Virtual training method and system for search, rescue and escape of firefighter

Abstract

The invention relates to the technical field of virtual training, in particular to a firefighter search, rescue and escape virtual training method and system, and the system comprises a virtual scene dynamic construction module which abstracts a room and a corridor intersection into nodes according to a planar graph of a building, abstracts a door and a channel into edges with geometric distance weights, and carries out the virtual scene dynamic construction module; and responding to a fire spreading or structure collapse event in the virtual training, adjusting the weight of a corresponding edge to be infinite, obtaining a shortest path, and establishing a reference rescue path set. According to the method, through dynamic analysis of building plan structure information, rooms and corridor intersections are abstracted as nodes, doors and channels are abstracted as edges with geometric weights, the accessibility of the edges is adjusted in real time after fire spreading or structure collapse occurs, impassable conditions are introduced during path calculation, and the path calculation accuracy is improved. And the real-time adaptability and rescue simulation precision of the path generation process are improved.

Description

Technical Field

[0001] This invention relates to the field of virtual training technology, and in particular to a virtual training method and system for firefighters in search and rescue and escape. Background Technology

[0002] The field of virtual training technology falls under the interdisciplinary research area of ​​simulation and human-computer interaction. It is dedicated to building highly immersive, controllable, and adjustable virtual environments for users through digital modeling, visualization, real-time interaction, and behavior capture.

[0003] Current technologies in virtual training often employ preset scenarios or static paths, lacking feedback mechanisms for real-time environmental changes. This leads to situations where, in emergencies such as fire spread or passageway blockage, path selection still relies on the initial settings, failing to realistically simulate the interference of dynamic changes on judgment and decision-making, thus reducing the realism of the training scenario and the effectiveness of on-site response training. At the behavioral recording level, only location trajectories or simple interaction logs are typically collected, ignoring fine-grained information on head movement behavior and visual exploration efficiency, making it difficult to recreate the complete chain of observation, judgment, and execution by firefighters in complex environments. Therefore, improvements are needed. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by proposing a virtual training system for firefighters to search, rescue, and escape.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a virtual training system for firefighters' search and rescue escape includes: The virtual scene dynamic construction module abstracts the intersections of rooms and corridors into nodes based on the building floor plan, and the doors and passages into edges with geometric distance weights. In virtual training, it responds to fire spread or structural collapse events and adjusts the weights of the corresponding edges to infinity, obtains the shortest path, and establishes a set of reference rescue paths. The search and rescue behavior sequence quantization module records the firefighter's three-dimensional coordinate sequence, head posture quaternion sequence, and interaction events to form a search and rescue behavior spatiotemporal data stream. Based on the search and rescue behavior spatiotemporal data stream, a search and rescue path deviation metric is generated. The team communication link simulation module extracts the real-time three-dimensional coordinates of any two firefighters based on the spatiotemporal data stream of the search and rescue behavior, obtains the multipath signal attenuation value, and generates a real-time communication quality index by superimposing the interference caused by wall obstruction and steel structure scattering based on the multipath signal attenuation value. The search and rescue effectiveness comprehensive evaluation module extracts the static duration of firefighters' position coordinates at path decision points based on the spatiotemporal data stream of the search and rescue behavior, obtains the lag degree of key decision points, and makes a judgment based on the lag degree of key decision points, combined with the search and rescue path deviation metric and the real-time communication quality index of the same time window, and constructs an individual search and rescue capability diagnostic profile.

[0006] Preferably, the step of obtaining the reference rescue path set is as follows: Based on the building floor plan, the location information of the intersection of rooms and corridors and the spatial connection information of doors and passages are extracted from the building floor plan. The locations of all room and corridor intersections are mapped as abstract nodes, and the connection relationships of all doors and passages are mapped as edges between nodes, thus obtaining the initial node-edge topology network. Based on the initial node edge topology network, the actual length information of each door and passage in the building plan is called, and the actual length of each door and passage is used as the geometric distance weight of the edge. The geometric distance weight values ​​of all edges in the initial node edge topology network are updated to generate a node edge topology network with geometric distance weights. Based on the node-edge topology network with assigned geometric distance weights, the location of fire spread and structural collapse events in the virtual training scenario is monitored in real time. Doors and passages that are impassable due to fire spread or structural collapse are identified. The geometric distance weights of the edges represented by the corresponding doors and passages are updated to infinity. The shortest path from the entrance node to all trapped personnel location nodes and safety exit nodes is calculated, and a set of reference rescue paths is generated.

[0007] Preferably, the steps for acquiring the spatiotemporal data stream of the search and rescue operation are as follows: Record the firefighters' three-dimensional coordinate positions, head posture quaternions, and timestamps of each interaction event throughout the entire virtual training process. Align and synchronize the three-dimensional coordinate position sequence, head posture quaternion sequence, and interaction event markers with timestamps as indexes to generate a spatiotemporal data stream of search and rescue behavior.

[0008] Preferably, the step of obtaining the search and rescue path deviation metric is as follows: Based on the three-dimensional coordinate position sequence in the spatiotemporal data stream of the search and rescue behavior, a single reference rescue path that matches the current mission scenario is selected from the set of reference rescue paths as a comparison benchmark. The current position of the firefighter at each time point is vertically projected onto the selected reference rescue path using the shortest Euclidean distance to obtain the shortest geometric distance from the current position to the path. Combined with the number of new grids and the number of repeated grids swept by the line of sight cone at each time point, the instantaneous deviation feature parameters of all time points are extracted. Based on the instantaneous deviation characteristic parameters, the search and rescue path deviation metric is calculated.

[0009] Preferably, the step of obtaining the multipath signal attenuation value is as follows: Based on the spatiotemporal data stream of the search and rescue behavior, the three-dimensional coordinate positions of any two firefighters at the same current timestamp are extracted, a spatial ray path is established with these two positions as endpoints, and the structural distribution information in the building plan is called to determine the type and thickness of all building structures passed through the ray path one by one, generating a ray path structural distribution sequence. Based on the ray path structure distribution sequence, the electromagnetic parameters corresponding to each building material are called respectively. The thickness of each building material that the ray path passes through in sequence is multiplied with the corresponding electromagnetic parameters point by point and accumulated one by one to form a multipath signal attenuation value.

[0010] Preferably, the step of obtaining the real-time communication quality index is as follows: Based on the multipath signal attenuation value, the attenuation caused by the ray path passing through the wall structure and the attenuation caused by the scattering interference caused by the steel structure are calculated respectively. The attenuation caused by the ray path passing through the wall structure and the attenuation caused by the scattering interference are superimposed on the multipath signal attenuation value to obtain the total attenuation of the radio communication signal between firefighters. The real-time communication quality index is then calculated based on the total attenuation.

[0011] Preferably, the step of obtaining the hysteresis of the key decision point is as follows: Based on the spatiotemporal data stream of the search and rescue behavior, the path decision point segments in which the firefighters remain stationary are extracted from all three-dimensional coordinate position sequences. Within each path decision point segment, the stationary state of the firefighter's coordinates in three-dimensional space is analyzed second by second. The duration of the time period in which no coordinate changes occur is counted. At the same time, the continuous change sequence of all head posture quaternions within the time period is extracted to form a path decision point stay feature set. Calculate the lag degree of key decision points based on the set of path decision point dwell characteristics.

[0012] Preferably, the steps for obtaining the individual search and rescue capability diagnostic profile are as follows: Based on the lag at the key decision points, the search and rescue path deviation metric and the real-time communication quality index within the same time window are jointly analyzed to determine whether firefighters exhibit hesitation at path decision points, thereby obtaining an individual search and rescue capability diagnostic profile.

[0013] This invention also provides a virtual training method for firefighters in search and rescue escape, including the following steps: Based on the building floor plan, the intersections of rooms and corridors are abstracted as nodes, and the doors and passages are abstracted as edges with geometric distance weights. In virtual training, in response to fire spread or structural collapse events, the weights of the corresponding edges are adjusted to infinity, and the shortest path is obtained to establish a set of reference rescue paths. Record the firefighter's three-dimensional coordinate sequence, head posture quaternion sequence and interaction events to form a search and rescue behavior spatiotemporal data stream, and generate a search and rescue path deviation metric based on the search and rescue behavior spatiotemporal data stream; Based on the spatiotemporal data stream of the search and rescue behavior, the real-time three-dimensional coordinates of any two firefighters are extracted, and the multipath signal attenuation value is obtained. Based on the multipath signal attenuation value, the interference caused by wall obstruction and steel structure scattering is superimposed to generate a real-time communication quality index. Based on the spatiotemporal data stream of the search and rescue behavior, the static duration of the firefighter's position coordinates is extracted at the path decision point to obtain the hysteresis of the key decision point. Based on the hysteresis of the key decision point, combined with the search and rescue path deviation metric and the real-time communication quality index of the same time window, a diagnostic profile of individual search and rescue capabilities is constructed.

[0014] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, by dynamically analyzing the structural information of the building floor plan, the intersections of rooms and corridors are abstracted as nodes, and doors and passages are abstracted as edges with geometric weights. The accessibility of these edges is adjusted in real time after fire spreads or structural collapse occurs. Impassable conditions are introduced when calculating paths, improving the real-time adaptability of the path generation process and the accuracy of rescue simulation. Firefighter behavior is captured as a three-dimensional coordinate sequence, a head posture quaternion sequence, and a set of interactive events, forming a multi-dimensional data stream with time synchronization characteristics. Dynamic distance deviation is compared with a static reference path, and combined with the scanning coverage and repetition rate analysis of the environment by the line-of-sight cone, a comprehensive quantification of path execution deviation and search strategy efficiency is achieved. The quality of radio communication between two points is determined based on multi-path signal detection using actual spatial coordinates. The electromagnetic properties of building materials and the physical characteristics of path penetration are considered, along with the effects of obstacle obstruction and structural scattering on the signal, ensuring a realistic reproduction of the communication environment modeling and signal attenuation assessment. In decision point analysis, the duration of the defense personnel's stay and changes in head posture angular velocity were extracted, the hysteresis was quantified, and the degree of behavioral deviation and communication interference status were combined to determine the ability to complete the task, which enhanced the ability to extract cognitive response and executive performance from multi-source data. Attached Figure Description

[0015] Figure 1 This is a system flowchart of the present invention. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0017] Please see Figure 1This invention provides a technical solution: a virtual training system for firefighters' search and rescue escape includes: The virtual scene dynamic construction module abstracts the intersections of rooms and corridors into nodes based on the building floor plan, and the doors and passages into edges with geometric distance weights. In virtual training, it responds to fire spread or structural collapse events and adjusts the weights of the corresponding edges to infinity, obtains the shortest path, and establishes a set of reference rescue paths. The search and rescue behavior sequence quantification module records the firefighter's three-dimensional coordinate sequence, head posture quaternion sequence, and interaction events to form a search and rescue behavior spatiotemporal data stream. Based on the search and rescue behavior spatiotemporal data stream, a search and rescue path deviation metric is generated. The team communication link simulation module extracts the real-time three-dimensional coordinates of any two firefighters based on the spatiotemporal data stream of search and rescue behavior, obtains the multipath signal attenuation value, and generates a real-time communication quality index by superimposing the interference caused by wall obstruction and steel structure scattering based on the multipath signal attenuation value. The comprehensive evaluation module for search and rescue effectiveness extracts the static duration of firefighters' position coordinates at path decision points based on the spatiotemporal data stream of search and rescue behavior, obtains the lag degree at key decision points, and makes a judgment based on the lag degree at key decision points, combined with the search and rescue path deviation metric and the real-time communication quality index within the same time window, thus constructing a diagnostic profile of individual search and rescue capabilities.

[0018] The steps to obtain the reference rescue route set are as follows: Based on the building floor plan, the location information of the intersection of rooms and corridors and the spatial connection information of doors and passages are extracted from the building floor plan. The locations of all room and corridor intersections are mapped as abstract nodes, and the connection relationships of all doors and passages are mapped as edges between nodes, thus obtaining the initial node-edge topology network. Based on the initial node edge topology network, the actual length information of each door and passage in the building floor plan is called. The actual length of each door and passage is used as the geometric distance weight of the edge. The geometric distance weight values ​​of all edges in the initial node edge topology network are updated to generate a node edge topology network with geometric distance weights. Based on a node-edge topology network with geometric distance weights, the location of fire spread and structural collapse events in the virtual training scenario is monitored in real time. Doors and passages that are impassable due to fire spread or structural collapse are identified. The geometric distance weights of the edges represented by the corresponding doors and passages are updated to infinity. The shortest path from the entrance node to all trapped personnel location nodes and safety exit nodes is calculated, and a set of reference rescue paths is generated.

[0019] Specifically, based on the digital architectural floor plans loaded by the system, such as DWG or DXF format CAD files, key architectural elements are first identified through layers and object attributes. The system automatically parses closed polygonal regions marked "ROOM_AREA" and specific blocks or geometric center points marked "CORRIDOR_JUNCTION". For each identified room area, the three-dimensional coordinates of its geometric center are calculated and a unique identifier is assigned. This center point is abstracted as a room node. For corridor intersections, their coordinate points marked on the drawing are directly extracted as intersection nodes. All these room nodes and intersection nodes are aggregated into a node set. Subsequently, the system scans the drawing... For elements in the passageway, such as door symbols defined in the "DOORS" layer or linear paths drawn in the "PASSAGEWAYS" layer, the system identifies the two spatial regions it connects to for each door or passageway element. That is, it determines the two nodes it connects to by querying which two defined room regions or intersection regions the element is geometrically adjacent to or overlaps with. Once the connection relationship is determined, an edge is established between the unique identifiers of the two nodes and stored in an edge set. This process traverses all rooms, intersections, doors, and passageways in the drawing, and finally combines all nodes and their connection relationships to form an initial node-edge topology network consisting of node identifiers and edge connection pairs.

[0020] Based on the initial node-edge topology network generated in the previous step, the system again calls upon the detailed geometric information of the digital building floor plan to quantify and assign weights to each edge in the network. Specifically, the system traverses each edge in the initial node-edge topology network and executes different length calculation logics based on the physical entity type represented by the edge. If an edge represents a door, the system will assign it a fixed, small passage cost as its length. For example, the equivalent passage length of all doors is set to 1 meter. This value is an empirical value derived from the time required for a firefighter to pass through a doorway and their normal walking speed, used to reflect the small cost required to open a door or squeeze through a narrow space in path planning. If an edge represents a corridor or passageway, the system will extract the precise three-dimensional coordinates of the two nodes connected by the edge in the building floor plan, i.e. and Then, by calculating the Euclidean distance between these two points... To obtain the actual geometric length of the channel, and to calculate the length value As the weight of this edge, the system applies this calculation process to all edges in the network, replacing the unweighted connections in the initial network one by one with the calculated actual length or equivalent passage length, and finally generating a node edge topology network in which each edge is assigned a precise geometric distance weight value.

[0021] Based on a node-edge topology network with assigned geometric distance weights, the system continuously receives real-time event data generated by the fire dynamics simulation engine and the structural damage physics engine during virtual training. This event data includes event type (such as fire spread, local explosion, load-bearing wall collapse), 3D spatial coordinates, radius of influence, and timestamp. The system captures this data through an event listener and executes the following judgment and processing flow: For a fire spread event, defined as a center point coordinate and a high-temperature lethal radius (e.g., 5 meters), the system traverses all edges representing doors and passages in the node-edge topology network, calculating the 3D spatial distance between the midpoint of each edge or multiple sampling points on its geometric path and the center point of the fire event. If the distance between any sampling point and the fire center is less than the preset high-temperature lethal radius, the passage is determined to be impassable due to high temperature or dense smoke. For a structural collapse event, defined as a polygon of the collapsed area, the system will determine whether there are edges... If any path overlaps or intersects with the polygon of the collapsed area, and there is an overlap, the passage is determined to be physically blocked and impassable. For any door or passage determined to be impassable, the system will immediately find the corresponding edge in the node edge topology network with geometric distance weights and dynamically modify its geometric distance weight value to a maximum value, such as 9999999.0. This value is equivalent to infinity in path calculation, representing a path interruption. After completing the weight update, the system will immediately use the preset entrance node in the scene as the starting point, and all known trapped personnel location nodes and all safe exit nodes as the ending points, and call the A search algorithm to calculate on the dynamically updated topology network. The heuristic function of the A algorithm is the straight-line Euclidean distance from each node to the target node. The algorithm will avoid edges with infinite weights, thereby finding the current optimal path from the entrance to every possible target point. These calculated shortest paths together constitute a set of reference rescue paths for subsequent evaluation.

[0022] The steps for acquiring the spatiotemporal data stream of search and rescue operations are as follows: Record the firefighters' three-dimensional coordinate positions, head posture quaternions, and timestamps of each interaction event throughout the entire virtual training process. Align and synchronize the three-dimensional coordinate position sequence, head posture quaternion sequence, and interaction event markers with timestamps as indexes to generate a spatiotemporal data stream of search and rescue behavior.

[0023] Specifically, by directly interacting with the application programming interface of the virtual reality engine, the world coordinates of the firefighter's head-mounted display in virtual 3D space are continuously captured at a fixed high-frequency sampling rate, such as 90 times per second, and recorded as... The sequence, simultaneously, acquires its attitude information from the inertial measurement unit of the head-mounted display at the same sampling frequency, and uses quaternions. The data is recorded in the form of a quaternion describing the real-time orientation and rotation of the firefighter's head. In addition, the training application software itself acts as an independent event source, logging every meaningful interaction performed by the firefighter in the virtual environment. These interactions include, but are not limited to, "opening a door," "closing a door," "marking trapped personnel," "using a fire extinguisher," "activating the thermal imager," or "pressing the team communication button." Each event is accompanied by a unique timestamp generated by the system's high-precision performance counter and its associated metadata, such as the ID or name of the interacted object. Subsequently, the system executes a time alignment and synchronization process, first defining a uniform analysis time grid, for example, at a frequency of 10 Hz, i.e., every 100 milliseconds as a time step. For each time step, The system uses nearest neighbor interpolation to select the data point whose timestamp is closest to the current time step's timestamp from the high-frequency sampled 3D coordinate position sequence and head pose quaternion sequence. This data point is then used as the valid pose and position data for that time step. Simultaneously, the system retrieves and aggregates all interaction event markers whose timestamps fall within the 100-millisecond time window, assigning these discrete events to the current time step. Through this process, the system integrates three potentially asynchronous independent data streams—continuous position data, continuous pose data, and discrete interaction events—onto a unified time axis, forming a structured data table. Each row represents a synchronized time point, and the columns contain the timestamp, 3D coordinates, head pose quaternion, and a list of interaction events that occurred at that time point, ultimately generating a spatiotemporal data stream for search and rescue operations.

[0024] The steps for obtaining the search and rescue path deviation metric are as follows: Based on the three-dimensional coordinate position sequence in the spatiotemporal data stream of search and rescue behavior, a single reference rescue path that matches the current mission scenario is selected from the set of reference rescue paths as a comparison benchmark. The current position of the firefighter at each time point is vertically projected onto the selected reference rescue path using the shortest Euclidean distance to obtain the shortest geometric distance from the current position to the path. Combined with the number of new grids and the number of repeated grids swept by the line of sight cone at each time point, the instantaneous deviation feature parameters of all time points are extracted. Based on the instantaneous deviation feature parameters, the search and rescue path deviation metric is calculated using the following formula: ; in, Measurement of deviation from search and rescue route. Total number at time points For the first Three-dimensional coordinate vector of firefighters at each time point , Current location of firefighters In the reference path The coordinate vector of the projection point on the surface. This represents the Euclidean distance from the current location to the path. For the first The number of grids repeatedly scanned by the line-of-sight cone at each time point. For the first The number of unique new grids swept by the view cone at each point in time. This is the penalty coefficient for visual search. To prevent extremely small positive constants with a denominator of zero.

[0025] Specifically, based on the search and rescue behavior spatiotemporal data stream generated in the previous step, the system first parses the metadata of the current virtual training task to determine the task objective, such as "searching and rescuing person number 2 trapped in the storage room on the third floor." Based on this objective, it then selects a unique single reference rescue path from the set of reference rescue paths. This path consists of a series of polyline segments connected by three-dimensional node coordinates, serving as the benchmark for subsequent analysis. Next, the system traverses each record in the search and rescue behavior spatiotemporal data stream, analyzing each record at each time point. Extract the firefighter's current three-dimensional coordinates. In order to calculate the deviation of this point from the reference path, the system will By comparing with each line segment on the reference path, vector geometric operations are used to find... For each line segment or its extension, find the perpendicular projection point and calculate the distance between them. After traversing all line segments, take the one with the smallest distance as the shortest geometric distance from the current position to the path, and its corresponding projection point is the shortest path. Meanwhile, the system utilizes a pre-defined, finely divided 3D grid within the virtual environment (e.g., cell sizes of 0.5m × 0.5m × 0.5m) to quantify firefighters' visual search behavior at specific time points. The system is based on the firefighter's current location. Using head pose quaternions, a vision cone representing the field of view is constructed, and the set of all 3D mesh cells covered by this cone is calculated. The system maintains a continuously updated global set of "explored meshes" throughout the training process. By comparing the set of meshes covered by the current vision cone with this global set, the number of newly scanned meshes can be identified. and the number of grids that were scanned repeatedly The newly scanned grid cells are added to the global set after being counted. Finally, for each time point, the system will calculate the shortest geometric distance and the number of newly scanned grid cells. Number of grids scanned repeatedly These three values ​​are stored as a data tuple. By combining the data tuples from all time points, the instantaneous deviation characteristic parameters of all time points can be extracted.

[0026] formula: The advantage of this formula lies in its fusion evaluation of two dimensions: physical path deviation and visual search efficiency. Traditional path deviation metrics only focus on geometric distance and cannot distinguish between the fundamentally different behaviors of "deviation for effective reconnaissance" and "deviation due to disorientation." This is addressed by introducing an exponential penalty term controlled by visual search behavior. This enables the quantification of the cognitive state behind deviant behavior, when firefighters repeatedly scan known areas ( (High) without acquiring new information ( When the value is low, the exponential term increases sharply, thus amplifying the penalty for physical deviation. This corresponds to the ineffective wandering behavior of firefighters in a fire scene due to tension and disorientation. Conversely, if the deviation from the path is accompanied by exploration of unknown areas ( If the distance is high, the penalty term will be very small, and the metric will focus more on pure geometric distance. This design makes... It measures not only how far we've gone astray, but also how confused we've been. The total number of time points represents the total number of data frames in the spatiotemporal data stream of the search and rescue activities covered in this analysis. It is determined by the total duration of the analysis segment and the sampling frequency of the data stream. This value is directly derived from the input data and does not require pre-setting. For example, if analyzing a 60-second search and rescue process, and the sampling frequency of the search and rescue spatiotemporal data stream is 10 times per second (10Hz), then... The value is calculated The result is 600, meaning a total of 600 time points will be used for calculation. In this example, a 20-second segment is analyzed with a sampling frequency of 10Hz. .

[0027] For the first The three-dimensional coordinate vector of the firefighter at the specified time point is directly read from the spatiotemporal data stream of the search and rescue operation, representing the coordinates of the firefighter at the specified time point. At each sampling moment, the position of the firefighter wearing a head-mounted display device in the world coordinate system of the virtual building scene is measured in meters. For example, the 5th sampling point at the 10th second of the search and rescue mission (i.e., The system reads from the data stream that the firefighter's position at that moment is a point in the corridor, with coordinates as follows: This data is the raw output of the virtual reality tracking system and is the foundation for all subsequent spatial calculations. It represents a specific point in time in this example. The recorded coordinates of the firefighters are .

[0028] Current location of firefighters In the reference path The coordinate vector of the projection point on the surface is obtained through geometric calculation. Specifically, the process involves projecting the firefighter's current position point... To reference path Project perpendicularly onto all the constituent line segments (consisting of a series of line segments), and calculate the distance from each projection point to the target line segment. The distance is calculated, and the projection point with the shortest distance is selected as the final projection point. This calculation process is completed in the step of extracting instantaneous deviation feature parameters, and the result is directly called. For example, if the reference path near the firefighter is a straight line along the Y-axis, its coordinates are... So, regarding the position of firefighters The coordinates of its nearest projection point on this path are In this example, this calculation result is used, that is .

[0029] For the first The number of grids repeatedly scanned by the view cone at a given time point quantifies how much of the firefighter's visual search behavior involves re-observing areas already seen. It is determined by comparing the virtual environment grid currently covered by the view cone with a global set recording all historically observed grids; the value is the number of elements in the intersection. A higher value indicates a higher probability of overlap. A value typically indicates that a firefighter may be loitering or confirming direction. For example, when a firefighter scans back and forth in a room that has already been searched, most of the grid within their field of vision is already present in the global set, and this value may be recorded. In this example, a scenario is set up where a firefighter is slightly disoriented and repeatedly observes a familiar intersection. The time recorded is... .

[0030] For the first The number of unique new grid cells swept by the view cone at each time point, this value is related to Correspondingly, this quantifies how much new visual-spatial information a firefighter acquires at any given moment. The value is the number of grids within the current visual cone that do not belong to the global set of previously observed grids. This applies when a firefighter enters a completely new room. The value will be very high, for example Conversely, if it stares at a wall, then The value could be 0. In this example scenario, while repeatedly observing the intersection, the firefighter's line of sight caught a previously unnoticed gap, thus acquiring a small amount of new information, which was recorded as 0. .

[0031] The visual search penalty coefficient is a dimensionless hyperparameter used to adjust the sensitivity of the path deviation metric to invalid visual search behavior. The coefficient is determined based on the analysis of a large amount of real training data and expert evaluation. The process involves collecting virtual training data from at least 100 firefighters, with 3-5 senior fire instructors subjectively scoring each firefighter's path selection and search efficiency (e.g., 1-10 points). Then, a suitable path selection method is found using techniques such as grid search or gradient optimization. The value that makes the calculated The strongest negative correlation with expert ratings (i.e.) (closest to -1.0), after this calibration process, it was found that when When set to 0.8, the model's evaluation results best match the expert judgment; therefore, this value is permanently used in this system. In this example, .

[0032] To prevent extremely small positive constants with denominators of zero, this is an engineering parameter used to ensure computational stability. Its value is set to a number much less than 1, typically a small positive value that can be represented by a single-precision or double-precision floating-point number, for example... The principle for setting this value is that it must be small enough that when When the integer is a positive integer greater than or equal to 1, it affects the fractional value. The impact is negligible, but it can ensure that... In extreme cases, the denominator is not zero, thus avoiding floating-point division by zero errors during calculations. In this example, we take... .

[0033] Calculation process: Using the parameters given in the example, calculate at a single time point The instantaneous deviation penalty value, this process will be in the entire The process is repeated at each time point, and the average value is calculated. : First, calculate the Euclidean distance from the current position to the path. : ; ; ; Next, calculate the exponential part of the exponential penalty term. : ; Then calculate the complete exponential penalty term. : ; Finally, calculate this time point. Instantaneous penalty value: ; This process will affect from arrive Execution at each time point yields a sequence containing 200 instantaneous values, for example... The final search and rescue route deviated from the measurement It is the average value of this sequence.

[0034] This result indicates that at time point Although the firefighter physically deviated from the planned route by only about half a meter, their visual search behavior exhibited an extremely high repetition rate and a very low rate of acquiring new information. This resulted in their instantaneous penalty value being amplified to 683.11, an extremely high value far exceeding their physical deviation distance. This reveals that the firefighter may have been in a state of high confusion or ineffective search at that moment, ultimately leading to the calculated... The value is a comprehensive evaluation indicator. The higher the value, the worse the firefighters' path selection and search strategies were throughout the search and rescue process. For example, if the final value is lower... If the score exceeds a preset threshold, such as 5.0, the system can determine that the firefighter has serious problems with path planning ability and environmental perception efficiency in this training and needs to receive targeted review and reinforcement training.

[0035] The steps for obtaining the multipath signal attenuation value are as follows: Based on the spatiotemporal data stream of search and rescue activities, the three-dimensional coordinate positions of any two firefighters at the same current timestamp are extracted, a spatial ray path with these two positions as endpoints is established, and the structural distribution information in the building floor plan is called to determine the type and thickness of all building structures passed through the ray path one by one, generating a ray path structural distribution sequence. Based on the ray path structure distribution sequence, the electromagnetic parameters corresponding to each building material are called respectively. The thickness of each building material that the ray path passes through in sequence is multiplied with the corresponding electromagnetic parameters point by point and accumulated one by one to form a multipath signal attenuation value.

[0036] Specifically, based on the spatiotemporal data stream of search and rescue operations, the system performs independent communication link analysis for each pair of firefighters participating in virtual training at a specified timestamp. First, the system extracts the precise three-dimensional coordinates of any two firefighters, such as firefighter A and firefighter B, at the current moment from the data stream, denoted as... and Then, a straight line segment is mathematically constructed between these two three-dimensional points. This straight line segment represents the spatial ray path of the direct radio signal between the two individuals. Next, the system calls upon the 3D geometric model of the building that was loaded and parsed before the virtual training began. This model originates from the building's floor plan and contains the 3D position, shape, size, and material information of all structural elements such as walls, floors, and columns. The system employs a ray tracing algorithm in 3D space to perform intersection calculations between the established spatial ray path and each geometric component in the building model. This algorithm starts from the origin of the ray. Begin, along the direction The system moves in a directional direction, detecting whether the ray enters or exits the solid area of ​​any building component. Each time a successful intersection is detected, the system records the three-dimensional coordinates of the ray's entry and exit points on the component's surface. Simultaneously, it reads the component's material type from its attribute data, such as "reinforced concrete" or "gypsum board partition," and calculates the actual distance the ray travels within the component based on the coordinates of the entry and exit points—that is, the effective thickness of the building structure. This process continues along the entire ray path until it reaches its endpoint. All building structures traversed by the ray path, along with their corresponding material types and thicknesses, are recorded in order from the starting point to the end point, forming an ordered list, such as [("plasterboard partition", 0.1 meters), ("air", 4.5 meters), ("reinforced concrete", 0.2 meters)]. This list is the ray path structure distribution sequence.

[0037] Based on the ray path structure distribution sequence generated in the previous step, the system queries a pre-configured database of electromagnetic property parameters for building materials. This database stores the signal attenuation coefficient values ​​of different building materials in specific radio frequency bands (such as the 400MHz band commonly used in fire protection). These coefficient values ​​are standard values ​​determined according to radio wave propagation data reports published by the International Telecommunication Union (ITU) or through actual physical measurement experiments. For example, the database may define the attenuation coefficient of "reinforced concrete" as 3.0 dB per meter, "gypsum board partition" as 0.5 dB per meter, and "glass" as... The attenuation coefficient of the "curtain wall" is 0.2 dB per meter, while that of "air" is 0 dB per meter. The system iterates through each element in the ray path structure distribution sequence. For each item in the sequence, such as "reinforced concrete", 0.2 meters, the system first extracts the material type "reinforced concrete" and its thickness of 0.2 meters. Then, it calls the parameter library to find the attenuation coefficient of 3.0 dB / meter corresponding to "reinforced concrete". It then multiplies the thickness value with the attenuation coefficient value to calculate the attenuation of the signal passing through this 0.2-meter-thick reinforced concrete wall. The system performs the same calculation for all non-air medium materials traversed in the sequence. For example, for ("plasterboard partition", 0.1 m), the calculated attenuation is... Finally, the calculated attenuation values ​​of all individual structures are summed to obtain a total penetration loss value. This sum is the basic multipath signal attenuation value without considering other complex propagation effects.

[0038] The steps to obtain the real-time communication quality index are as follows: Based on the multipath signal attenuation value, the attenuation caused by the ray path passing through the wall structure and the attenuation caused by the scattering interference caused by the steel structure are calculated separately. The attenuation caused by the ray path passing through the wall structure and the attenuation caused by the scattering interference are superimposed on the multipath signal attenuation value to obtain the total attenuation of the radio communication signal between firefighters. The real-time communication quality index is then calculated based on the total attenuation.

[0039] Specifically, based on the multipath signal attenuation value calculated in the previous step, the system further introduces a more refined attenuation model to simulate the complexity of the real environment. First, it calculates occlusion attenuation, which mainly simulates the shadow effect generated by large obstacles. The system analyzes the ray path structure distribution sequence to identify the "main structures" it traverses. The criterion is any wall or floor slab with a thickness exceeding 0.15 meters. For example, when a ray passes through a 0.2-meter-thick reinforced concrete load-bearing wall, in addition to its own 0.6 dB penetration loss, the system will also add a fixed occlusion attenuation penalty value, set to 4 dB. This value is derived from statistical analysis of a large amount of signal propagation experimental data within buildings, representing a typical average occlusion loss. Next, the system calculates scattering interference attenuation, which is specifically designed for... To address the scattering effect of radio signals caused by the steel mesh within reinforced concrete structures, the system checks for the presence of "reinforced concrete" type materials in the structural distribution sequence of the ray path. If present, a fixed scattering interference attenuation value of 7 dB is added for each traversal of such a structure. This value is an empirical constant derived from an experimental physical model calibrated based on the propagation of 400MHz signals in reinforced concrete. Subsequently, the basic multipath signal attenuation value, all identified obstruction attenuation values, and all scattering interference attenuation values ​​are summed to obtain a total attenuation value. Finally, the system calculates a real-time communication quality index ranging from 0 to 100 based on this total attenuation value. The calculation process follows a preset mapping function that defines a non-linear relationship between communication quality and total signal attenuation, specifically: ; The 10 dB and 90 dB thresholds are set based on the receiver sensitivity of the handheld walkie-talkie. A total attenuation below 10 dB is considered ideal communication with a signal quality score of 100, while a total attenuation above 90 dB is considered a complete loss of signal, making it impossible to establish effective communication with a quality score of 0.

[0040] The steps for obtaining the hysteresis of critical decision points are as follows: Based on the spatiotemporal data stream of search and rescue behavior, the path decision point segments in which the firefighters remain stationary are extracted from all three-dimensional coordinate position sequences. Within each path decision point segment, the stationary state of the firefighter's coordinates in three-dimensional space is analyzed second by second. The duration of the time period in which no coordinate changes occur is counted. At the same time, the continuous change sequence of all head posture quaternions within the time period is extracted to form a path decision point stay feature set. Based on the set of dwell characteristics at path decision points, the hysteresis of key decision points is calculated using the following formula: ; in, The latency at critical decision points (unit: seconds). This represents the total number of path decision points. For the first The duration of stationary stay within each path decision point segment (in seconds). For the first The cumulative head rotation angular displacement (in radians) within the path decision point segment. This represents the average angular velocity over that period of time (unit: radians / second). Angular velocity adjustment factor (unit: seconds / radians) used to control the sensitivity of search behavior.

[0041] Specifically, based on the spatiotemporal data stream of search and rescue activities, the system first calls predefined geometric data of path decision points. This data marks key locations in the building model, such as corridor intersections, stairwells, and doorways of rooms with multiple exits, as path decision points, and sets a spherical trigger area with a radius of 1.5 meters for each point. The system then scans the three-dimensional coordinate position sequence in the spatiotemporal data stream of search and rescue activities point by point to determine whether the firefighter's real-time coordinates have entered the trigger area of ​​any path decision point. Once entered, the system begins to monitor the firefighter's movement. The criterion for determining a stationary state is that the total displacement of the firefighter's three-dimensional coordinates is less than 0.1 meters within a continuous 1-second time window. This threshold is calibrated based on the natural sway range of the firefighter's body when performing static actions such as observation and communication. The system detects that when a firefighter remains stationary for more than 2 seconds within a certain path decision point area, this period is marked as a valid "stationary stay segment," and its start and end timestamps are accurately recorded. The duration of the stationary stay in this segment is then calculated. For each identified stationary stay segment, the system extracts all head posture quaternions recorded within that time period from the search and rescue behavior spatiotemporal data stream, forming a continuous change sequence sorted by timestamp. Finally, all relevant information for each stationary stay event, including its path decision point ID, stationary stay duration, and the corresponding continuous change sequence of head posture quaternions, is packaged into a structured data object. All these objects together constitute the path decision point stay feature set.

[0042] formula: The advantage of this formula lies in introducing an amplification factor adjusted by head rotation behavior, which dynamically penalizes the firefighter's dwell time at the decision point. This allows for the differentiation between effective dwelling and ineffective hesitation. Traditional dwell time statistics cannot distinguish whether firefighters are conducting necessary observations and communications or engaging in meaningless pacing due to disorientation or psychological stress. The hyperbolic tangent function is used to address this issue. The characteristics of this will result in the average head rotation angular velocity during the dwell period. This is converted into a modulating factor between 0 and 1. When a firefighter remains still but experiences intense head movement (high angular velocity), it means they may be anxiously looking around for clues, exhibiting obvious hesitation. When the function value approaches 1, the hysteresis contribution of that pause is close to twice its actual duration. Conversely, if a firefighter pauses with their head essentially still (low angular velocity), they may be communicating with command center or taking a short rest. When the function value approaches 0, the hysteresis contribution is approximately equal to the actual dwell time. This design makes... It is not just a measure of time, but also a quantitative assessment of cognitive load and decision-making efficiency during the stay, identifying those difficult moments of decision-making that are masked by high-intensity search behavior; The total number of path decision points represents the total number of times firefighters, identified by the system, effectively remained stationary at path decision points throughout the entire search and rescue mission. This value is obtained by processing the path decision point dwelling feature set generated in the previous stage and counting the independent dwelling events within it. This value reflects the frequency with which firefighters need to make path selection decisions and dwell for these decisions during the mission, and is a fundamental indicator for measuring mission complexity and firefighter decision-making load. It is dynamically generated based on the actual behavior of firefighters. For example, in a complete search and rescue training session, if the system identifies 3 stationary dwelling events that meet the criteria, then... The value of is 3.

[0043] For the first The duration of stationary stay within each path decision point segment, measured in seconds, quantifies the firefighter's position in the [number of]th [location / section]. The time spent in each identified decision-point dwelling event is directly derived from the duration of each event recorded in the path decision-point dwelling feature set. This set is calculated in the previous step by analyzing the firefighter's three-dimensional coordinate position sequence and based on a preset static judgment threshold (e.g., displacement less than 0.1 meters for more than 2 consecutive seconds). This parameter is the basis for calculating hysteresis and directly reflects the firefighter's physical dwell time at the decision point. For example, through analysis, the durations of the three dwelling events were found to be: 4.5 seconds for the first dwelling, 3.0 seconds for the second, and 6.2 seconds for the third. Therefore... , , .

[0044] For the first The cumulative head rotation angular displacement within each path decision point segment, in radians, is used to quantify the firefighter's position in the [number of] path decision points. The intensity of head search activity during the second static dwell period is calculated as follows: First, the intensity of head search activity during the second static dwell period is extracted from the path decision point dwell feature set. The head pose quaternion sequence corresponding to each event For every two consecutive quaternions in the sequence and Calculate the relative rotation quaternions between them. ,in yes The conjugate, then, from Extract its scalar portion Calculate the angle of this tiny rotation. Finally, by summing up all the calculated minute rotation angles during the entire dwell time, the total angular displacement is obtained. For example, in the three stops mentioned above, the firefighters actively observed and calculated during the first stop. The curve; during the second call with teammates, the head remained almost completely still. The arc, for the third time, caused him to anxiously scan the area repeatedly due to the limited visibility caused by the thick smoke. radian.

[0045] An angular velocity adjustment coefficient, measured in seconds per radian, is used to control the sensitivity of search behavior. This is a hyperparameter used to adjust the weight of the average head angular velocity in hysteresis calculation. Its value is set based on an expert calibration method. The specific process is as follows: 50 virtual training videos of firefighters containing varying degrees of hesitation are selected. Five senior fire training instructors are invited to independently score the degree of hesitation of the firefighters at the decision point in each video (1-10 points, 10 points being extreme hesitation). The average score is taken as the "expert hesitation score" for that segment. Then, an optimization loop is implemented through programming, traversing the range from 0.1 to 2.0 with a step size of 0.05. The possible values ​​for each The value is used to calculate the hysteresis at the critical decision points of all 50 segments. And calculate the group The Pearson correlation coefficient between the expert's hesitant subsequence and the overall sequence was selected, and ultimately, the one that resulted in the largest absolute value of the correlation coefficient was chosen. The value is used as the final configuration parameter of the system and is determined through this process. At that time, the formula calculation results showed the highest degree of fit with the expert judgment.

[0046] Calculation process: According to the calculation example, , The parameters for the three stops are as follows: , , The calculation process is as follows: First, calculate the hysteresis contribution value for each pause: For the first stay ( ): ; ; For the second stay ( ): ; ; For the third stay ( ): ; ; Then, calculate the total average hysteresis. : ; ; The results indicate that the firefighter's average critical decision-making delay during this mission was 6.78 seconds, a value that is not a simple average of his physical dwell time. It is not a single second, but rather an "effective hesitation time" weighted by the intensity of top search behavior during the decision-making process, a higher... Values ​​exceeding, for example, those set in advance through statistical analysis of a large number of trainees' data (such as 5.0 seconds, representing the 80th percentile level), strongly suggest that the firefighter has significant lag in route decision-making, and may have problems such as difficulty in identifying directions, indecisive tactical decisions, or psychological tension.

[0047] The steps for obtaining an individual search and rescue capability diagnostic profile are as follows: Based on the lag at key decision points, the search and rescue path deviation metric and the real-time communication quality index within the same time window are jointly analyzed to determine whether firefighters hesitate at path decision points, thus obtaining an individual search and rescue capability diagnostic profile.

[0048] Specifically, based on the calculated critical decision point delay, the system further integrates the search and rescue path deviation metric and the real-time communication quality index within the same time window as the decision point obtained in previous steps. It uses a preset set of logical rules to comprehensively determine whether a firefighter exhibits hesitation and analyze its causes. This rule set is built upon a knowledge base of fire and rescue experts. The specific judgment logic is as follows: if the critical decision point delay exceeds a high-risk threshold, such as 6.0 seconds (this threshold is set as the 85th percentile of this indicator for all trainees), and the search and rescue path deviation metric is also higher than its corresponding high-risk threshold, such as 5.0 (also set based on historical data distribution), the system determines it as "disorientation-type hesitation" and records in the diagnostic profile that the firefighter has shortcomings in spatial perception and autonomous navigation capabilities. If the critical decision point delay is very high, but the search and rescue path deviation metric is high... If the quantity is within the normal range, the system will check the real-time communication quality index for the same time window. If the index is below 40 (indicating severe communication obstruction), it will be judged as "communication-dependent hesitation." The profile will indicate that the firefighter lacks independent decision-making ability when unable to effectively obtain team instructions or information. If the communication quality index is still good (e.g., above 70) even with high latency and low path deviation, it will be judged as "decision-difficulty hesitation." The profile will describe the possible cognitive bottleneck in tactical selection or risk assessment. By summarizing the analysis results of all decision points throughout the mission, the system will finally generate a structured text report, namely, an individual search and rescue capability diagnostic profile. This profile not only gives an overall performance rating but also lists the time and location of specific hesitation events, along with possible causes and corresponding capability shortcomings analyzed based on the above rules.

Claims

1. A virtual training system for firefighters' search and rescue escape, characterized in that, The system includes: The virtual scene dynamic construction module abstracts the intersections of rooms and corridors into nodes based on the building floor plan, and the doors and passages into edges with geometric distance weights. In virtual training, it responds to fire spread or structural collapse events and adjusts the weights of the corresponding edges to infinity, obtains the shortest path, and establishes a set of reference rescue paths. The search and rescue behavior sequence quantization module records the firefighter's three-dimensional coordinate sequence, head posture quaternion sequence, and interaction events to form a search and rescue behavior spatiotemporal data stream. Based on the search and rescue behavior spatiotemporal data stream, a search and rescue path deviation metric is generated. The team communication link simulation module extracts the real-time three-dimensional coordinates of any two firefighters based on the spatiotemporal data stream of the search and rescue behavior, obtains the multipath signal attenuation value, and generates a real-time communication quality index by superimposing the interference caused by wall obstruction and steel structure scattering based on the multipath signal attenuation value. The search and rescue effectiveness comprehensive evaluation module extracts the static duration of firefighters' position coordinates at path decision points based on the spatiotemporal data stream of the search and rescue behavior, obtains the lag degree of key decision points, and makes a judgment based on the lag degree of key decision points, combined with the search and rescue path deviation metric and the real-time communication quality index of the same time window, and constructs an individual search and rescue capability diagnostic profile.

2. The firefighter search and rescue escape virtual training system according to claim 1, characterized in that, The steps for obtaining the reference rescue path set are as follows: Based on the building floor plan, the location information of the intersection of rooms and corridors and the spatial connection information of doors and passages are extracted from the building floor plan. The locations of all room and corridor intersections are mapped as abstract nodes, and the connection relationships of all doors and passages are mapped as edges between nodes, thus obtaining the initial node-edge topology network. Based on the initial node edge topology network, the actual length information of each door and passage in the building plan is called, and the actual length of each door and passage is used as the geometric distance weight of the edge. The geometric distance weight values ​​of all edges in the initial node edge topology network are updated to generate a node edge topology network with geometric distance weights. Based on the node-edge topology network with assigned geometric distance weights, the location of fire spread and structural collapse events in the virtual training scenario is monitored in real time. Doors and passages that are impassable due to fire spread or structural collapse are identified. The geometric distance weights of the edges represented by the corresponding doors and passages are updated to infinity. The shortest path from the entrance node to all trapped personnel location nodes and safety exit nodes is calculated, and a set of reference rescue paths is generated.

3. The firefighter search and rescue escape virtual training system according to claim 1, characterized in that, The steps for acquiring the spatiotemporal data stream of the search and rescue operation are as follows: Record the firefighters' three-dimensional coordinate positions, head posture quaternions, and timestamps of each interaction event throughout the entire virtual training process. Align and synchronize the three-dimensional coordinate position sequence, head posture quaternion sequence, and interaction event markers with timestamps as indexes to generate a spatiotemporal data stream of search and rescue behavior.

4. The firefighter search and rescue escape virtual training system according to claim 1, characterized in that, The steps for obtaining the search and rescue path deviation metric are as follows: Based on the three-dimensional coordinate position sequence in the spatiotemporal data stream of the search and rescue behavior, a single reference rescue path that matches the current mission scenario is selected from the set of reference rescue paths as a comparison benchmark. The current position of the firefighter at each time point is vertically projected onto the selected reference rescue path using the shortest Euclidean distance to obtain the shortest geometric distance from the current position to the path. Combined with the number of new grids and the number of repeated grids swept by the line of sight cone at each time point, the instantaneous deviation feature parameters of all time points are extracted. Based on the instantaneous deviation characteristic parameters, the search and rescue path deviation metric is calculated.

5. The firefighter search and rescue escape virtual training system according to claim 1, characterized in that, The steps for obtaining the multipath signal attenuation value are as follows: Based on the spatiotemporal data stream of the search and rescue behavior, the three-dimensional coordinate positions of any two firefighters at the same current timestamp are extracted, a spatial ray path is established with these two positions as endpoints, and the structural distribution information in the building plan is called to determine the type and thickness of all building structures passed through the ray path one by one, generating a ray path structural distribution sequence. Based on the ray path structure distribution sequence, the electromagnetic parameters corresponding to each building material are called respectively. The thickness of each building material that the ray path passes through in sequence is multiplied with the corresponding electromagnetic parameters point by point and accumulated one by one to form a multipath signal attenuation value.

6. The firefighter search and rescue escape virtual training system according to claim 1, characterized in that, The steps for obtaining the real-time communication quality index are as follows: Based on the multipath signal attenuation value, the attenuation caused by the ray path passing through the wall structure and the attenuation caused by the scattering interference caused by the steel structure are calculated respectively. The attenuation caused by the ray path passing through the wall structure and the attenuation caused by the scattering interference are superimposed on the multipath signal attenuation value to obtain the total attenuation of the radio communication signal between firefighters. The real-time communication quality index is then calculated based on the total attenuation.

7. The firefighter search and rescue escape virtual training system according to claim 1, characterized in that, The steps for obtaining the hysteresis of the key decision point are as follows: Based on the spatiotemporal data stream of the search and rescue behavior, the path decision point segments in which the firefighters remain stationary are extracted from all three-dimensional coordinate position sequences. Within each path decision point segment, the stationary state of the firefighter's coordinates in three-dimensional space is analyzed second by second. The duration of the time period in which no coordinate changes occur is counted. At the same time, the continuous change sequence of all head posture quaternions within the time period is extracted to form a path decision point stay feature set. Calculate the lag degree of key decision points based on the set of path decision point dwell characteristics.

8. The firefighter search and rescue escape virtual training system according to claim 1, characterized in that, The steps for obtaining the individual search and rescue capability diagnostic profile are as follows: Based on the lag at the key decision points, the search and rescue path deviation metric and the real-time communication quality index within the same time window are jointly analyzed to determine whether firefighters exhibit hesitation at path decision points, thereby obtaining an individual search and rescue capability diagnostic profile.

9. The firefighter search and rescue escape virtual training method of the firefighter search and rescue virtual training system according to any one of claims 1-8, characterized in that, Includes the following steps: Based on the building floor plan, the intersections of rooms and corridors are abstracted as nodes, and the doors and passages are abstracted as edges with geometric distance weights. In virtual training, in response to fire spread or structural collapse events, the weights of the corresponding edges are adjusted to infinity, and the shortest path is obtained to establish a set of reference rescue paths. Record the firefighter's three-dimensional coordinate sequence, head posture quaternion sequence and interaction events to form a search and rescue behavior spatiotemporal data stream, and generate a search and rescue path deviation metric based on the search and rescue behavior spatiotemporal data stream; Based on the spatiotemporal data stream of the search and rescue behavior, the real-time three-dimensional coordinates of any two firefighters are extracted, and the multipath signal attenuation value is obtained. Based on the multipath signal attenuation value, the interference caused by wall obstruction and steel structure scattering is superimposed to generate a real-time communication quality index. Based on the spatiotemporal data stream of the search and rescue behavior, the static duration of the firefighter's position coordinates is extracted at the path decision point to obtain the hysteresis of the key decision point. Based on the hysteresis of the key decision point, combined with the search and rescue path deviation metric and the real-time communication quality index of the same time window, a diagnostic profile of individual search and rescue capabilities is constructed.

Images