporal variation features to generate more comprehensive and accurate information. The input signal strength distribution prediction map is generated by the spatial feature processing module, showing the spatial distribution of the signal at the fire scene. The evolution trend curve is generated by the temporal feature processing module, reflecting the temporal variation of signal strength.

[0100] First, the signal strength distribution prediction map is discretized in its spatial dimension. Since the signal strength distribution prediction map is a continuous image, it needs to be converted into a discrete set of spatial grid intensity values ​​to facilitate subsequent processing and analysis. This process is similar to the spatial grid division process in step S122, which divides the continuous signal strength distribution into multiple discrete spatial grid cells and assigns a corresponding signal strength value to each grid cell. Thus, the signal strength distribution prediction map is converted into a set consisting of multiple spatial grid cells and their corresponding signal strength values.

[0101] Next, the evolution trend curve is discretized in the time dimension. The evolution trend curve is a continuous curve and needs to be converted into a set of discrete time node strength change values. The evolution trend curve is sampled at set intervals to obtain the signal strength change value corresponding to each time node. This converts the evolution trend curve into a set consisting of multiple time nodes and their corresponding signal strength change values.

[0102] Next, a spatiotemporal index relationship is established between the set of spatial grid strength values ​​and the set of time node strength change values. This spatiotemporal index relationship represents the expected signal strength value of the target spatial grid cell at the target time node. This index relationship allows information in both spatial and temporal dimensions to be linked. For example, what is the expected signal strength value for a spatial grid cell at a specific time node? The process of establishing the spatiotemporal index relationship must consider the continuity and correlation of space and time to ensure the accuracy and reliability of the index relationship.

[0103] Next, the spatial grid position and time node information of the actual signal subunits collected in the multi-source signal data set are extracted to obtain the expected signal strength value at the corresponding spatiotemporal index position. The actual signal subunits contain real signal strength information. By matching this information with the spatiotemporal index relationship, the expected signal strength value at the same spatial grid position and time node can be obtained.

[0104] The deviation between the actual signal strength and the expected signal strength is calculated and used as a spatiotemporal consistency constraint parameter. This deviation reflects the difference between the actual and expected signal conditions. A small deviation indicates good spatiotemporal signal consistency; a large deviation may indicate an anomaly, such as the movement of trapped personnel or a sudden change in environmental conditions.

[0105] Finally, by combining the spatial grid intensity value set, the time node intensity change value set, and the spatiotemporal consistency constraint parameters, a fused feature set is generated, which includes spatial intensity distribution, temporal evolution trends, and spatiotemporal deviation information. This fused feature set integrates information from both spatial and temporal dimensions and takes into account spatiotemporal consistency constraints, enabling a more comprehensive reflection of the spatiotemporal characteristics of fire scene signals.

[0106] Step S134: calling the position prediction module of the personnel positioning model to perform position probability calculation processing on the fused feature set to generate a personnel presence probability value for each position point in the fire scene area.

[0107] The location prediction module is the final output of the personnel location model. It calculates the probability of personnel presence at each location within the fire scene based on the input fused feature set. The fused feature set includes spatial intensity distribution, temporal evolution trends, and spatiotemporal deviation information. This information reflects the spatiotemporal characteristics of the signal and is closely related to the location of trapped personnel.

[0108] The location prediction module uses a classifier based on machine learning or deep learning. This classifier can be a support vector machine (SVM), decision tree, neural network, or other methods. The classifier is trained based on a large amount of historical data, which includes a set of fused features from different scenarios and the corresponding person location labels. By learning from this historical data, the classifier can establish a mapping relationship between the fused features and the person's location.

[0109] When a fused feature set is input, the classifier evaluates each location point. It compares the fused features at that location point with the learned patterns and calculates the probability that the location point belongs to the human presence category or the unpersonal category. Specifically, the classifier determines whether the signal strength at that location point meets the signal characteristics of a human presence based on the spatial intensity distribution information in the fused features; determines whether changes in signal strength are related to human activity based on temporal evolution trends; and determines whether there are any anomalies in the signal at that location point based on spatiotemporal deviation information, which could affect the probability of human presence.

[0110] The support vector machine (SVM) classifier can classify data into different categories by finding an optimal hyperplane. When calculating the probability of a person's presence, the distance between the input fused features and the hyperplane can be used to determine which category the location belongs to and calculate the corresponding probability. The decision tree classifier can recursively classify the data based on the different attribute properties of the fused features to ultimately determine the category and probability of the location. The neural network classifier uses multi-layer neuron calculations and nonlinear transformations to map the input fused features to different categories and output corresponding probability values.

[0111] The classifier generates probability values ​​for the presence of people at each location within the fire scene. These values ​​indicate the likelihood of a person being present at each location. Higher values ​​indicate a greater likelihood of a person being present at that location. By analyzing these probability values, we can preliminarily determine the locations where trapped people may be located.

[0112] Step S135: extracting location points whose probability of person presence exceeds a preset threshold as potential person location candidate points, and performing aggregation processing on the physical coordinate information of the potential person location candidate points to generate a potential person location candidate set.

[0113] After obtaining the probability of human presence at each location within the fire scene, we need to filter potential human location candidates based on a preset threshold. This threshold is determined based on actual conditions and experience, and is used to distinguish between locations with a high probability of human presence and those with a low probability.

[0114] The probability of a person being present at each location is compared with a preset threshold. If the probability of a person being present at a location exceeds the threshold, it indicates a high likelihood of a person being present at that location, and the location is selected as a potential candidate location. These potential candidate locations are selected based on the results of the fusion feature and classifier calculations and have a high degree of credibility.

[0115] The physical coordinate information of these potential personnel location candidate points is then aggregated. The physical coordinate information has been accurately recorded in step S111 and represents the actual location of each location point at the fire scene. By collecting and organizing the physical coordinate information of the potential personnel location candidate points, a set is formed, namely the potential personnel location candidate set. This potential personnel location candidate set contains the physical coordinate information of all locations where trapped people may be located.

[0116] Step S140: determining target location information of persons to be rescued at the fire scene based on the potential personnel location candidate set.

[0117] The set of potential location candidates is only a preliminary candidate range and may contain some false candidate points, or the actual location of the trapped person may not completely overlap with the candidate points. Therefore, the set of potential location candidates needs to be further processed to determine the target location information of the people to be rescued at the fire scene.

[0118] Step S141: performing signal source uniqueness verification processing on each potential personnel position candidate point in the potential personnel position candidate set to determine whether different scanning point signal sub-units point to the same position point.

[0119] At a fire scene, due to factors such as signal propagation and reflection, signal subunits at different scanning points may point to the same location, or a single location may be affected by multiple signal sources, resulting in false candidate points. Therefore, it is necessary to verify the uniqueness of the signal source for each potential person location candidate in the potential person location candidate set.

[0120] For each potential candidate location, the signal subunits from different scanning points associated with it are analyzed. The characteristics of these subunits, such as signal strength, frequency, and phase, are examined to determine whether they originate from the same source. If multiple subunits have similar characteristics, and their propagation paths and temporal relationships are consistent with originating from the same source, then they can be assumed to be pointing to the same location.

[0121] During verification, factors such as signal multipath and attenuation can be considered. Multipath causes signals to undergo multiple reflections and refractions during propagation, resulting in variations in their arrival time and intensity. By analyzing signal arrival time differences and signal intensity attenuation patterns, the uniqueness of the signal source can be more accurately determined. If a potential candidate location is affected by multiple signal sources, or if the signal subunit characteristics vary significantly across different scanning points, the candidate may be a false candidate and require further screening or exclusion.

[0122] Step S142: extracting potential personnel position candidate points that have passed uniqueness verification as valid candidate points, and calculating the confidence score of each valid candidate point in the fused feature set.

[0123] After the signal source uniqueness verification process, the potential personnel location candidate points that pass the verification are extracted as valid candidate points. These valid candidate points have higher credibility and are more likely to be the actual location of the trapped person.

[0124] To further evaluate the reliability of each valid candidate point, it is necessary to calculate its confidence score in the fused feature set. The confidence score is an indicator that takes into account multiple factors and reflects the degree of match between the valid candidate point and the information in the fused feature set.

[0125] When calculating the confidence score, factors such as the spatial intensity distribution, temporal evolution trend, and spatiotemporal deviation information in the fused feature set can be considered. For example, if the signal strength of a valid candidate point spatially matches the high-value area in the signal strength distribution prediction map, and its temporal evolution trend of signal strength is consistent with the evolution trend curve, and the spatiotemporal deviation information is small, then the confidence score of the valid candidate point will be high. Conversely, if the signal characteristics of a valid candidate point do not match the information in the fused feature set, or there is a large spatiotemporal deviation, then its confidence score will be low.

[0126] The specific calculation method can be a weighted summation approach. A weight is assigned to each factor in the fused feature set. The valid candidate point is scored based on its performance on each factor. These scores are then multiplied by the corresponding weight and summed to obtain the confidence score for the valid candidate point. The weight assignment is determined based on actual conditions and experience; different factors may have different importance in personnel location.

[0127] Step S143: sorting the confidence scores of the valid candidate points, and selecting the top K valid candidate points with the highest confidence scores as target candidate points, where K is a positive integer greater than 0.

[0128] After obtaining the confidence score of each valid candidate point, these scores are sorted. They can be sorted in ascending or descending order. Here, descending order is used, that is, valid candidate points with higher confidence scores are ranked first.

[0129] After sorting, the top K valid candidate points with the highest confidence scores are selected as target candidate points. The value of K is determined based on the actual situation and rescue needs, and represents the number of candidate points that require special attention. Selecting the top K valid candidate points with the highest confidence scores improves the accuracy and efficiency of subsequent positioning and reduces unnecessary search range.

[0130] Step S144: converting the physical coordinate information of the target candidate point into a two-dimensional plane coordinate point set, performing density cluster analysis on the two-dimensional plane coordinate point set, and calculating the number of other coordinate points within a preset neighborhood around each coordinate point as a density value.

[0131] While the fire scene is three-dimensional, the actual rescue process focuses on two-dimensional location information. Therefore, the physical coordinates of the target candidate points are converted into a set of two-dimensional coordinate points. This process can be achieved through projection, where the coordinate points in three-dimensional space are projected onto a two-dimensional plane to obtain the corresponding two-dimensional coordinate points.

[0132] After obtaining the set of two-dimensional plane coordinate points, density cluster analysis is performed on them. Density cluster analysis is a clustering method based on the density of data points, which can divide data points into different clusters, each cluster representing a potential gathering area for people.

[0133] In density cluster analysis, a predefined neighborhood is first determined. For each 2D coordinate point, the number of other coordinate points within the predefined neighborhood is calculated. This number represents the density of that point. The density value reflects the concentration of people around that point. A high density value for a point indicates that a large number of trapped people are likely nearby.

[0134] Step S145: Filter out coordinate points whose density values ​​exceed a preset density threshold as core points, and select coordinate points whose distances from the core points are less than a preset distance threshold as boundary points.

[0135] Based on the calculated density value for each coordinate point, coordinate points with a density exceeding a preset density threshold are selected as core points. The preset density threshold is determined based on actual conditions and experience, and represents the minimum density requirement for gathering people. Core points are the center of high-density areas. They are surrounded by a large number of coordinate points and are more likely to be the gathering areas for trapped people.

[0136] Then, coordinate points whose distance from the core point is less than a preset distance threshold are considered boundary points. The preset distance threshold is also determined based on actual conditions and represents the influence range around the core point. Although the density of boundary points may not be high, their proximity to the core point may indicate that they belong to the same gathering area as the core point.

[0137] Step S146: construct a candidate point clustering area based on the core points and boundary points, and extract the minimum enclosing rectangle of the candidate point clustering area as the initial area boundary.

[0138] After determining the core points and boundary points, a candidate point clustering region is constructed based on these points. The core points and boundary points can be connected to form a closed region, which is the candidate point clustering region.

[0139] To more conveniently describe and analyze the candidate point cluster area, we extract the minimum enclosing rectangle of the area as the initial region boundary. The minimum enclosing rectangle is a rectangle that completely encloses the candidate point cluster area and has the smallest area. The initial region boundary is obtained by calculating the coordinates of the vertices in the candidate point cluster area and finding the minimum rectangle that can contain these vertices. The initial region boundary provides a preliminary range for subsequent region optimization and target location determination.

[0140] Step S147: performing contour optimization processing on the initial region boundary, adjusting the coordinates of the boundary vertices to make the boundary more consistent with the actual dense region where the candidate points are distributed, and generating the final dense region boundary where the candidate points are distributed.

[0141] The initial region boundary is only a preliminary estimate and may deviate from the actual dense area where the candidate points are distributed. Therefore, it is necessary to optimize the contour of the initial region boundary.

[0142] During the optimization process, the coordinates of the vertices on the initial region boundary can be adjusted based on the distribution of candidate points. By analyzing the density distribution and positional relationships of candidate points within the region, the boundary vertices are moved toward the direction where the candidate points are more densely distributed, so that the boundary is more closely aligned with the actual dense area of ​​candidate point distribution.

[0143] When adjusting vertex coordinates, multiple factors can be considered. For example, we can ensure that the adjusted boundary still completely encloses the candidate point cluster area while minimizing the boundary area to improve positioning accuracy. The smoothness of the boundary is also considered to avoid overly sharp or irregular boundary shapes. Through multiple iterative adjustments, the boundary shape is continuously optimized until it closely matches the actual dense area of ​​candidate point distribution, generating the final boundary of the dense area of ​​candidate point distribution.

[0144] Step S148: Calculate the geometric center coordinates of the dense area boundary as the target position coordinates, and combine them with the coverage range parameters of the dense area boundary to generate target position information including the target position coordinates and coverage range information.

[0145] After obtaining the final dense region boundary for the candidate point distribution, calculate the geometric center coordinates of this boundary. The geometric center coordinates are the average of the coordinates of all vertices in the dense region boundary and represent the center of the region. These geometric center coordinates are the target location coordinates, which are the approximate location of the person to be rescued at the fire scene.

[0146] At the same time, the coverage parameters of the dense area boundary, such as its length, width, and area, are combined to generate target location information containing both the target's coordinates and coverage information. This coverage information helps rescuers better understand the distribution of those in need of rescue, providing more detailed information for developing rescue plans. For example, rescuers can rationally allocate rescue forces and routes based on the size and shape of the coverage area, improving rescue efficiency.

[0147] Step S150: Generate a rescue guidance instruction including the location coordinates based on the target location information, and send the rescue guidance instruction to the rescue terminal to trigger a personnel search and rescue operation.

[0148] After obtaining the target location information including the target location coordinates and coverage information, it needs to be converted into specific rescue guidance instructions so that rescuers can accurately reach the location of the person to be rescued according to the instructions.

[0149] Step S151: extracting the target location coordinates and coverage information from the target location information, and generating positioning data containing a specific location description.

[0150] Extract the target location coordinates and coverage information from the target location information. The target location coordinates are the geometric center coordinates of the previously calculated dense area boundary, which clearly define the approximate location of the person to be rescued. The coverage information includes parameters such as the length, width, and area of ​​the dense area boundary, which describe the possible distribution range of the person to be rescued.

[0151] The target location coordinates and coverage information are combined to generate positioning data containing a specific location description. When generating positioning data, the target location coordinates and coverage information can be converted into a more understandable location description based on the actual layout and coordinate system of the commercial complex. For example, if the commercial complex is managed using floor and area numbering, the positioning data will clearly indicate the floor and specific area where the target location is located, as well as the approximate scope of that area.

[0152] Step S152: Obtain environmental map data of the fire scene, perform coordinate matching processing on the positioning data and the environmental map data, and generate a rescue map superimposed with a target location mark.

[0153] Obtain environmental map data for the fire scene. This can be architectural blueprints of the commercial complex or a 3D map previously scanned using equipment such as LiDAR. The environmental map data includes detailed layout information for the commercial complex, such as floor plans, corridor locations, and room layouts.

[0154] Perform coordinate matching between the positioning data and the environmental map data. Since the target location coordinates in the positioning data are based on the internal coordinate system of the commercial complex, and the environmental map data also has its own coordinate system, coordinate conversion and matching are required to ensure that the positioning data can be accurately superimposed on the environmental map. During the matching process, the coordinates of the positioning data are converted to the coordinate system of the environmental map data based on information such as the origin, coordinate axis direction, and scale of the two coordinate systems.

[0155] After coordinate matching is complete, the positioning data is overlaid onto the environmental map as a marker, generating a rescue map with the target location marker. The target location marker can be a prominent icon or a colored area, effectively displaying the location and distribution of the person to be rescued within the commercial complex. The rescue map provides an intuitive visual reference for rescuers, allowing them to quickly understand the relationship between the person's location and the surrounding environment.

[0156] Step S153: Analyze the path accessibility information in the rescue map and plan the shortest feasible path from the current location of the rescue terminal to the target location.

[0157] Rescue maps contain not only target location information but also path accessibility information within the fire scene. This information can be obtained by analyzing environmental map data, such as whether passages are blocked by obstacles and whether doors can be opened normally. To analyze path accessibility information, image processing and machine learning algorithms can be used to identify and analyze images within the rescue map, marking accessible paths and impassable areas.

[0158] Based on the parsed path accessibility information, the shortest feasible path from the rescue terminal's current location to the target location is planned. This can be accomplished using graph search algorithms, such as Dijkstra's algorithm or the A* algorithm. These algorithms abstract the rescue map into a graph, where nodes represent locations on the map and edges represent connections between nodes and their cost. The cost can be determined based on factors such as the length of the path and the presence of obstacles.

[0159] When planning a route, you can start from the rescue terminal's current location and gradually expand the search range to find the shortest path to the target location. During the search, you can prioritize paths with lower travel costs while avoiding inaccessible areas. Ultimately, you will find the shortest feasible path from the rescue terminal's current location to the target location.

[0160] Step S154: converting the coordinate sequence of the shortest feasible path into navigation instruction information, where the navigation instruction information includes path direction guidance and key node prompts.

[0161] After obtaining the shortest feasible path, its coordinate sequence is converted into navigation instruction information. Navigation instruction information is a form of instruction that is easier for rescuers to understand and execute, and it contains path direction guidance and key node prompts.

[0162] First, the coordinate sequence of the shortest feasible path is analyzed to determine the path's direction. Based on the relative positions of adjacent coordinate points, the path is determined to indicate whether it is forward, backward, left, or right. This directional information is converted into specific text descriptions, such as "go straight ahead," "turn left," or "turn right."

[0163] At the same time, key nodes on the shortest feasible path are identified, such as intersections, elevator entrances, and stairwells. For each key node, corresponding prompts can be added, such as "Turn left at the intersection ahead" or "Reach the elevator entrance and take the elevator to the designated floor." These key node prompts can help rescuers accurately find the path in complex environments.

[0164] The path direction guidance and key node prompts are integrated to generate navigation instruction information. The navigation instruction information will be arranged in sequence according to the path to form a clear navigation instruction sequence.

[0165] Step S155: Fuse the positioning data, rescue map and navigation instruction information to generate a rescue guidance instruction including location coordinates, environmental map marks and navigation path instructions.

[0166] The generated positioning data, rescue map, and navigation instruction information are integrated to generate rescue guidance instructions containing location coordinates, environmental map markers, and navigation path instructions. During the integration process, the specific location description of the positioning data, the visual information of the rescue map, and the text instructions of the navigation instruction information can be combined to form a complete instruction package.

[0167] Rescue guidance instructions can be stored as digital files, containing the location of the person to be rescued, a rescue map with the target location overlaid on it, and detailed navigation route instructions. This instruction package can be sent to the rescue terminal via a wireless communication network, providing rescuers with comprehensive rescue guidance information.

[0168] Step S156: The rescue guidance instruction is sent to the display module of the rescue terminal via the wireless communication network. The display module graphically displays the rescue guidance instruction content to assist rescue personnel in performing search and rescue operations.

[0169] The rescue guidance instructions are sent to the display module of the rescue terminal via a wireless communication network. The wireless communication network can be Wi-Fi, Bluetooth, or a cellular network, ensuring that the rescue guidance instructions can be transmitted to the rescue terminal quickly and stably.

[0170] After receiving the rescue guidance instructions, the rescue terminal's display module graphically displays the instructions. The display module displays a rescue map with the target location overlaid on it, allowing rescuers to intuitively monitor the location of the person to be rescued and the layout of the surrounding environment. The display module also marks the navigation path from the rescue terminal's current location to the target location on the map, and displays real-time path direction guidance and key node prompts in the order of the navigation instructions.

[0171] Rescuers can quickly and accurately reach the location of the person being rescued based on the rescue guidance instructions displayed on the display module. During the rescue process, rescuers can also adjust the navigation path according to the actual situation. The display module will update the path information in real time to ensure the smooth progress of the rescue operation.

[0172] Throughout the entire process of personnel detection at fire scenes, data security and privacy protection must be considered in all aspects of data collection, processing, and transmission. The collected communication signal data, which may contain personal information of trapped individuals, such as mobile phone numbers, requires encryption technology to prevent data leakage. For data storage, secure and reliable storage devices and methods can be used to ensure data integrity and availability. Furthermore, secure communication protocols, such as SSL / TLS, can be used to encrypt data transmission to prevent theft or tampering during transmission.

[0173] Furthermore, training a personnel location model requires a large amount of historical data and simulated scenario data. Historical data can include signal data and corresponding personnel location information collected in similar fire scenarios, while simulated scenario data can be generated using computer simulation software. During training, this data can be divided into training, validation, and test sets. The training set is used to train the model's parameters, the validation set is used to adjust the model's structure and hyperparameters, and the test set is used to evaluate the model's performance.

[0174] The structure of the personnel localization model includes a spatial feature processing module, a temporal feature processing module, a feature fusion module, and a location prediction module. Data is transmitted and processed between these modules through predefined connections. During training, optimization algorithms such as backpropagation can be used to continuously adjust model parameters to ensure that the model output matches the actual personnel location information as closely as possible. Training parameters include the learning rate, number of iterations, and batch size; their selection affects the model's training results and performance. Through continuous training and optimization, the personnel localization model learns the complex relationship between signal characteristics and personnel location, improving the accuracy and reliability of personnel localization at fire scenes.

[0175] Figure 2 A schematic diagram illustrates exemplary hardware and software components of a personnel detection system 100 for fire scenes, which can implement the concepts of the present application, provided in some embodiments of the present application. For example, a processor 120 can be used in the personnel detection system 100 for fire scenes and used to perform the functions of the present application.

[0176] The personnel detection system 100 applied to a fire scene can be a general-purpose server or a special-purpose server, both of which can be used to implement the personnel detection method applied to a fire scene of the present application. Although only one server is shown in this application, for convenience, the functions described in this application can be implemented in a distributed manner on multiple similar platforms to balance the processing load.

[0177] For example, a personnel detection system 100 for use at a fire scene may include a network port 110 connected to a network, one or more processors 120 for executing program instructions, a communication bus 130, and various forms of storage media 140, such as a disk, ROM, or RAM, or any combination thereof. For example, the personnel detection system 100 for use at a fire scene may also include program instructions stored in ROM, RAM, or other types of non-transitory storage media, or any combination thereof. The method of the present application may be implemented based on these program instructions. The personnel detection system 100 for use at a fire scene may also include an I / O interface 150 between the computer and other input and output devices.

[0178] For ease of explanation, only one processor is described in the personnel detection system 100 for use at a fire scene. However, it should be noted that the personnel detection system 100 for use at a fire scene in this application may also include multiple processors, and therefore the steps described herein as being performed by one processor may also be performed jointly or individually by multiple processors. For example, if the processor of the personnel detection system 100 for use at a fire scene performs steps A and B, it should be understood that steps A and B may also be performed jointly by two different processors or individually by a single processor. For example, the first processor may perform step A, the second processor may perform step B, or the first and second processors may perform steps A and B jointly.

[0179] In addition, an embodiment of the present invention further provides a readable storage medium, in which computer-executable instructions are preset. When a processor executes the computer-executable instructions, the personnel detection method applied to a fire scene as described above is implemented.

[0180] It should be noted that in order to simplify the description of the present invention and thus help understand one or more embodiments of the invention, in the foregoing description of the embodiments of the present invention, multiple features are sometimes combined into one embodiment, figure or description thereof.

Claims

1. A method for detecting people at a fire scene, characterized in that: The method comprises: Acquire a multi-source signal data set at a fire scene, wherein the multi-source signal data set includes communication signal units with time stamps collected at different locations; Performing signal feature extraction processing on the multi-source signal data set to obtain spatial distribution features and temporal variation features of the communication signal unit; Calling a pre-trained personnel positioning model to perform joint positioning analysis on the spatial distribution characteristics and the temporal variation characteristics to generate a set of potential personnel position candidates corresponding to the communication signal unit; Determine target location information of persons to be rescued at the fire scene based on the potential personnel location candidate set; generating a rescue guidance instruction including location coordinates based on the target location information, and sending the rescue guidance instruction to a rescue terminal to trigger a personnel search and rescue operation; in, The performing signal feature extraction processing on the multi-source signal data set to obtain the spatial distribution features and temporal variation features of the communication signal units includes: Performing signal strength value extraction processing on each signal subunit in the multi-source signal data set to obtain a signal strength parameter of each signal subunit; Based on the physical coordinate information of the scanning points and the signal strength parameters corresponding to the signal subunits, a corresponding relationship matrix between signal strength and spatial position is constructed as a spatial distribution feature, wherein the spatial distribution feature includes a mapping relationship between different scanning points and corresponding signal strengths; Performing time dimension analysis on the continuous time series signal subunits in the multi-source signal data set, and calculating a signal strength change rate parameter between adjacent time stamp signal subunits; Extracting a signal strength fluctuation frequency parameter of the signal subunit within a preset time window, wherein the fluctuation frequency parameter represents a change period of the signal strength in a time dimension; Performing feature fusion processing on the signal strength change rate parameter and the fluctuation frequency parameter to generate a time series change feature reflecting the law of signal strength change over time; The calling of the pre-trained personnel positioning model to perform a joint positioning analysis on the spatial distribution features and the temporal variation features to generate a set of potential personnel position candidates corresponding to the communication signal unit includes: Inputting the spatial distribution features into the spatial feature processing module of the personnel positioning model, and generating a signal intensity distribution prediction map covering the fire scene area through a spatial interpolation algorithm; Inputting the time series variation characteristics into the time feature processing module of the personnel positioning model, and generating an evolution trend curve of the signal strength in the time dimension through a time series prediction algorithm; Inputting the signal strength distribution prediction graph and the evolution trend curve into the feature fusion module of the personnel positioning model, performing feature association analysis based on preset spatiotemporal association rules, and generating a fusion feature set with spatiotemporal consistency constraints; Calling the position prediction module of the personnel positioning model to perform position probability calculation processing on the fused feature set to generate a probability value of personnel presence at each position point in the fire scene area; The location points where the probability of the person existing exceeds a preset threshold are extracted as potential person location candidate points, and the physical coordinate information of the potential person location candidate points is aggregated to generate a potential person location candidate set.

2. The method for detecting people at a fire scene according to claim 1, wherein: The multi-source signal data set of the fire scene is obtained, wherein the multi-source signal data set includes communication signal units with time stamps collected at different locations, including: Mobile detection equipment deployed at the fire scene performs signal scanning operations at different spatial points and records the physical coordinate information of each scanning point; During the signal scanning process at each scanning point, the communication signal in the environment is sampled and processed according to the preset time interval to generate a signal sampling segment with a continuous time series; Add the corresponding scanning point physical coordinate information and sampling timestamp information to each signal sampling segment to form a basic signal unit with a temporal and spatial correlation relationship; Performing communication protocol parsing on the basic signal units to screen out valid signal units that meet preset communication standards, wherein the preset communication standards include cellular network communication protocols and short-range wireless communication protocols; The effective signal units are sequentially reorganized in the order of sampling timestamps to generate a multi-source signal data set containing multiple spatiotemporally correlated signal sub-units, each signal sub-unit in the multi-source signal data set carries the corresponding scanning point physical coordinate information and sampling timestamp information.

3. The method for detecting people at a fire scene according to claim 1, wherein: The method of constructing a corresponding relationship matrix between signal intensity and spatial position as a spatial distribution feature based on the physical coordinate information of the scanning points and the signal strength parameters corresponding to the signal subunits includes: Extracting the physical coordinate information of all scanning points in the signal subunit to form a spatial sampling point set containing multiple coordinate points; Associating each coordinate point in the set of spatial sampling points with the corresponding signal strength parameter one by one to generate a target association data pair; Performing spatial grid division processing on the target associated data pair to divide the fire scene area into a plurality of evenly distributed spatial grid units; Calculate the average signal strength of all target-associated data pairs within each spatial grid cell to generate an average signal strength value for each spatial grid cell; The coordinate range information of the spatial grid unit and the corresponding average signal strength value are matrix-processed to generate a corresponding relationship matrix between signal strength and spatial position reflecting the law of signal strength changing with spatial position as a spatial distribution feature.

4. The method for detecting people at a fire scene according to claim 1, wherein: The performing time dimension analysis processing on the continuous time series signal subunits in the multi-source signal data set and calculating the signal strength change rate parameter between adjacent time stamp signal subunits includes: Extracting timestamp information and corresponding signal strength parameters of the continuous time series signal subunits to generate time-intensity series data; performing time interval normalization processing on the time-intensity series data so as to make the time intervals of adjacent time stamp signal subunits consistent; Traversing each pair of adjacent signal subunits in the time-intensity series data, and calculating the difference between the signal intensity parameter of the subsequent timestamp signal subunit and the signal intensity parameter of the previous timestamp signal subunit; Dividing the difference by the time interval between adjacent timestamps to obtain a signal strength change rate parameter between adjacent timestamp signal subunits; Smoothing filtering is performed on the signal strength change rate parameter to generate a stable signal strength change rate parameter sequence.

5. The method for detecting people at a fire scene according to claim 1, wherein: The signal strength distribution prediction graph and the evolution trend curve are input into the feature fusion module of the personnel positioning model, and feature association analysis is performed based on preset spatiotemporal association rules to generate a fusion feature set with spatiotemporal consistency constraints, including: Performing spatial dimension discretization processing on the signal strength distribution prediction map to convert the continuous signal strength distribution into a discrete set of spatial grid strength values; Performing time dimension discretization processing on the evolution trend curve to convert the continuous time evolution trend into a discrete set of time node intensity change values; Establishing a spatiotemporal index relationship between the spatial grid strength value set and the time node strength change value set, the spatiotemporal index relationship representing an expected value of the signal strength of a target spatial grid cell at a target time node; Extracting the spatial grid position and time node information of the signal subunit actually collected in the multi-source signal data set, and obtaining the expected value of the signal strength at the corresponding spatiotemporal index position; Calculating a deviation between an actual signal strength value and an expected signal strength value, and using the deviation as a spatiotemporal consistency constraint parameter; The spatial grid intensity value set, the time node intensity change value set and the spatiotemporal consistency constraint parameter are combined to generate a fusion feature set including spatial intensity distribution, time evolution trend and spatiotemporal deviation information.

6. The method for detecting people at a fire scene according to claim 1, characterized in that: The step of determining target location information of persons to be rescued at the fire scene based on the potential personnel location candidate set includes: Performing signal source uniqueness verification processing on each potential personnel position candidate point in the potential personnel position candidate set to determine whether different scanning point signal subunits point to the same position point; Extracting potential personnel position candidate points that have passed uniqueness verification as valid candidate points, and calculating the confidence score of each valid candidate point in the fused feature set; Sorting the confidence scores of the valid candidate points, and selecting the top K valid candidate points with the highest confidence scores as target candidate points, where K is a positive integer greater than 0; Converting the physical coordinate information of the target candidate point into a set of two-dimensional plane coordinate points, performing density cluster analysis on the set of two-dimensional plane coordinate points, and calculating the number of other coordinate points within a preset neighborhood around each coordinate point as a density value; Filter out coordinate points whose density values ​​exceed the preset density threshold as core points, and select coordinate points whose distance from the core points is less than the preset distance threshold as boundary points; Constructing a candidate point clustering region based on the core points and the boundary points, and extracting the minimum enclosing rectangle of the candidate point clustering region as an initial region boundary; Performing contour optimization processing on the initial region boundary, adjusting the coordinates of the boundary vertices so that the boundary is more closely aligned with the actual dense region where the candidate points are distributed, and generating a final dense region boundary where the candidate points are distributed; The geometric center coordinates of the dense area boundary are calculated as the target position coordinates, and combined with the coverage range parameters of the dense area boundary to generate target position information including the target position coordinates and coverage range information.

7. The method for detecting people at a fire scene according to claim 1, wherein: The generating of a rescue guidance instruction including position coordinates based on the target position information, and sending the rescue guidance instruction to a rescue terminal to trigger a personnel search and rescue operation, includes: Extracting target location coordinates and coverage information from the target location information to generate positioning data containing a specific location description; Obtaining environmental map data of the fire scene, performing coordinate matching processing on the positioning data and the environmental map data, and generating a rescue map superimposed with target location marks; Analyzing the path accessibility information in the rescue map and planning the shortest feasible path from the current location of the rescue terminal to the target location; Converting the coordinate sequence of the shortest feasible path into navigation instruction information, wherein the navigation instruction information includes path direction guidance and key node prompts; Fusing the positioning data, rescue map and navigation instruction information to generate rescue guidance instructions including location coordinates, environmental map markings and navigation path instructions; The rescue guidance instruction is sent to the display module of the rescue terminal through the wireless communication network, and the display module displays the rescue guidance instruction content in a graphical manner to assist rescue personnel in performing search and rescue operations.

8. A personnel detection system used in a fire scene, characterized in that: It includes a processor and a memory, the memory is connected to the processor, the memory is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the memory to implement the personnel detection method applied to the fire scene as described in any one of claims 1 to 7 above.

Images