Dynamic travel reconstruction and guidance method based on degree of congestion and walking safety

Abstract

The invention relates to the field of path planning, and discloses a crowding degree and walking safety-based travel dynamic reconstruction and guidance method, which comprises the following steps of: acquiring pedestrian density, speed vector, motion direction and retention unit in a target area in real time; recording density, speed, direction and stay change of each space unit in a continuous time window based on the space state matrix, and establishing a space-time crowd state database in combination with a historical flow track and a scene structure; based on the space-time crowd state database and the basic behavior characteristic mode library, the panic index, the crowd intensity and the reverse migration tendency of each space unit are calculated; executing path solving operation according to the cost value system, and outputting an evacuation path set and a rescue priority channel set; continuously acquiring density, speed, direction and psychological state indexes of each path unit in a personnel moving process, and generating a corresponding path adjustment instruction according to a threshold calculation result; the method has the advantages of improving the emergency evacuation efficiency and the rescue response speed.

Description

Technical Field

[0001] This invention relates to the field of route planning, specifically to a method for dynamic route reconstruction and guidance based on congestion and pedestrian safety. Background Technology

[0002] In high-density public areas such as large transportation hubs, commercial complexes, and sports stadiums, emergencies such as fires, stampedes, explosions, and emergency medical needs occur frequently. After an incident, crowd density changes rapidly, and local congestion spreads to surrounding areas. Traditional methods relying on fixed signage and pre-set evacuation routes are ineffective in dealing with dynamic situations, easily causing traffic bottlenecks, evacuation delays, and even secondary safety risks. Furthermore, existing systems often only divert and evacuate ordinary pedestrians, insufficiently guaranteeing access for emergency rescue forces. They lack a coordination mechanism that simultaneously considers rapid evacuation and priority rescue in emergency scenarios. Existing route planning is mostly based on the shortest distance or static traffic capacity, failing to fully consider changes in congestion levels, pedestrian safety risks, and crowd flow trends, thus unable to respond to emergencies in real time. In addition, current evacuation areas are mostly fixed, making it difficult to dynamically adjust according to the accident location, on-site capacity changes, and crowd behavior characteristics, thus limiting emergency response efficiency and safety. Therefore, it is essential to design a dynamic route reconstruction and guidance method based on congestion levels and pedestrian safety to improve emergency evacuation efficiency and rescue response speed. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention provides a method for dynamic reconstruction and guidance of travel based on congestion and pedestrian safety, which has the advantages of improving emergency evacuation efficiency and rescue response speed, and solves the problems mentioned in the background technology.

[0004] To achieve the aforementioned goals of improving emergency evacuation efficiency and rescue response speed, this invention provides the following technical solution: a method for dynamic reconstruction and guidance of travel based on congestion and pedestrian safety, comprising the following steps: Preferably, the process of constructing the spatial state matrix is ​​as follows: Based on the target area structure, obtain the channel network, node locations, obstacle distribution, and exit connectivity. Cameras, sensors, and positioning modules are deployed within the area to obtain pedestrian location coordinates, movement trajectories, and hotspots where they stop; The collected pedestrian motion features are mapped to a spatial grid structure and assigned to corresponding nodes and connected edges to generate a spatial state matrix containing spatial indexes, node attributes, and flow parameters.

[0005] Preferably, the process of establishing a spatiotemporal population status database by combining historical movement trajectories and scene structure is as follows: Based on the spatial state matrix, the real-time collected density, velocity, direction and residence data are continuously stored in time series and combined with spatial unit numbering to form regional time grids. Trajectory reconstruction and event labeling of pedestrian movement and dwelling trajectories within a continuous time window; Based on the regional topology, the trajectory is mapped to the corresponding access unit, boundary unit, and event source node; By introducing historical scene operation data, including the distribution of crowd flow, entrance opening strategies and regional functional attributes at different times, real-time trajectories are matched and integrated with historical scene structure information to establish a multi-dimensional spatiotemporal crowd status database that includes spatial location index, time series labels, behavioral event identifiers and environmental conditions.

[0006] Preferably, the process of extracting behavioral feature vectors to generate a basic behavioral feature pattern library is as follows: In a multidimensional spatiotemporal population status database, density changes, velocity changes, directional shifts, and dwell time series data within a continuous time window are extracted, and corresponding individual and group behavior segments are located based on spatial unit indexes and time labels. Behavioral patterns are decomposed into individual trajectories and local state fluctuations to identify typical micro-behavioral features such as acceleration, sudden stop, aggregation and stagnation, deflection and reverse movement, and encoded in vector form. Based on spatial topology and node relationships, the group behavior is analyzed in a structured manner to extract macro-level group behavior variables; Multi-scale clustering analysis was performed on the extracted individual and group behavioral feature vectors. The clustering results were combined with scene area attributes, time period classification, and event labels to construct a basic behavioral feature pattern library.

[0007] The preferred process for introducing group psychodynamic mechanisms is as follows: Based on the basic behavioral feature pattern library, a mapping model between behavioral features and psychological parameters is established, including the correspondence between behavioral acceleration, density gradient change rate, path deviation fluctuation amplitude, conformity rate, and reverse escape tendency. We perform emotional feature enhancement processing on continuous time slice behavioral data, and infer the group stress gradient and the direction of anxiety propagation through local density mutation detection, velocity anomaly clustering identification and directional perturbation analysis. Psychological parameters are calculated from individual and group behavioral inputs to obtain psychodynamic indicators.

[0008] Preferably, the process of forming the psychological behavior weight matrix is ​​as follows: Based on psychodynamic indicators, the panic index, herd mentality intensity, and reverse migration impulse of each spatial unit are weighted and integrated with the node access attributes. The weights of the merged nodes are adjusted locally and globally. A dynamic correction mechanism is constructed by combining the neighborhood relationship between nodes, path connectivity and event level information, and the weights of psychological behavior are iteratively updated within a continuous time window. For high-density congested areas or potentially dangerous nodes, a weighted attenuation and flow direction correction strategy is introduced to spatially differentiate the psychological behavior weight matrix, thus forming a psychological behavior weight matrix.

[0009] The preferred process for forming a value system that distinguishes between evacuation and rescue routes is as follows: Based on the psychological behavior weight matrix, the velocity gradient, local conflict probability, psychological behavior weight and reachability parameters are processed in a unified manner. Calculate the weights for population dispersion, exit accessibility, and local congestion sensitivity for evacuation routes; Calculate the accessibility level, blockage risk, interference index, and time sensitivity weight for rescue routes; A dual-purpose path value system is generated by using a joint evaluation method at the node and path levels.

[0010] The preferred method, according to the value system, is to solve the computational process as follows: Based on the value system, graph search, shortest path or multi-objective optimization algorithms are used to solve the path; In the solution process, path continuity, node turning complexity, regional accessibility, and real-time psychological behavior weights are incorporated into the constraints. Output the set of evacuation routes and the set of rescue routes, and generate a multi-level route candidate set.

[0011] Preferably, the process of forming the dual-path boot configuration table is as follows: The multi-level path candidate set is organized according to the node index and path sequence structure to generate path guidance configuration information; Index and label each node in the path set, and calculate the connectivity between nodes, the turning complexity, and the local traffic load. The path level weights and priorities are dynamically adjusted based on real-time psychological state factors, density gradients, and flow trends. During the dynamic adjustment process, an updated path priority sequence is generated, and the level changes, applicable time windows, and key node identifiers of each path are recorded to form a dual-path guidance configuration table.

[0012] Preferably, the process of generating the corresponding path adjustment instruction based on the threshold calculation result is as follows: The density, speed, direction, and psychological state parameters of each path unit are continuously collected and compared in real time with the thresholds in the dual-path guidance configuration table. The path adjustment range and scope are calculated based on the deviation coefficient, node congestion level, and path priority. The calculation results are then parsed into an executable set of path adjustment instructions.

[0013] Compared with existing technologies, this invention provides a method for dynamic reconstruction and guidance of travel routes based on congestion and pedestrian safety, which has the following beneficial effects: This invention, by comprehensively considering pedestrian density distribution, local flow velocity gradient, spatial topological resistance, and psychological and behavioral factors, can achieve intelligent avoidance and path reconstruction in situations of high congestion, sudden events, or spatial blockage, thereby effectively reducing the probability of congestion. By establishing a spatiotemporal crowd state database and a basic behavioral characteristic pattern library, the system can learn and predict crowd migration trends and behavioral patterns in different scenarios, thus supporting adaptive path updates and dynamic guidance optimization, improving the accuracy and response speed of path adjustments. By introducing a psychodynamic weight matrix to quantitatively model crowd panic index, herd mentality, and reverse migration tendency, path planning is not only based on physical spatial constraints but also reflects the psychological response characteristics of the crowd, thereby achieving hierarchical guidance of evacuation and rescue paths, balancing safety and efficiency. Through a dual-path guidance configuration table and a path change trigger threshold mechanism, flow direction deviation and safety risks can be detected in real time during personnel movement, enabling adaptive path correction and improving the robustness and continuity of the guidance system. It can achieve intelligent perception, risk prediction, and safe guidance of crowd flow in complex public spaces, significantly improving emergency evacuation efficiency and pedestrian safety, and has broad practical application value and social benefits. Attached Figure Description

[0014] Figure 1 This is a schematic diagram of the method of the present invention. Detailed Implementation

[0015] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0016] Example 1: Please refer to Figure 1 As shown in the embodiment of the present invention, a method for dynamic reconstruction and guidance of a journey based on congestion and pedestrian safety includes: S1: Real-time data collection of pedestrian density, velocity vector, movement direction, and lingering units within the target area, and loading of exit nodes, obstacle nodes, and event source nodes based on the area topology to construct a spatial state matrix.

[0017] The process of constructing the spatial state matrix in S1 is as follows: Based on the target area structure, obtain the channel network, node locations, obstacle distribution, and exit connectivity. Before actual deployment, obtain the building floor plan, floor structure diagram and area functional division diagram of the target area, and convert information such as passages, rooms, squares, stairs, escalators, entrances and exits into computable traffic network data. Number all walking passages in the area, record the passage width, length, tilt angle and bidirectional or unidirectional traffic attributes, and mark the coordinate position of nodes and the connectivity between nodes. Record obstacles in a structured way to form a basic topology data file that can be used for calculation.

[0018] Cameras, sensors, and positioning modules are deployed within the area to obtain pedestrian location coordinates, movement trajectories, and hotspots where they stop; Video cameras, millimeter-wave sensors, and wireless positioning modules are deployed within the area. The cameras are installed at the top of the passageway, near entrances and exits, and in areas where people tend to gather, with coverage areas designed to minimize obstruction. The wireless positioning modules determine the location of people carrying mobile devices by detecting changes in signal strength. The millimeter-wave sensors monitor obstructed areas and nighttime scenes. The system records each person's two-dimensional location coordinates, timestamp, direction of movement, and speed at a frequency of no less than ten times per second. It also counts the number of people staying in fixed areas and the duration of their stay, which is used as a basis for determining loitering. The collected data is written to a real-time buffer and then processed for noise filtering, duplicate trajectory deduplication, and time synchronization.

[0019] The collected pedestrian motion features are mapped to a spatial grid structure and assigned to corresponding nodes and connected edges to generate a spatial state matrix containing spatial indexes, node attributes, and flow parameters. The region is divided into equal-area grid cells, each cell corresponding to a unique index number. An association table between cells and topological nodes is established. Using the real-time coordinates of pedestrians, each pedestrian is assigned to their respective grid cell. The nodes and connecting edges crossed by their movement are determined based on their direction of movement, forming a mapping result from location to node. In each calculation cycle, the number of people, average speed, and main direction of movement in each grid are counted. These values ​​are then summarized into the corresponding access nodes and edge attributes to generate a spatial state matrix.

[0020] S2: Based on the spatial state matrix, record the density, speed, direction and dwell time changes of each spatial unit within a continuous time window. Combine historical flow trajectories and scene structure to establish a spatiotemporal population state database, and extract behavioral feature vectors to generate a basic behavioral feature pattern library.

[0021] The process of establishing a spatiotemporal population status database in S2 by combining historical flow trajectories and scene structure is as follows: Based on the spatial state matrix, the real-time collected density, velocity, direction and residence data are continuously stored in time series and combined with spatial unit numbering to form regional time grids. The system writes the density, average velocity, main direction, and dwell count of each spatial unit in the spatial state matrix at the corresponding time to the time series storage at a fixed sampling period. The sampling period is from 200 milliseconds to 1 second, with a default of once every 500 milliseconds. The written record includes fields such as timestamp, spatial unit index, current number of people, average velocity value, direction vector magnitude, number of dwellers, and cumulative dwell time. The time series records are appended in ascending order of time and stored in regional and time partitions to support fast retrieval. To reduce storage costs, a time series aggregation strategy is used for the original trajectory that exceeds the historical retention period. Aggregated indicators, including average density, maximum density, velocity variance, and dwell event count, are calculated and saved in five-minute and thirty-minute windows. The aggregation record includes start time and end time to form a regional time raster.

[0022] Trajectory reconstruction and event labeling of pedestrian movement and dwelling trajectories within a continuous time window; Trajectory preprocessing, interpolation, and filtering are performed on the original position sequence within a continuous time window to complete trajectory reconstruction. Preprocessing includes time synchronization, outlier removal, merging of duplicate device numbers, and coordinate adjustment. Interpolation uses kinematically constrained linear interpolation or Kalman filtering to fill in missing points. Filtering is used to smooth position noise and calculate instantaneous velocity and acceleration values. Subsequently, the trajectory is segmented according to a configurable window length. Commonly used window lengths are 30 seconds, 2 minutes, and 15 minutes to adapt to different analysis granularities. After trajectory segmentation, an event detection algorithm is executed to identify dwell events, reverse movement events, and velocity change events. Detected events are added to the trajectory record with an event label field and written to the event index table.

[0023] Based on the regional topology, the trajectory is mapped to the corresponding access unit, boundary unit, and event source node; The reconstructed trajectory is mapped to the passage cells and boundary cells through spatial indexing to achieve the coupling of trajectory and topology. First, a spatial index lookup structure is constructed for the regional topology. Spatial hashing or quadtree indexing of grid cells and node coordinates is used to support fast point-to-cell mapping. For each location point in the trajectory, the nearest neighbor cell lookup is performed and the mapping result is recorded. At the same time, the passage record of passing through the connected edge is generated according to the cell changes of adjacent location points. For trajectory points that pass through the regional boundary or exit location, they are marked as boundary cell events and the entrance or exit number, passage time and personnel identifier are recorded in the event source node table. The mapping result is written into the trajectory mapping table in the form of row vectors, and the edge passage count and edge passage duration are stored in the edge passage table to construct edge flow statistics.

[0024] By introducing historical scene operation data, including the distribution of crowd flow, entrance opening strategies and regional functional attributes at different times, real-time trajectories are matched and integrated with historical scene structure information to establish a multi-dimensional spatiotemporal crowd status database that includes spatial location index, time series label, behavioral event identifier and environmental conditions. Real-time trajectory mapping data is matched and integrated with historical scene operation data, and a multi-dimensional database structure is constructed accordingly. Historical scene operation data includes crowd flow records at different time periods, entrance opening and closing logs, regional function label information, and historical event records. Data files are divided into independent storage partitions according to time and region dimensions, and columnar storage is adopted to reduce access latency for concurrent queries of multiple fields. During the data entry process, scene label fields are added to each trajectory mapping record. At the same time, environmental condition fields such as lighting level, security checkpoint status, and meteorological observation values ​​are aligned with the time raster and written into the environmental record table. The database structure adopts a multi-table organization mode, and each table is accessed by association based on a unified time index and spatial coding system, forming a multi-dimensional spatiotemporal crowd status database with time index, spatial coding, historical labels, and environmental attribute structure.

[0025] The process of extracting behavioral feature vectors from S2 to generate a basic behavioral feature pattern library is as follows: In a multidimensional spatiotemporal population status database, density changes, velocity changes, directional shifts, and dwell time series data within a continuous time window are extracted, and corresponding individual and group behavior segments are located based on spatial unit indexes and time labels. From the established multidimensional spatiotemporal population status database, a specified time window is selected according to the time index, for example, every two seconds or every five seconds as a sampling period. The density value, average velocity, direction vector and dwell count sequence of the corresponding spatial unit are read. Based on the spatial unit number, these time series are associated with the individual movement records identified in the trajectory table. Continuous and complete trajectory segments are filtered out, and the scene label field of the segment is loaded at the same time. For each trajectory segment, the density change, velocity change and direction angle change of adjacent sampling points are extracted. At the same time, the duration of dwell state is recorded, and these indicators are arranged in chronological order to form a behavioral time series input structure.

[0026] Behavioral patterns are decomposed into individual trajectories and local state fluctuations to identify typical micro-behavioral features such as acceleration, sudden stop, aggregation and stagnation, deflection and reverse movement, and encoded in vector form. Motion state segmentation is performed on each individual trajectory time series. Behavioral segments are identified based on velocity difference and direction change threshold. In the specific encoding process, indicators such as acceleration magnitude, emergency stop duration, turning angle, and reverse distance are quantized into fixed-dimensional behavioral feature vectors. These vectors, along with the timestamps and spatial unit numbers of the corresponding trajectory segments, are written into the intermediate behavioral result table.

[0027] Based on spatial topology and node relationships, the group behavior is analyzed in a structured manner to extract macro-level group behavior variables; By utilizing regional spatial topology data and node connectivity, multiple trajectories located in adjacent spatial units within the same time window are aggregated. The concentration of group movement direction, local density gradient changes, and flow direction differences between neighboring nodes are statistically analyzed within each unit. Taking a channel-type area as an example, the average movement direction consistency index of the group is calculated from the entrance node to the exit node. In an open hall area, the trend of multiple directions converging towards a hot spot is statistically analyzed, and the hot spot location index is recorded. During the mapping stage, features such as the group gathering index, direction consistency ratio, and frequency of local reverse flow are written into the group behavior feature table and stored synchronously with the corresponding regional function label and event background label.

[0028] Multi-scale clustering analysis was performed on the extracted individual and group behavioral feature vectors. The clustering results were combined with scene area attributes, time period classification and event labels to construct a basic behavioral feature pattern library. The extracted and encoded individual and group behavioral feature vectors are summarized according to the sampling period, and the data of each dimension are normalized to ensure consistency of different indicator dimensions. The processed feature set is input into the clustering analysis program, and clustering operations are performed based on the time scale and the spatial scale respectively. The clustering method adopts a combination of density clustering and hierarchical clustering to identify stable behavioral clusters. After the clustering is completed, the center vector, feature distribution range, representative behavioral segment index and occurrence probability of each behavioral cluster are calculated and recorded. Then, the cluster category is associated with scene area attributes, time period classification and event label to generate behavioral pattern annotation records. The sorted behavioral pattern data is written into the behavioral pattern master table to complete the generation of the basic behavioral feature pattern library.

[0029] S3: Based on the spatiotemporal population state database and basic behavioral characteristic pattern library, a group psychological dynamic mechanism is introduced to calculate the panic index, herd mentality intensity and reverse migration tendency of each spatial unit, forming a psychological behavior weight matrix. The flow velocity gradient, local conflict probability, accessibility and psychological behavior weight are weighted and integrated to form a value system that distinguishes evacuation and rescue paths.

[0030] The process of introducing group psychodynamics in S3 is as follows: Based on the basic behavioral feature pattern library, a mapping model between behavioral features and psychological parameters is established, including the correspondence between behavioral acceleration, density gradient change rate, path deviation fluctuation amplitude, conformity rate, and reverse escape tendency. Feature dimensions for psychological inference are selected from the existing basic behavioral feature pattern library, including individual acceleration sequences, unit density gradient time series, path offset angle fluctuation amplitude, neighborhood directional consistency ratio, and reverse migration incidence rate. A mapping model is established using a combination of supervised learning and statistical regression. The supervised learning part uses historical event labeled data as training labels, and the training objectives are three continuous values: panic index, herd mentality intensity, and reverse migration tendency. Model candidates include multiple linear regression, ridge regression, tree-based regression, and gradient boosting regression. Cross-validation is used to select the optimal model. The statistical regression part performs correlation analysis and regression coefficient estimation on key features to obtain initial mapping weights. The model input format is a standardized feature vector, with the vector fields in the following order: timestamp, spatial unit number, mean acceleration, mean density gradient, directional fluctuation amplitude, neighborhood consistency value, and reverse migration rate. The output is a floating-point value of the three psychological parameters. The training process records the training set size, validation set error, and model version number. The training results are stored in the model library as model files, and the training data time range and label source are written in the model metadata table.

[0031] We perform emotional feature enhancement processing on continuous time slice behavioral data, and infer the group stress gradient and the direction of anxiety propagation through local density mutation detection, velocity anomaly clustering identification and directional perturbation analysis. For continuous time-slice data entering the psychological inference process, three enhancement steps are first performed: local density mutation detection, velocity anomaly clustering identification, and directional perturbation analysis. Local density mutation detection uses the sliding window difference method to calculate the density change rate within adjacent time windows and compare it with a threshold. An example of the density change rate threshold is that an increase rate exceeding 0.5 people per square meter per second is judged as a mutation event. Velocity anomaly clustering identification uses a density-based clustering algorithm to divide the velocity vector set, marking isolated clusters and low-frequency clusters as abnormal velocity groups, and recording the average velocity and occurrence frequency of abnormal clusters. Directional perturbation analysis calculates and normalizes the standard deviation of the direction angle within a short time window. Window with a standard deviation exceeding 30 degrees is marked as a directional perturbation. The above three detection results are added to the original behavioral feature vector as emotion enhancement features. The newly added fields are mutation event identifier, abnormal velocity intensity, and directional perturbation intensity. The enhanced data is written into a temporary emotion feature table according to the time series, with the field format being timestamp, spatial unit number, enhancement identifier set, and enhancement value.

[0032] Psychological parameters are calculated from individual and group behavioral inputs to obtain psychodynamic indicators; The psychological parameter calculation module reads input vectors from the temporary emotional feature table and the behavioral feature table, performs forward calculations according to the mapping model, and obtains the time-series values ​​of three indicators: panic index, herd mentality strength, and reverse migration tendency. The calculation process includes four sub-steps: data standardization, feature weighted summation, model forward inference, and threshold calibration. Standardization uses historical mean and standard deviation for zero-mean standardization. Feature weighted summation uses coefficients provided by the mapping model to sum the input vectors according to their weights to obtain the original score. Model forward inference inputs the original score into the trained model to obtain continuous value output. Threshold calibration performs linear scaling and boundary clipping on the output according to the calibration parameters of the deployment scenario to keep the indicators between zero and one. After the calculation is completed, the psychodynamic indicators are recorded in time series and written into the psychological parameter table. The table fields include timestamp, spatial unit number, individual or group identifier, panic index value, herd mentality strength value, reverse migration tendency value, and model version number.

[0033] The process of forming the psychological behavior weight matrix in S3 is as follows: Based on psychodynamic indicators, the panic index, herd mentality intensity, and reverse migration impulse of each spatial unit are weighted and integrated with the node access attributes. The panic index, herd mentality intensity, and reverse migration impulse values ​​corresponding to each spatial unit are read from the psychological parameter table. At the same time, node accessibility data, including accessibility capacity, passable width, node buffer area, and historical accessibility rate, are read. The above indicators are synthesized into basic psychological weight values ​​in a linear weighting manner. The weight synthesis adopts a standardized weighted summation method. First, the historical mean of each input indicator is subtracted and divided by the historical standard deviation to achieve zero mean standardization. Then, the basic weight value is calculated by combining the weight coefficients. Example weight ratios can use point calibration values, such as 0.5, 0.3, and 0.2. The synthesis result is written into the node psychological weight temporary table, including the node number, timestamp, and basic psychological weight value. This operation is performed in batches within each sampling period. The calculation process records the input vector dimension, standardization parameters, and the version of the coefficients used.

[0034] The weights of the merged nodes are adjusted locally and globally. A dynamic correction mechanism is constructed by combining the neighborhood relationship between nodes, path connectivity and event level information, and the weights of psychological behavior are iteratively updated within a continuous time window. The system performs local-global joint correction on the fused basic psychological weights. The local part uses spatial neighborhood smoothing, linearly fusing each node with its own weight using the weighted average of the weights of its neighboring nodes and the local smoothing coefficient, and normalizing the fusion result to maintain a uniform magnitude. The global part introduces connectivity score and event level factor to multiplicatively adjust the node weights. The connectivity score is derived from path reachability calculation, and the event level is mapped to discrete levels according to historical scenarios, such as levels one to three, with the weight correction factor amplified as the level increases. Local smoothing and global calibration are performed cyclically according to continuous time windows. After each window is completed, the difference before and after correction is calculated. When the difference is greater than the convergence threshold, the next iteration begins. The system records the node psychological weights after each update and writes them into the weight update table, including the node index, corrected weight, timestamp, and iteration round fields.

[0035] For high-density congested areas or potentially dangerous nodes, a weighted attenuation and flow direction correction strategy is introduced to spatially differentiate the psychological behavior weight matrix, thus forming a psychological behavior weight matrix; For detected high-density congested nodes or potentially dangerous nodes, a spatially differentiated adjustment strategy is implemented to limit the abnormal spread of weights and correct the flow direction. High density is determined based on density thresholds and density gradients. When the population density per unit area exceeds four people per square meter or the density increase rate exceeds 0.5 people per square meter per second, it is recorded as a high-density node. An attenuation factor is applied to the psychological weight of the marked node. The attenuation coefficient can be set to 0.7. The ratio of the node's capacity to its adjacent paths is used to calculate the flow direction correction. The correction result is written into the outgoing edge value field. After adjustment, the node's psychological weight, attenuation factor, flow direction correction, and update time are written into the psychological weight matrix table and stored and organized according to spatial unit index and time index. A spatial matrix structure is constructed using row and column encoding. The matrix unit corresponds to the node weight value and its outgoing edge correction value. After the table completes the data writing, it forms the psychological behavior weight matrix at the current time.

[0036] The process of forming a value system that distinguishes between evacuation and rescue routes in S3 is as follows: Based on the psychological behavior weight matrix, the velocity gradient, local conflict probability, psychological behavior weight and reachability parameters are processed in a unified manner. Based on the psychological behavior weight matrix at the current moment, the real-time flow velocity and the flow velocity gradient within the continuous time segment are extracted node by node. The velocity change per unit time is calculated by the difference between adjacent time segments. The local conflict probability is calculated based on the direction deviation statistics. The variance of the direction vector distribution within the node is used as the conflict index input. At the same time, the accessibility parameters are calculated based on the average psychological behavior weight in the neighborhood of the node and the road traffic status.

[0037] Calculate the weights for population dispersion, exit accessibility, and local congestion sensitivity for evacuation routes; Within the evacuation route solution area, the population dispersion index is calculated for each node. The dispersion is calculated based on the population density variance of the node and its adjacent nodes. The exit accessibility score is calculated based on the distance and width of the nearest exit. The distance is converted to meter-level path length from the spatial grid index, and the exit width is taken from the on-site layout archive data. The local congestion sensitivity is determined by the node density and the density increase rate. When the density exceeds three people per square meter and the growth rate is greater than 0.3 people per square meter per second, the sensitivity level is increased. The calculation results are written into the evacuation route cost table.

[0038] Calculate the accessibility level, blockage risk, interference index, and time sensitivity weight for rescue routes; For the passage routes of rescue vehicles or personnel, the smoothness level and obstruction risk are calculated for each passable node and connecting path segment. The smoothness level is based on the passage width, obstacle layout and real-time crowd flow assessment recorded by the on-site security system. The obstruction risk is judged based on emergency information, fence location and temporary control records. If there is a security checkpoint or warning line deployed in front of the node, the obstruction risk level is increased. The interference index is determined by referring to the disturbance amplitude of the surrounding crowd direction and the noise event retrieval results. The time sensitivity is set according to the rescue type. For example, the time sensitivity of the rescue route is higher in medical emergency scenarios than in equipment transportation scenarios. All indicators are allocated according to the weight table and recorded in the rescue route value table.

[0039] A dual-purpose path cost system is generated using a joint evaluation method at the node and path levels. The node value records of evacuation and rescue routes are used as the node layer input. At the same time, the path value is calculated for each path segment according to its length, slope and the comprehensive weight of the nodes. The path segment value is calculated by weighting the mean node value with the path parameters. Value tables are generated for evacuation routes and rescue routes respectively, and then merged into a dual-purpose path value system. After storage, a value system that distinguishes between evacuation and rescue routes is formed.

[0040] S4: Perform path solving calculations according to the value system, output the set of evacuation paths and the set of rescue priority channels, and update the path parameters based on regional traffic capacity, real-time psychological state factors and flow trends to form a dual-path guidance configuration table.

[0041] The process of solving the calculation according to the value system in S4 is as follows: Based on the value system, graph search, shortest path or multi-objective optimization algorithms are used to solve the path; The regional topology and cost table are mapped to a weighted directed graph structure. The nodes of the graph correspond to spatial unit numbers, and the directed edges correspond to connectivity edges. The edge attributes include two sets of weights for evacuation and rescue costs, edge length, and maximum throughput. For the origin-endpoint pairs requiring online solution, a search subgraph is constructed to limit the search range. The search subgraph extracts nodes within its neighborhood using the influence domain of the origin radius. The solution algorithm can employ graph search algorithms or multi-objective optimization algorithms. Depending on the scenario, a single-objective heuristic search or a multi-objective ranking method is selected. The single-objective heuristic search uses a heuristic function that is a linear combination of the Euclidean distance between nodes and the edge weight. The heuristic weight can be configured between 0.5 and 1.0 to adjust for speed priority or cost priority. The multi-objective optimization uses a weight vector method to synthesize the evacuation and rescue costs into a composite cost value according to pre-configured coefficients and then solves for the shortest path. The algorithm input format includes the origin number, destination number, timestamp, cost matrix position index, and path length constraint field. After the algorithm runs, it outputs the path sequence and the cumulative path cost.

[0042] In the solution process, path continuity, node turning complexity, regional accessibility, and real-time psychological behavior weights are incorporated into the constraints. During path generation, each candidate path is superimposed with a turning complexity penalty, a continuity penalty, and a traffic capacity constraint on top of the accumulated cost value. The turning complexity penalty is calculated based on the angle between three adjacent points in the path. When the angle is less than 130 degrees, a turning cost value is added. The example value of the turning penalty is an increase of 0.1 to 0.3 in cost value for each turning. The continuity penalty scores the occurrence of discontinuous nodes or the need for route switching in the path segment. The discontinuity determination is based on the maximum walkable distance between nodes and the edge connectivity identifier. During path evaluation, the traffic capacity constraint marks path segments that exceed the maximum traffic capacity of nodes or edges as infeasible and removes the path segment. The traffic capacity threshold is set based on the number of people passing through per minute. The real-time psychological behavior weight uses the node weight matrix value as the multiplier coefficient of the dynamic cost value.

[0043] Output the set of evacuation routes and the set of rescue routes, and generate a multi-level route candidate set; After the solution is completed, all feasible paths are post-processed. The output structure of the path is path identifier, node sequence, total path length, cost details of each segment, estimated travel time, list of the most congested segments, and cost version number used when the path was generated. The system sorts the path set according to evacuation purpose and rescue purpose respectively and generates a multi-level path candidate set.

[0044] The process of creating a dual-path boot configuration table in S4 is as follows: The multi-level path candidate set is organized according to the node index and path sequence structure to generate path guidance configuration information; The candidate path set is indexed and organized. Each path is assigned a unique path identifier and the node sequence, starting node number, ending node number, total path distance, and total agency value are recorded. A path guidance table is established in the data storage system, with fields including path identifier, node sequence, start timestamp, path purpose, total agency value, and configuration version record. The path purpose field is used to distinguish between evacuation paths and rescue paths. The time field is recorded at the millisecond level to support real-time updates. The node sequence field stores the node numbers passed by the path in sequence. The organization process involves traversing the candidate path set, generating a structured record for each path, and writing it into the table.

[0045] Index and label each node in the path set, and calculate the connectivity between nodes, the turning complexity, and the local traffic load. For each node in the path, index labeling is performed and three types of indicators are calculated. The connectivity indicator is obtained through neighborhood topology query. The calculation formula is the ratio of the number of adjacent traversable edges to the node degree and is recorded as an integer score value. The turning complexity is represented by the statistics of the angle formed by three adjacent points in the path. The neighboring angle is calculated for each node and the turning count of the turning angle is accumulated as the turning number. Then, the turning number is multiplied by the turning penalty coefficient as the turning complexity value. The penalty coefficient can be set to 0.2 in an example. The local flow load is obtained by dividing the number of people flowing in per unit time by the available passage width of the node. The time window used is three seconds. When the load ratio exceeds the threshold, such as 250 people per meter per minute, it is recorded as high load.

[0046] The path level weights and priorities are dynamically adjusted based on real-time psychological state factors, density gradients, and flow trends. The read psychological weight value and density gradient value are multiplicatively combined to form an adjustment factor. The adjustment factor is equal to the base factor multiplied by the psychological weight multiplied by the density gradient amplification factor. The density gradient is defined as the density difference per unit time. The flow trend is calculated by using the magnitude of the neighborhood mainstream vector and the consistency coefficient to reduce the priority of the direction opposite to the mainstream. To avoid drastic fluctuations in weight, the adjustment factor is restricted and pruned. The pruning range is set to 0.5 to 1.5. After adjustment, the adjustment factor is aggregated by the path layer nodes. The overall priority weight of the path is calculated and mapped to the path level weight value. After each adjustment, the updated record is written back to the weight field of the path guidance table. The written fields include the update timestamp, the source identifier of the adjustment factor, the pruned factor value, and the new path level weight.

[0047] During the dynamic adjustment process, an updated path priority sequence is generated, and the level changes, applicable time windows, and key node identifiers of each path are recorded to form a dual-path guidance configuration table. After each weight update, the adjustment module reorders the path set, generates priority sequences and assigns candidate levels for evacuation and rescue paths respectively, and stores the level numbers in the order of No. 1, No. 2, and No. 3. At the same time, it extracts the list of key nodes for each path. Key nodes are composed of nodes with high turning complexity, high local traffic load, or prominent psychological weight and writes them into the key node field. Finally, a dual-path guidance configuration table is created in the database.

[0048] S5: During personnel movement, continuously collect density, speed, direction and psychological state indicators of each path unit, and calculate the path change trigger threshold in combination with the dual path guidance configuration table, and generate corresponding path adjustment instructions based on the threshold calculation results.

[0049] The process of generating the corresponding path adjustment instruction based on the threshold calculation result in S5 is as follows: The density, speed, direction, and psychological state parameters of each path unit are continuously collected and compared in real time with the thresholds in the dual-path guidance configuration table. During operation, the system collects path unit status indicators at fixed time intervals, with the collection period set to the range of 500 milliseconds to 1 second. The collected data includes the population density value, average speed value, main direction vector, and psychological state indicators of the area where the node is located. The psychological state indicators are read from the psychological weight matrix, including anxiety coefficient, conformity tendency coefficient, and contrarian tendency coefficient. All collected data are timestamped and encoded with path unit and written to a temporary buffer.

[0050] The path adjustment range and scope are calculated based on the deviation coefficient, node congestion level and path priority, and the calculation results are parsed into an executable path adjustment instruction set. The deviation magnitude calculation logic is executed according to the proportional model. If the density exceeds the threshold of 20%, the node is marked as moderately congested, and if it exceeds 40%, it is marked as severely congested. Combining the current priority of the path and the node position, the adjustment scope is determined to be a single node or a continuous node link. The module then parses the calculation results into a structured path adjustment instruction. The instruction parameters include the target path number, the target node number range, the adjustment type field, the adjustment magnitude value, the execution duration, and the instruction effective timestamp. After the instruction is generated, it is written into the path control instruction table and pushed to the guidance terminal and the command platform through the message queue.

[0051] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0052] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for dynamic reconstructing and guiding travel based on congestion and pedestrian safety, characterized in that, Includes the following steps: Real-time data collection is performed on pedestrian density, velocity vector, direction of movement, and loitering units within the target area. Exit nodes, obstacle nodes, and event source nodes are loaded according to the area topology to construct a spatial state matrix. Based on the spatial state matrix, the density, speed, direction and dwell time changes of each spatial unit within a continuous time window are recorded. A spatiotemporal population state database is established by combining historical flow trajectories and scene structure, and behavioral feature vectors are extracted to generate a basic behavioral feature pattern library. Based on a spatiotemporal population state database and a basic behavioral characteristic pattern library, a group psychodynamic mechanism is introduced to calculate the panic index, herd mentality intensity, and reverse migration tendency of each spatial unit, forming a psychological behavior weight matrix. The flow velocity gradient, local conflict probability, accessibility, and psychological behavior weight are weighted and integrated to form a value system that distinguishes evacuation and rescue paths. The path calculation is performed according to the value system, and the set of evacuation paths and the set of rescue priority channels are output. The path parameters are updated according to the regional traffic capacity, real-time psychological state factors and flow trend to form a dual-path guidance configuration table. During personnel movement, the density, speed, direction, and psychological state indicators of each path unit are continuously collected. The path change trigger threshold is calculated in conjunction with the dual-path guidance configuration table, and the corresponding path adjustment instruction is generated based on the threshold calculation results.

2. The method for dynamic reconstruction and guidance of travel based on congestion and pedestrian safety according to claim 1, characterized in that, The process of constructing the spatial state matrix is ​​as follows: Based on the target area structure, obtain the channel network, node locations, obstacle distribution, and exit connectivity. Cameras, sensors, and positioning modules are deployed within the area to obtain pedestrian location coordinates, movement trajectories, and hotspots where they stop; The collected pedestrian motion features are mapped to a spatial grid structure and assigned to corresponding nodes and connected edges to generate a spatial state matrix containing spatial indexes, node attributes, and flow parameters.

3. The method for dynamic reconstruction and guidance of travel based on congestion and pedestrian safety according to claim 2, characterized in that, The process of establishing a spatiotemporal population status database by combining historical movement trajectories and scene structures is as follows: Based on the spatial state matrix, the real-time collected density, velocity, direction and residence data are continuously stored in time series and combined with spatial unit numbering to form regional time grids. Trajectory reconstruction and event labeling of pedestrian movement and dwelling trajectories within a continuous time window; Based on the regional topology, the trajectory is mapped to the corresponding access unit, boundary unit, and event source node; By introducing historical scene operation data, including the distribution of crowd flow, entrance opening strategies and regional functional attributes at different times, real-time trajectories are matched and integrated with historical scene structure information to establish a multi-dimensional spatiotemporal crowd status database that includes spatial location index, time series labels, behavioral event identifiers and environmental conditions.

4. The method for dynamic reconstruction and guidance of travel based on congestion and pedestrian safety according to claim 3, characterized in that, The process of extracting behavioral feature vectors to generate a basic behavioral feature pattern library is as follows: In a multidimensional spatiotemporal population status database, density changes, velocity changes, directional shifts, and dwell time series data within a continuous time window are extracted, and corresponding individual and group behavior segments are located based on spatial unit indexes and time labels. Behavioral patterns are decomposed into individual trajectories and local state fluctuations to identify typical micro-behavioral features such as acceleration, sudden stop, aggregation and stagnation, deflection and reverse movement, and encoded in vector form. Based on spatial topology and node relationships, the group behavior is analyzed in a structured manner to extract macro-level group behavior variables; Multi-scale clustering analysis was performed on the extracted individual and group behavioral feature vectors. The clustering results were combined with scene area attributes, time period classification, and event labels to construct a basic behavioral feature pattern library.

5. The method for dynamic reconstruction and guidance of travel based on congestion and pedestrian safety according to claim 4, characterized in that, The process of introducing group psychodynamic mechanisms is as follows: Based on the basic behavioral feature pattern library, a mapping model between behavioral features and psychological parameters is established, including the correspondence between behavioral acceleration, density gradient change rate, path deviation fluctuation amplitude, conformity rate, and reverse escape tendency. We perform emotional feature enhancement processing on continuous time slice behavioral data, and infer the group stress gradient and the direction of anxiety propagation through local density mutation detection, velocity anomaly clustering identification and directional perturbation analysis. Psychological parameters are calculated from individual and group behavioral inputs to obtain psychodynamic indicators.

6. The method for dynamic reconstruction and guidance of travel based on congestion and pedestrian safety according to claim 5, characterized in that, The process of forming the psychological behavior weight matrix is ​​as follows: Based on psychodynamic indicators, the panic index, herd mentality intensity, and reverse migration impulse of each spatial unit are weighted and integrated with the node access attributes. The weights of the merged nodes are adjusted locally and globally. A dynamic correction mechanism is constructed by combining the neighborhood relationship between nodes, path connectivity and event level information, and the weights of psychological behavior are iteratively updated within a continuous time window. For high-density congested areas or potentially dangerous nodes, a weighted attenuation and flow direction correction strategy is introduced to spatially differentiate the psychological behavior weight matrix, thus forming a psychological behavior weight matrix.

7. The method for dynamic reconstruction and guidance of travel based on congestion and pedestrian safety according to claim 6, characterized in that, The process of forming a value system that distinguishes between evacuation and rescue routes is as follows: Based on the psychological behavior weight matrix, the velocity gradient, local conflict probability, psychological behavior weight and reachability parameters are processed in a unified manner. Calculate the weights for population dispersion, exit accessibility, and local congestion sensitivity for evacuation routes; Calculate the accessibility level, blockage risk, interference index, and time sensitivity weight for rescue routes; A dual-purpose path value system is generated by using a joint evaluation method at the node and path levels.

8. The method for dynamic reconstruction and guidance of travel based on congestion and pedestrian safety according to claim 7, characterized in that, The process of solving the calculation according to the value system execution path is as follows: Based on the value system, graph search, shortest path or multi-objective optimization algorithms are used to solve the path; In the solution process, path continuity, node turning complexity, regional accessibility, and real-time psychological behavior weights are incorporated into the constraints. Output the set of evacuation routes and the set of rescue routes, and generate a multi-level route candidate set.

9. The method for dynamic reconstruction and guidance of a journey based on congestion and pedestrian safety according to claim 8, characterized in that, The process of creating a dual-path boot configuration table is as follows: The multi-level path candidate set is organized according to the node index and path sequence structure to generate path guidance configuration information; Index and label each node in the path set, and calculate the connectivity between nodes, the turning complexity, and the local traffic load. The path level weights and priorities are dynamically adjusted based on real-time psychological state factors, density gradients, and flow trends. During the dynamic adjustment process, an updated path priority sequence is generated, and the level changes, applicable time windows, and key node identifiers of each path are recorded to form a dual-path guidance configuration table.

10. The method for dynamic reconstruction and guidance of a journey based on congestion and pedestrian safety according to claim 9, characterized in that, The process of generating the corresponding path adjustment instruction based on the threshold calculation result is as follows: The density, speed, direction, and psychological state parameters of each path unit are continuously collected and compared in real time with the thresholds in the dual-path guidance configuration table. The path adjustment range and scope are calculated based on the deviation coefficient, node congestion level, and path priority. The calculation results are then parsed into an executable set of path adjustment instructions.

Images