Method and system for dynamic recommendation of crowd evacuation path based on visual data

Abstract

The application is suitable for the field of computer vision, and provides a crowd evacuation path dynamic recommendation method and system based on visual data. The application discloses a crowd evacuation path dynamic recommendation method and system based on visual data in the field of computers. Real-time crowd motion video streams are collected, and pedestrian head orientation vectors are extracted after preprocessing to form a crowd motion velocity field and establish a crowd motion state database. A visual inertia dynamics model is constructed to predict future crowd evacuation path distribution changes. Real-time carrying rates of each channel are calculated, and a saturation time is generated by combining a capacity threshold value to generate a graded early warning. When the channel reaches the optimal capacity threshold value, an upstream bifurcation node is determined, a diversion start time window and intensity are calculated, and a control instruction is generated and transmitted to an indication execution end. The application realizes accurate dynamic path recommendation and improves evacuation safety and efficiency.

Description

Technical Field

[0001] This invention belongs to the field of computer vision, and in particular relates to a method and system for dynamic recommendation of crowd evacuation routes based on visual data. Background Technology

[0002] With the routine operation of large public venues, transportation hubs, and commercial complexes, emergency evacuation safety in densely populated areas has become a core concern in the field of public safety. In recent years, the rapid development of computer vision technology, the widespread adoption of video acquisition equipment, and the maturity of target detection and motion analysis algorithms have driven the transformation of crowd monitoring from traditional manual patrols to automation and intelligence. Currently, the industry has formed basic technical capabilities such as real-time crowd density statistics and evacuation route delineation based on video streams. Some solutions have begun to attempt simple path recommendations based on crowd movement trajectories, but overall, they remain at the static guidance stage, lacking in-depth analysis of the intrinsic driving factors and dynamic trends of crowd movement, making it difficult to meet the needs of efficient evacuation in complex scenarios.

[0003] Current technologies cannot capture the visual inertial characteristics of crowd movement, relying solely on trajectory or density data to predict changes in pedestrian flow distribution. This results in insufficient accuracy in predicting the dynamic trends of crowds along future evacuation routes. It is difficult to combine real-time channel load with future pedestrian flow changes to accurately determine the rationality of the use of each evacuation channel, and it is impossible to promptly identify situations where some channels are overcrowded while others are idle. Guidance strategies lack support for predicting channel saturation times, and guidance nodes are mostly fixed settings without early initiation of guidance at upstream forks. This can easily lead to problems such as untimely guidance or diversion to already saturated channels causing secondary congestion, making it impossible to achieve true dynamic path optimization. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for dynamic recommendation of crowd evacuation routes based on visual data, aiming to solve the technical problems existing in the prior art as identified in the background art.

[0005] This invention is implemented as follows: a dynamic crowd evacuation path recommendation method based on visual data, the method comprising:

[0006] Real-time acquisition of crowd movement video streams and preprocessing to obtain standardized video frames; extraction of head orientation vectors for each pedestrian; simultaneous calculation of the group movement velocity field; and establishment of a crowd movement state database.

[0007] Establish a visual inertial dynamics model, calculate the angle distribution between the head orientation vector and the direction of movement, determine the visual inertial intensity parameters, and predict the distribution changes of the crowd on the evacuation path in the next 5-10 minutes.

[0008] Based on real-time video streams, identify the areas of each evacuation route, calculate the real-time crowd density of each route, and calculate the real-time carrying capacity by combining the physical parameters of the route.

[0009] Based on the real-time carrying capacity of each evacuation route, a path saturation time prediction model is established in combination with the route capacity threshold. The time point when each evacuation route reaches the maximum saturation is calculated, and a graded early warning signal is generated.

[0010] When it is predicted that any evacuation route will reach 85% of its optimal capacity threshold, the coordinates of the bifurcation nodes of the upstream path are determined based on the changes in the distribution of people on the evacuation route and the graded early warning signals. The optimal time window for diversion and the diversion intensity parameters are calculated, a set of control instructions is generated, and the instructions are transmitted to the execution end.

[0011] As a further aspect of the present invention, the establishment of a population movement state database specifically includes:

[0012] Continuously and in real time, the original video stream of crowd movement is acquired, and the received original video stream of crowd movement is preprocessed to generate standardized video frames.

[0013] Standardized video frames are input into a pre-trained head pose estimation neural network to detect and locate the two-dimensional pixel coordinates of the head in the image, output the pitch angle, yaw angle and roll angle of each visible pedestrian head, and convert them into a head orientation vector in three-dimensional space.

[0014] Based on a standardized video frame sequence, the motion vector of each pixel between adjacent frames is calculated. Based on the motion vectors of all pedestrian area pixels, a group motion velocity field representing the overall movement direction is formed.

[0015] Establish a unified spatiotemporal coordinate system, synchronize the head orientation vector with the group's motion velocity field under the unified spatiotemporal coordinate system for time stamping and spatial alignment, and establish a database of crowd motion state containing timestamps, spatial positions, head orientation vectors and local motion vectors.

[0016] As a further aspect of the present invention, the prediction of changes in the distribution of people along evacuation routes within the next 5-10 minutes specifically includes:

[0017] Based on the head orientation vector and the group motion velocity field, the angle between the head orientation vector of each pedestrian and the corresponding local motion direction of the pedestrian is calculated.

[0018] Statistically analyze the distribution histogram of the angles of all pedestrians at the current time and within the preset time window in the past, and calculate the visual inertia intensity;

[0019] Using visual inertial intensity, real-time crowd density, and environmental visibility as inputs, a visual inertial dynamics model is constructed. The parameters in the current crowd motion state database are used as initial conditions, and numerical iterations are performed over the next 5-10 minutes to predict the distribution changes of the crowd along the evacuation path.

[0020] As a further embodiment of the present invention, the calculation of real-time load factor specifically includes:

[0021] Based on real-time video streams, combined with building structure topology maps, the system identifies passage boundaries and entrances / exits, divides the scene into different evacuation passage areas, and calculates the real-time crowd density in each evacuation passage area.

[0022] The physical parameters of each evacuation route area are read from the pre-set building information database, and the real-time carrying capacity of each evacuation route area is calculated by combining the real-time population density.

[0023] As a further embodiment of the present invention, the generation of graded early warning signals specifically includes:

[0024] Based on the real-time population density and real-time carrying capacity of each evacuation route area, time series data of the real-time carrying capacity of each evacuation route area are generated;

[0025] Read the historical capacity threshold of each evacuation route area. The historical capacity threshold is the maximum carrying capacity when the route reaches full saturation. Set the optimal capacity threshold to 85% of the historical capacity threshold data.

[0026] By performing pattern matching between real-time carrying capacity time series data and historical capacity thresholds, a path saturation time prediction model is established. Based on the current real-time carrying capacity and the predicted distribution changes of the population on the evacuation path, the future load growth of each evacuation channel area is simulated and deduced, and the time point when the real-time carrying capacity of each evacuation channel area reaches the optimal capacity threshold is calculated.

[0027] Based on the difference between the calculated time when the real-time load factor reaches the optimal capacity threshold and the current time, a tiered early warning signal is generated, with the warning level determined by the magnitude of the difference.

[0028] As a further embodiment of the present invention, the set of generation control instructions specifically includes:

[0029] When a graded early warning signal appears in any evacuation route area, combined with the distribution changes of people on the evacuation route, based on the building structure topology map, the nearest path fork node is found upstream of the evacuation route area, and the coordinates of the fork node are determined as the coordinates of the fork node of the upstream path.

[0030] Determine whether there are effective graded early warning signals in other evacuation channel areas guided by the location coordinates of the path fork node. If there are, take the path fork node as a new starting point and continue to search for the next path fork node upstream along the building structure topology map, and repeat the judgment process; if there are no, determine the path fork node location coordinates as the fork node location of the upstream path used for diversion.

[0031] Based on the time point when the real-time carrying rate in the graded early warning signal reaches the optimal capacity threshold, the deadline for initiating guidance is calculated by working backwards, and the crowd response delay is taken into account to calculate the optimal time window for initiating guidance.

[0032] The required diversion intensity parameters are calculated based on the difference between the real-time carrying capacity and the optimal capacity threshold, as well as the predicted population density level.

[0033] The coordinates of the bifurcation node of the upstream path, the optimal time window for diversion initiation, and the diversion intensity parameters are encapsulated into a structured set of control instructions and transmitted to the instruction execution terminal located at the coordinates of the bifurcation node.

[0034] Another object of the present invention is to provide a dynamic recommendation system for crowd evacuation routes based on visual data, the system comprising:

[0035] The video data extraction module is used to collect real-time video streams of crowd movement and preprocess them to obtain standardized video frames, extract the head orientation vector of each pedestrian, calculate and form a group movement velocity field, and establish a crowd movement state database.

[0036] The crowd prediction module is used to establish a visual inertial dynamics model, calculate the angle distribution between the head orientation vector and the direction of movement, determine the visual inertial intensity parameters, and predict the distribution changes of the crowd on the evacuation path in the next 5-10 minutes.

[0037] The carrying capacity calculation module is used to identify the areas of each evacuation route based on real-time video streams, calculate the real-time crowd density of each route, and calculate the real-time carrying capacity in combination with the physical parameters of the route.

[0038] The early warning generation module is used to establish a path saturation time prediction model based on the real-time carrying capacity rate of each evacuation route and the route capacity threshold, calculate the time point when each evacuation route reaches the maximum saturation, and generate graded early warning signals.

[0039] The instruction transmission module is used to determine the coordinates of the bifurcation node of the upstream path based on the distribution changes of the crowd on the evacuation path and the graded early warning signals when it is predicted that any evacuation channel will reach the optimal capacity threshold of 85%. It also calculates the optimal time window for diversion and diversion intensity parameters, generates a set of control instructions, and transmits them to the instruction execution end.

[0040] As a further embodiment of the present invention, the crowd prediction module includes:

[0041] The angle calculation module is used to calculate the angle between the head orientation vector of each pedestrian and the local motion direction corresponding to the pedestrian, based on the head orientation vector and the group motion velocity field.

[0042] The angle distribution statistics module is used to statistically analyze the distribution histogram of the angles of all pedestrians at the current time and within a preset time window in the past, and to calculate the visual inertial intensity.

[0043] The crowd evacuation flow prediction module is used to construct a visual inertial dynamics model by taking visual inertial intensity, real-time crowd density and environmental visibility as inputs. It uses the parameters in the current crowd motion state database as initial conditions and performs numerical iterations over the next 5-10 minutes to predict the distribution changes of the crowd along the evacuation path.

[0044] As a further embodiment of the present invention, the early warning generation module includes:

[0045] The carrying capacity time series generation module is used to generate time series data of the real-time carrying capacity of each evacuation route area based on the real-time population density and real-time carrying capacity of each evacuation route area;

[0046] The capacity threshold setting module is used to read the historical capacity threshold of each evacuation route area. The historical capacity threshold is the maximum carrying capacity when the route reaches full saturation, and the optimal capacity threshold is set to 85% of the historical capacity threshold data.

[0047] The threshold achievement time calculation module is used to perform pattern matching between real-time carrying capacity time series data and historical capacity thresholds, establish a path saturation time prediction model, simulate and extrapolate the future load growth of each evacuation channel area based on the current real-time carrying capacity and the predicted distribution changes of the population on the evacuation path, and calculate the time point when the real-time carrying capacity of each evacuation channel area reaches the optimal capacity threshold.

[0048] The graded early warning signal generation module is used to generate graded early warning signals based on the difference between the calculated time point when the real-time load rate reaches the optimal capacity threshold and the current time. The early warning level is divided according to the magnitude of the difference.

[0049] As a further embodiment of the present invention, the instruction transmission module includes:

[0050] The path bifurcation node identification module is used to find the nearest path bifurcation node upstream of any evacuation route area when a graded early warning signal appears in any evacuation route area, combined with the distribution changes of people on the evacuation route, based on the building structure topology map, and determine the bifurcation node location coordinates as the bifurcation node location coordinates of the upstream path.

[0051] The effective graded early warning signal judgment module is used to determine whether there are effective graded early warning signals in other evacuation channel areas guided by the location coordinates of the path branch node. If there are, the path branch node is used as a new starting point to continue searching for the next path branch node upstream along the building structure topology map, and the judgment process is repeated. If there are no, the path branch node is determined as the location coordinates of the branch node of the upstream path used for diversion.

[0052] The diversion start-up time calculation module is used to calculate the optimal time window for diversion start-up by working backward from the time point when the real-time carrying rate in the graded early warning signal reaches the optimal capacity threshold, and taking into account the crowd response delay.

[0053] The flow guidance intensity parameter calculation module is used to calculate the required flow guidance intensity parameters based on the difference between the real-time carrying capacity and the optimal capacity threshold, as well as the predicted population density level.

[0054] The control command encapsulation and transmission module is used to encapsulate the bifurcation node position coordinates of the upstream path, the optimal time window for diversion initiation, and the diversion intensity parameters into a structured set of control commands, and transmit them to the instruction execution terminal located at the bifurcation node position coordinates.

[0055] The beneficial effects of this invention are:

[0056] This invention effectively addresses the shortcomings of existing technologies in crowd flow prediction and dynamic guidance by deeply integrating visual data with crowd visual inertial characteristics. The solution first extracts pedestrian head orientation vectors and combines them with the group's motion velocity field to construct a visual inertial dynamics model. This model accurately captures the inherent patterns of "eye-guided movement" in crowds, significantly improving the accuracy of predicting changes in crowd distribution along evacuation routes within the next 5-10 minutes, providing a reliable trend basis for subsequent guidance strategy formulation. Simultaneously, by calculating the capacity of each evacuation channel in real time and combining historical capacity thresholds to predict channel saturation times, the solution can clearly understand the usage status of each channel, promptly identify overcrowding or underutilization, and ensure the rational utilization of evacuation resources. In the guidance and execution phase, the solution dynamically retrieves effective upstream branch nodes to avoid diverting crowds to saturated channels. It also determines the optimal guidance time window by combining crowd response latency and command execution latency, and matches the guidance intensity with load requirements. The guidance operation is initiated at the upstream branch point in advance, which not only ensures the timeliness of guidance but also avoids secondary congestion. Ultimately, it achieves full-process, accurate dynamic recommendation of crowd evacuation paths based on visual data, which significantly improves the safety and efficiency of evacuation in dense crowd scenarios. Attached Figure Description

[0057] Figure 1A flowchart of a dynamic crowd evacuation route recommendation method based on visual data provided in an embodiment of the present invention;

[0058] Figure 2 A flowchart for establishing a population movement status database provided in an embodiment of the present invention;

[0059] Figure 3 A flowchart for predicting changes in the distribution of people along evacuation routes within the next 5-10 minutes, provided as an embodiment of the present invention;

[0060] Figure 4 A flowchart for calculating real-time load factor is provided in an embodiment of the present invention;

[0061] Figure 5 A flowchart for generating graded early warning signals provided in an embodiment of the present invention;

[0062] Figure 6 A flowchart for generating a set of control instructions provided in an embodiment of the present invention;

[0063] Figure 7 This is a structural block diagram of a dynamic recommendation system for crowd evacuation routes based on visual data provided in an embodiment of the present invention.

[0064] Figure 8 This is a structural block diagram of the crowd prediction module provided in an embodiment of the present invention;

[0065] Figure 9 This is a structural block diagram of the early warning generation module provided in an embodiment of the present invention;

[0066] Figure 10 This is a structural block diagram of the instruction transmission module provided in an embodiment of the present invention. Detailed Implementation

[0067] 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.

[0068] Figure 1 A flowchart of the dynamic crowd evacuation route recommendation method based on visual data provided in this embodiment of the invention is shown below. Figure 1 As shown, the method includes:

[0069] S100: Real-time acquisition of crowd movement video streams and preprocessing to obtain standardized video frames; extraction of head orientation vectors for each pedestrian; simultaneous calculation of the group movement velocity field; and establishment of a crowd movement state database.

[0070] We continuously and in real-time acquire raw video streams of crowd movement. Based on this, we perform image preprocessing, generating standardized video frames by unifying image resolution, adjusting brightness and contrast, removing noise, and unifying frame rate, thus eliminating interference caused by factors such as changes in ambient light, equipment differences, and image distortion.

[0071] Standardized video frames are input into a pre-trained head pose estimation neural network, which can accurately detect and locate the two-dimensional pixel coordinates of each visible pedestrian's head, and further output pitch angle, yaw angle and roll angle. It also obtains a three-dimensional head orientation vector through spatial coordinate transformation, quantifies the pedestrian's visual attention direction into a computable spatial vector, captures the pedestrian's subjective movement intention, and can not only observe the pedestrian's actual movement trajectory, but also perceive its walking tendency in advance.

[0072] Based on standardized video frame sequences, the motion vector of each pixel between adjacent frames is calculated using optical flow. Then, based on the pixels in the pedestrian area, a group motion velocity field is formed through grid division and average velocity statistics. By aggregating the motion information of individual pixels and filtering out irrelevant behaviors such as random swaying and brief pauses, the overall movement direction and speed characteristics of the crowd are extracted, and the macroscopic laws of group movement are grasped.

[0073] Establish a unified spatiotemporal coordinate system, synchronize the head orientation vector with the group's motion velocity field using timestamps and align their spatial positions, integrate key information such as timestamps, spatial positions, head orientation vectors, and local motion vectors, construct a complete database of crowd motion states, and achieve the organic integration of individual intentions and group trend data.

[0074] S200: Establish a visual inertial dynamics model, calculate the angle distribution between the head orientation vector and the direction of movement, determine the visual inertial intensity parameters, and predict the distribution changes of the crowd on the evacuation path in the next 5-10 minutes.

[0075] Based on a crowd movement state database, the head orientation vector and corresponding local movement direction vector of each pedestrian are extracted, and the angle between the two is calculated to quantify the degree of matching between the pedestrian's subjective line of sight guidance and objective movement behavior. When the angle is small, it indicates that the pedestrian's movement direction is consistent with their line of sight, and the behavior has clear autonomy and predictability. When the angle is large, it often means that the pedestrian may be affected by factors such as crowding, environmental interference, or following others, and the movement behavior shows a certain degree of randomness.

[0076] By statistically analyzing the histograms of the angle distribution of all pedestrians within the current time and past preset time windows, and by analyzing the proportion of pedestrians in the low angle range of the histograms, the intensity of visual inertia is determined. The core function of visual inertia intensity is to aggregate individual-level behavioral characteristics into group motion attributes, reflecting the degree of inertia of the group as a whole guided by line of sight. If the visual inertia intensity is high, it indicates that the group's overall movement is orderly and follows its own line of sight guidance; if the intensity is low, it suggests that the group's movement may be disordered, and more random motion factors need to be included in the prediction.

[0077] Based on this, a visual inertial dynamics model is constructed, using visual inertial intensity, real-time crowd density, and environmental visibility as core input parameters. Visual inertial intensity determines the model's judgment logic on the dominant trend of crowd movement, real-time crowd density affects the rate of crowd flow and the probability of congestion, and environmental visibility is related to pedestrians' ability to recognize paths. In low-visibility scenarios, pedestrians may rely more on following behavior, leading to a decrease in visual inertial intensity, and the model needs to adjust the prediction weights through the visibility parameter. Simultaneously, using parameters such as timestamps, spatial locations, head orientation vectors, and local motion vectors from the crowd movement state database as initial conditions, the model is solved through numerical iteration. The numerical iteration process is a dynamic simulation of the crowd movement state. Each iteration updates the prediction results for the next moment based on the crowd distribution, movement direction, and environmental parameters of the previous moment, gradually extrapolating key information such as changes in crowd density and shifts in flow direction along various evacuation paths over the next 5-10 minutes.

[0078] S300, based on real-time video streams, identifies each evacuation route area, calculates the real-time crowd density of each route, and calculates the real-time carrying capacity by combining the physical parameters of the route.

[0079] Accurate identification of passage areas is achieved through the synergy of real-time video stream and building structure topology map. The building structure topology map provides physical boundary information of passages in the scene, including wall locations, entrance and exit distribution, and connection relationships of each passage. Based on this information, the scene space in the real-time video stream can be divided into different evacuation passage areas, ensuring that each area corresponds completely to the actual physical passage and avoiding the misinclusion of non-passage areas or transition areas between passages in the statistical scope.

[0080] Subsequently, based on the segmented channel area, the real-time crowd density is calculated using visual information in standardized video frames. The number of pedestrians in the area is identified through a target detection algorithm, and the number of pedestrians per unit area is obtained by combining the effective area of ​​the channel area. Since the video preprocessing in the early stage has eliminated interference such as noise and brightness differences, the accuracy of pedestrian recognition and the accuracy of the number statistics are guaranteed at this time, thereby ensuring that the density calculation results can truly reflect the degree of crowding in the channel.

[0081] The physical parameters of each evacuation route area are retrieved from the pre-set building information database. These parameters include the effective width and effective length of the route, as well as the presence of columns, obstacles, and other factors that affect the actual passage space. Based on these parameters, the maximum capacity of each route can be calculated, that is, the maximum number of pedestrians that the route can carry under safe conditions. The real-time pedestrian count is then compared with the maximum capacity to obtain the real-time load factor of each route. This load factor directly reflects the current load level and remaining carrying capacity of the route.

[0082] S400, based on the real-time carrying capacity of each evacuation route, and combined with the route capacity threshold, establishes a path saturation time prediction model, calculates the time point when each evacuation route reaches its maximum saturation, and generates graded early warning signals.

[0083] The system continuously collects and records load factor data at different time points, forming real-time load factor time series data specific to each channel. This time series data can intuitively reflect the changing pattern of channel load over time. In the early stages of event dissipation, the load factor time series of the main evacuation channel will show a rapid upward trend, while the secondary channels may show a slow upward trend or a stable state. By analyzing the trend of the time series data, the growth rate and potential pressure of the channel load can be preliminarily judged.

[0084] Historical capacity thresholds for each evacuation route are read. These thresholds are derived from the maximum carrying capacity data of the routes when they reached full saturation in various past scenarios. They have strong scenario adaptability and safety. Based on this, the optimal capacity threshold is set as the core standard for determining whether a route has entered the warning range.

[0085] By performing pattern matching between real-time load factor time-series data and historical capacity thresholds, and comparing the process of channel saturation under similar historical load growth trends, combined with changes in crowd distribution along evacuation routes, the model determines how many people will flow into the current channel and at what speed in the future, thereby constructing a path saturation time prediction model. This model starts with the current real-time load factor and uses the predicted crowd distribution as the basis for load growth to simulate and extrapolate the future change curve of channel load.

[0086] The warning levels are determined by the difference between the calculated time point and the current time. The smaller the time difference, the higher the warning level, indicating that the channel is about to reach the warning threshold. The larger the time difference, the lower the warning level. By classifying the channels, the urgency of different channels is clearly defined, providing a clear priority reference for subsequent resource allocation and traffic diversion strategy formulation.

[0087] S500, when it is predicted that any evacuation route will reach the optimal capacity threshold of 85%, determines the coordinates of the bifurcation node of the upstream path based on the distribution changes of the crowd on the evacuation path and the graded early warning signals, calculates the optimal time window for diversion and diversion intensity parameters, generates a set of control commands, and transmits them to the instruction execution end.

[0088] By combining the distribution changes of the crowd along the evacuation path with the building structure topology map, the upstream path branching nodes can be located. The distribution changes of the crowd along the evacuation path can clearly identify the source area of ​​the crowd that is about to flow into the saturation channel, while the building structure topology map provides the connection relationship between channels. Based on both, the branching node closest to the saturation channel can be identified. The location of this node needs to cover the main potential inflow crowd in order to intercept the subsequent pedestrians flowing into the saturation channel as early as possible and reduce the load pressure on the channel.

[0089] Determine whether there are effective graded warning signals for other evacuation routes led to by the branching node. If so, it indicates that these routes are also close to saturation. Diverting people to such routes will cause secondary congestion. Therefore, it is necessary to take this node as a new starting point and continue to search for the next branching node upstream along the building structure topology map until a node leading to a route without effective warning signals is found. This ensures that the diverted crowd can enter a route with sufficient capacity and guarantees the actual effect of the diversion operation.

[0090] After identifying the effective bifurcation node, the optimal time window for diversion initiation is calculated based on the time when the channel reaches the optimal capacity threshold in the tiered early warning signal. This requires deducting the crowd response delay and command execution delay. The crowd response delay is the time it takes for pedestrians to receive the guidance signal and begin moving, while the command execution delay is the time it takes for the instruction device to start operation after receiving the command. Taking these delays into account ensures that the diversion operation starts before the channel reaches saturation, allowing sufficient time for pedestrians to complete path transitions and preventing diversion failure due to late initiation. Simultaneously, the scale of people requiring diversion is determined based on the difference between the real-time carrying capacity and the optimal capacity threshold. This, combined with the predicted crowd density level, assesses the pressure of subsequent crowds flowing into the saturated channel. Both factors jointly determine the diversion intensity parameter. If the difference is significant and the subsequent crowd density level is high, the diversion intensity needs to be increased.

[0091] Finally, the coordinates of the bifurcation node of the upstream path, the optimal time window for diversion initiation, and the diversion intensity parameters are encapsulated into a structured set of control instructions, which are then transmitted to the instruction execution terminal located at the bifurcation node. The instruction execution terminal outputs guidance signals at a specified time and intensity according to the instructions, such as guide lights flashing at a set frequency and voice devices broadcasting turning prompts, guiding pedestrians from the saturated channel to other effective channels.

[0092] like Figure 2 As shown, the establishment of the crowd movement status database specifically includes:

[0093] S110 continuously and in real time acquires raw crowd movement video streams and performs image preprocessing on the received raw crowd movement video streams to generate standardized video frames.

[0094] S120 inputs standardized video frames into a pre-trained head pose estimation neural network, detects and locates the two-dimensional pixel coordinates of the head in the image, outputs the pitch angle, yaw angle and roll angle of each visible pedestrian head, and converts them into a head orientation vector in three-dimensional space.

[0095] The transformation of the head orientation vector is based on the mapping relationship between Euler angles, rotation matrix, and unit vector. The specific steps are as follows:

[0096] 1. Establish a right-handed coordinate system with the centroid of the pedestrian's head as the origin: the x-axis points to the right of the pedestrian, the y-axis points upward, and the z-axis points to the front of the pedestrian (the initial line of sight is the positive z-axis direction).

[0097] 2. Based on the pitch angle (α), yaw angle (β), and roll angle (γ), rotate sequentially around the y-axis (pitch), z-axis (yaw), and x-axis (roll) to obtain the combined rotation matrix. :

[0098] ;

[0099] in:

[0100] Pitch and Rotation Matrix ;

[0101] Yaw rotation matrix ;

[0102] Tumbling Rotation Matrix ;

[0103] The initial line of sight is a unit vector. (Positive z-axis direction), the 3D head orientation vector is obtained through rotation matrix transformation. ;

[0104] 3. Normalize to a unit vector to ensure consistent direction.

[0105] S130, based on a standardized video frame sequence, calculates the motion vector of each pixel between adjacent frames, and forms a group motion velocity field representing the overall movement direction based on the motion vectors of all pedestrian area pixels.

[0106] The group motion velocity field is a statistical aggregation of the pixel motion vectors in the pedestrian region. The specific steps are as follows:

[0107] 1. For continuous standardized video frames ( Each pixel is calculated using optical flow. motion vector , which represents the displacement rate of a pixel in the x / y direction.

[0108] 2. Extract pedestrian mask regions from video frames using object detection. (Only pedestrian pixels are retained).

[0109] 3. Divide the scene into sections of size [size missing]. The grid, for each grid Calculate the average motion vector for the pedestrian region pixels within the area:

[0110] ;

[0111] in For grid The number of pixels for an expert.

[0112] 4. Average velocity of each grid Assigning values ​​to the grid center coordinates creates a group motion velocity field with grid units, containing both the magnitude and direction of the velocity. .

[0113] S140. Establish a unified spatiotemporal coordinate system, synchronize the head orientation vector with the group motion velocity field under the unified spatiotemporal coordinate system for time stamping and spatial alignment, and establish a database of crowd motion state containing timestamps, spatial positions, head orientation vectors and local motion vectors.

[0114] like Figure 3 As shown, the prediction of the distribution changes of the crowd along the evacuation route in the next 5-10 minutes specifically includes:

[0115] S210, based on the head orientation vector and the group motion velocity field, calculate the angle between the head orientation vector of each pedestrian and the local motion direction corresponding to the pedestrian;

[0116] pedestrian The head orientation vector is Its corresponding local motion direction vector is (Taken from the average velocity vector of the grid where the pedestrian is located in the group velocity field), then the included angle The calculation formula is:

[0117] ;

[0118] in:

[0119] : The magnitude of the vector in which the head faces;

[0120] The magnitude of the local motion direction vector represents the local motion rate;

[0121] :pedestrian The angle between the head's orientation and the direction of movement, range .

[0122] S220: Calculate the distribution histogram of the angles of all pedestrians at the current time and within the preset time window in the past, and calculate the visual inertia intensity.

[0123] Visual inertial intensity (denoted as) The percentage of pedestrians whose direction of movement aligns with their line of sight is the proportion of the total number of pedestrians, reflecting the degree of inertia in how the crowd's line of sight guides their movement.

[0124] like A value close to 1 indicates that most pedestrians move in the same direction as their heads (direction of vision), and that crowd movement has strong visual guidance and high predictability.

[0125] like A value close to 0 indicates a significant deviation between the direction of crowd movement and the direction of line of sight, which may indicate confusion, following behavior, or external interference.

[0126] S230 uses visual inertial intensity, real-time crowd density, and environmental visibility as inputs to construct a visual inertial dynamics model. It uses parameters from the current crowd motion state database as initial conditions to perform numerical iterations over the next 5-10 minutes to predict the distribution changes of the crowd along the evacuation path.

[0127] Crowd density is described using a visually inertial modified convection-diffusion equation. Spatiotemporal evolution:

[0128] ;

[0129] in:

[0130] : Time coordinates Population density at the location;

[0131] : Rate of change of density over time;

[0132] Convection term dominated by visual inertia Visual inertial strength, For the group velocity field;

[0133] The crowd diffusion coefficient reflects the degree of random movement and is determined by the degree of environmental crowding.

[0134] : The Laplacian operator, used to describe diffusion caused by density gradient;

[0135] The coefficient of recovery reflects the population's tendency to adapt to the changing environment. The rate of approximation;

[0136] The maximum carrying capacity of the channel area is determined by the channel's physical parameters.

[0137] Solving this equation through numerical iteration yields the results for the next 5-10 minutes. The distribution of .

[0138] like Figure 4 As shown, the calculation of the real-time load factor specifically includes:

[0139] S310, based on real-time video streams and combined with building structure topology maps, identifies channel boundaries and entrances / exits, divides the scene into different evacuation channel areas, and calculates the real-time crowd density of each evacuation channel area.

[0140] S320 reads the physical parameters of each evacuation route area from the preset building information database, and calculates the real-time carrying capacity of each evacuation route area by combining the real-time population density.

[0141] Real-time load factor The ratio of the channel's actual load to its maximum load is given by the formula:

[0142] ;

[0143] in:

[0144] Real-time crowd density in the passageway area;

[0145] : The effective area of ​​the passageway;

[0146] Real-time number of people in the passageway;

[0147] The maximum number of people that the passageway can accommodate;

[0148] Real-time load factor (range) ).

[0149] like Figure 5 As shown, the generation of graded early warning signals specifically includes:

[0150] S410, based on the real-time population density and real-time carrying capacity of each evacuation route area, forms time series data of the real-time carrying capacity of each evacuation route area;

[0151] S420, Read the historical capacity threshold of each evacuation route area. The historical capacity threshold is the maximum carrying capacity when the route reaches full saturation, and set the optimal capacity threshold to 85% of the historical capacity threshold data.

[0152] The 85% threshold reserves a 15% capacity buffer to avoid congestion and stampede risks when the passage reaches 100% saturation, adhering to the safety margin principle of emergency evacuation. When the carrying capacity exceeds 85%, the flow rate of people will decrease significantly, and early warning can prevent people from reaching the congestion threshold. From the warning to the execution of diversion instructions and the response of the crowd, there is a time required (usually 30 seconds to 2 minutes), and the 85% threshold provides sufficient time window for emergency operations.

[0153] S430 uses pattern matching between real-time carrying capacity time series data and historical capacity thresholds to establish a path saturation time prediction model. Based on the current real-time carrying capacity and the predicted distribution changes of the population on the evacuation path, it simulates and extrapolates the future load growth of each evacuation channel area and calculates the time point when the real-time carrying capacity of each evacuation channel area reaches the optimal capacity threshold.

[0154] Assuming the channel capacity increases linearly with time, then:

[0155] ;

[0156] make (Optimal capacity threshold), solution yields arrival time ;

[0157] in:

[0158] : Real-time load factor at any given moment;

[0159] Initial time ( The carrying capacity of )

[0160] The carrying capacity growth rate is determined by the inflow of people and the density of upstream areas.

[0161] Time from the current moment until the load factor reaches 85%.

[0162] S440 generates a graded early warning signal based on the difference between the calculated time when the real-time load factor reaches the optimal capacity threshold and the current time. The warning level is divided according to the magnitude of the difference.

[0163] like Figure 6 As shown, the set of generation control instructions specifically includes:

[0164] S510: When a graded warning signal appears in any evacuation route area, combined with the distribution changes of the population on the evacuation route, based on the building structure topology map, the nearest path fork node is found upstream of the evacuation route area, and the coordinates of the fork node position are determined as the coordinates of the fork node position of the upstream path.

[0165] S520: Determine whether there are effective graded early warning signals in other evacuation channel areas guided by the location coordinates of the path branch node. If so, take the path branch node as a new starting point and continue to search for the next path branch node upstream along the building structure topology map, and repeat the judgment process. If not, determine the path branch node as the location coordinates of the branch node of the upstream path used for diversion.

[0166] If other channels pointed to by the branching node are already saturated (with a warning), diversion will cause secondary congestion. Therefore, it is necessary to find a branching point for an unsaturated channel upstream. Finding the nearest effective branching point upstream along the building topology can minimize the detour distance for the crowd and reduce diversion costs. Through iterative judgment, ensure that the finally selected branching point can cover enough upstream crowds and divert them to safe channels, which is in line with the evacuation strategy of source control.

[0167] S530, based on the time point when the real-time carrying rate in the graded early warning signal reaches the optimal capacity threshold, reverse the time when the guidance needs to be initiated, and consider the crowd response delay to calculate the optimal time window for the guidance to be initiated.

[0168] ;

[0169] Wherein: the start time of traffic diversion startup ;

[0170] The point in time when the channel's load factor reaches 85%;

[0171] Crowd response delay, which is the time from when pedestrians see the guidance signal to when they begin to move;

[0172] Command execution delay, 5-10 seconds, which is the time required for the device to start and for signals to be transmitted;

[0173] The time frame for traffic redirection startup must be within [the specified time frame]. to Completed in between.

[0174] S540 calculates the required diversion intensity parameters based on the difference between the real-time carrying capacity and the optimal capacity threshold, as well as the predicted population density level.

[0175] The flow guidance intensity parameter characterizes the strength of the flow guidance operation, taking into account the number of people requiring flow guidance, the acceptance capacity of the target channel, the urgency of time, and the signal transmission strength. It is used to quantify the scale of flow guidance and guide the output intensity of the equipment (high-frequency voice, high-brightness indicator light).

[0176] ;

[0177] in:

[0178] : Conductivity parameters;

[0179] : Remaining capacity of the target channel;

[0180] The current number of passengers exceeding the capacity of the passage ( );

[0181] Signal strength coefficient, levels 1-5, corresponding to guide light brightness, voice frequency, etc.

[0182] S550 encapsulates the bifurcation node coordinates of the upstream path, the optimal time window for diversion initiation, and diversion intensity parameters into a structured set of control commands and transmits them to the instruction execution terminal located at the bifurcation node coordinates.

[0183] Figure 7 The structural block diagram of the dynamic recommendation system for crowd evacuation paths based on visual data provided in the embodiments of the present invention is as follows: Figure 7 As shown, the system includes:

[0184] The video data extraction module 100 is used to collect real-time crowd movement video streams and preprocess them to obtain standardized video frames, extract the head orientation vector of each pedestrian, calculate and form a group movement velocity field, and establish a crowd movement state database.

[0185] The crowd prediction module 200 is used to establish a visual inertial dynamics model, calculate the angle distribution between the head orientation vector and the direction of movement, determine the visual inertial intensity parameters, and predict the distribution changes of the crowd on the evacuation path in the next 5-10 minutes.

[0186] The carrying capacity calculation module 300 is used to identify the areas of each evacuation route based on real-time video streams, calculate the real-time crowd density of each route, and calculate the real-time carrying capacity in combination with the physical parameters of the route.

[0187] The early warning generation module 400 is used to establish a path saturation time prediction model based on the real-time carrying capacity of each evacuation route and the route capacity threshold, calculate the time point when each evacuation route reaches the maximum saturation, and generate graded early warning signals.

[0188] The instruction transmission module 500 is used to determine the coordinates of the bifurcation node of the upstream path based on the distribution changes of the crowd on the evacuation path and the graded early warning signals when it is predicted that any evacuation channel will reach the optimal capacity threshold of 85%. It also calculates the optimal time window for diversion and diversion intensity parameters, generates a set of control instructions, and transmits them to the instruction execution end.

[0189] like Figure 8 As shown, the crowd prediction module 200 includes:

[0190] The angle calculation module 210 is used to calculate the angle between the head orientation vector of each pedestrian and the local motion direction corresponding to the pedestrian based on the head orientation vector and the group motion velocity field.

[0191] The angle distribution statistics module 220 is used to statistically analyze the distribution histogram of the angles of all pedestrians at the current time and within the past preset time window, and to calculate the visual inertial intensity.

[0192] The crowd evacuation flow prediction module 230 is used to construct a visual inertial dynamics model by taking visual inertial intensity, real-time crowd density and environmental visibility as inputs. It uses the parameters in the current crowd motion state database as initial conditions to perform numerical iterations for the next 5-10 minutes to predict the distribution changes of the crowd on the evacuation path.

[0193] like Figure 9 As shown, the early warning generation module 400 includes:

[0194] The carrying capacity time series generation module 410 is used to generate time series data of the real-time carrying capacity of each evacuation channel area based on the real-time population density and real-time carrying capacity of each evacuation channel area.

[0195] The capacity threshold setting module 420 is used to read the historical capacity threshold of each evacuation channel area. The historical capacity threshold is the maximum carrying capacity when the channel reaches full saturation, and the optimal capacity threshold is set to 85% of the historical capacity threshold data.

[0196] The threshold achievement time calculation module 430 is used to perform pattern matching between real-time carrying capacity time series data and historical capacity thresholds, establish a path saturation time prediction model, simulate and extrapolate the future load growth of each evacuation channel area based on the current real-time carrying capacity and the predicted distribution changes of the population on the evacuation path, and calculate the time point when the real-time carrying capacity of each evacuation channel area reaches the optimal capacity threshold.

[0197] The graded early warning signal generation module 440 is used to generate graded early warning signals based on the difference between the calculated time point when the real-time load rate reaches the optimal capacity threshold and the current time. The early warning level is divided according to the size of the difference.

[0198] like Figure 10 As shown, the instruction transmission module 500 includes:

[0199] The path branch node identification module 510 is used to find the nearest path branch node upstream of the evacuation channel area when a graded early warning signal appears in any evacuation channel area, combined with the distribution changes of the crowd on the evacuation path, based on the building structure topology map, and determine the location coordinates of the branch node as the location coordinates of the branch node of the upstream path.

[0200] The effective graded early warning signal judgment module 520 is used to determine whether there is an effective graded early warning signal in other evacuation channel areas guided by the location coordinates of the path branch node. If there is, the path branch node is used as a new starting point to continue searching for the next path branch node upstream along the building structure topology map, and the judgment process is repeated. If there is no, the path branch node is determined as the location coordinates of the branch node of the upstream path used for diversion.

[0201] The diversion start-up time calculation module 530 is used to calculate the optimal time window for diversion start-up by working backward from the time point when the real-time carrying rate in the graded early warning signal reaches the optimal capacity threshold, and taking into account the crowd response delay.

[0202] The flow intensity parameter calculation module 540 is used to calculate the required flow intensity parameters based on the difference between the real-time carrying capacity and the optimal capacity threshold, as well as the predicted population density level.

[0203] The control command encapsulation and transmission module 550 is used to encapsulate the bifurcation node position coordinates of the upstream path, the optimal time window for diversion initiation, and the diversion intensity parameters into a structured set of control commands, and transmit them to the instruction execution terminal located at the bifurcation node position coordinates.

[0204] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0205] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

[0206] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A dynamic recommendation method for crowd evacuation routes based on visual data, characterized in that, The method includes: Real-time acquisition of crowd movement video streams and preprocessing to obtain standardized video frames; extraction of head orientation vectors for each pedestrian; simultaneous calculation of the group movement velocity field; and establishment of a crowd movement state database. Establish a visual inertial dynamics model, calculate the angle distribution between the head orientation vector and the direction of movement, determine the visual inertial intensity parameters, and predict the distribution changes of the crowd on the evacuation path in the next 5-10 minutes. Based on real-time video streams, identify the areas of each evacuation route, calculate the real-time crowd density of each route, and calculate the real-time carrying capacity by combining the physical parameters of the route. Based on the real-time carrying capacity of each evacuation route, a path saturation time prediction model is established in combination with the route capacity threshold. The time point when each evacuation route reaches the maximum saturation is calculated, and a graded early warning signal is generated. When it is predicted that any evacuation route will reach 85% of its optimal capacity threshold, the coordinates of the bifurcation nodes of the upstream path are determined based on the distribution changes of the population on the evacuation route and the graded early warning signals. The optimal time window for diversion and the diversion intensity parameters are calculated, a set of control instructions is generated, and the instructions are transmitted to the execution end. The prediction of changes in the distribution of people along evacuation routes within the next 5-10 minutes specifically includes: Based on the head orientation vector and the group motion velocity field, the angle between the head orientation vector of each pedestrian and the corresponding local motion direction of the pedestrian is calculated. Statistically analyze the distribution histogram of the angles of all pedestrians at the current time and within the preset time window in the past, and calculate the visual inertia intensity; Using visual inertial intensity, real-time crowd density, and environmental visibility as inputs, a visual inertial dynamics model is constructed. The parameters in the current crowd motion state database are used as initial conditions, and numerical iterations are performed over the next 5-10 minutes to predict the distribution changes of the crowd along the evacuation path.

2. The method according to claim 1, characterized in that, The establishment of the population movement status database specifically includes: Continuously and in real time, the original video stream of crowd movement is acquired, and the received original video stream of crowd movement is preprocessed to generate standardized video frames. Standardized video frames are input into a pre-trained head pose estimation neural network to detect and locate the two-dimensional pixel coordinates of the head in the image, output the pitch angle, yaw angle and roll angle of each visible pedestrian head, and convert them into a head orientation vector in three-dimensional space. Based on a standardized video frame sequence, the motion vector of each pixel between adjacent frames is calculated. Based on the motion vectors of all pedestrian area pixels, a group motion velocity field representing the overall movement direction is formed. Establish a unified spatiotemporal coordinate system, synchronize the head orientation vector with the group's motion velocity field under the unified spatiotemporal coordinate system for time stamping and spatial alignment, and establish a database of crowd motion state containing timestamps, spatial positions, head orientation vectors and local motion vectors.

3. The method according to claim 2, characterized in that, The calculation of the real-time load factor specifically includes: Based on real-time video streams, combined with building structure topology maps, the system identifies passage boundaries and entrances / exits, divides the scene into different evacuation passage areas, and calculates the real-time crowd density in each evacuation passage area. The physical parameters of each evacuation route area are read from the pre-set building information database, and the real-time carrying capacity of each evacuation route area is calculated by combining the real-time population density.

4. The method according to claim 3, characterized in that, The generation of graded early warning signals specifically includes: Based on the real-time population density and real-time carrying capacity of each evacuation route area, time series data of the real-time carrying capacity of each evacuation route area are generated; Read the historical capacity threshold of each evacuation route area. The historical capacity threshold is the maximum carrying capacity when the route reaches full saturation. Set the optimal capacity threshold to 85% of the historical capacity threshold data. By performing pattern matching between real-time carrying capacity time series data and historical capacity thresholds, a path saturation time prediction model is established. Based on the current real-time carrying capacity and the predicted distribution changes of the population on the evacuation path, the future load growth of each evacuation channel area is simulated and deduced, and the time point when the real-time carrying capacity of each evacuation channel area reaches the optimal capacity threshold is calculated. Based on the difference between the calculated time when the real-time load factor reaches the optimal capacity threshold and the current time, a tiered early warning signal is generated, with the warning level determined by the magnitude of the difference.

5. The method according to claim 4, characterized in that, The set of generation control instructions specifically includes: When a graded early warning signal appears in any evacuation route area, combined with the distribution changes of people on the evacuation route, based on the building structure topology map, the nearest path fork node is found upstream of the evacuation route area, and the coordinates of the fork node are determined as the coordinates of the fork node of the upstream path. Determine whether there are effective graded early warning signals in other evacuation channel areas guided by the location coordinates of the path fork node. If there are, take the path fork node as a new starting point and continue to search for the next path fork node upstream along the building structure topology map, and repeat the judgment process; if there are no, determine the path fork node location coordinates as the fork node location of the upstream path used for diversion. Based on the time point when the real-time carrying rate in the graded early warning signal reaches the optimal capacity threshold, the deadline for initiating guidance is calculated by working backwards, and the crowd response delay is taken into account to calculate the optimal time window for initiating guidance. The required diversion intensity parameters are calculated based on the difference between the real-time carrying capacity and the optimal capacity threshold, as well as the predicted population density level. The coordinates of the bifurcation node of the upstream path, the optimal time window for diversion initiation, and the diversion intensity parameters are encapsulated into a structured set of control instructions and transmitted to the instruction execution terminal located at the coordinates of the bifurcation node.

6. A dynamic recommendation system for crowd evacuation routes based on visual data, characterized in that, The system includes: The video data extraction module is used to collect real-time video streams of crowd movement and preprocess them to obtain standardized video frames, extract the head orientation vector of each pedestrian, calculate and form a group movement velocity field, and establish a crowd movement state database. The crowd prediction module is used to establish a visual inertial dynamics model, calculate the angle distribution between the head orientation vector and the direction of movement, determine the visual inertial intensity parameters, and predict the distribution changes of the crowd on the evacuation path in the next 5-10 minutes. The carrying capacity calculation module is used to identify the areas of each evacuation route based on real-time video streams, calculate the real-time crowd density of each route, and calculate the real-time carrying capacity in combination with the physical parameters of the route. The early warning generation module is used to establish a path saturation time prediction model based on the real-time carrying capacity rate of each evacuation route and the route capacity threshold, calculate the time point when each evacuation route reaches the maximum saturation, and generate graded early warning signals. The instruction transmission module is used to determine the coordinates of the bifurcation node of the upstream path based on the distribution changes of the crowd on the evacuation path and the graded early warning signals when it is predicted that any evacuation channel will reach the optimal capacity threshold of 85%. It also calculates the optimal time window and diversion intensity parameters for diversion initiation, generates a set of control instructions, and transmits them to the instruction execution end. The population prediction module includes: The angle calculation module is used to calculate the angle between the head orientation vector of each pedestrian and the local motion direction corresponding to the pedestrian, based on the head orientation vector and the group motion velocity field. The angle distribution statistics module is used to statistically analyze the distribution histogram of the angles of all pedestrians at the current time and within a preset time window in the past, and to calculate the visual inertial intensity. The crowd evacuation flow prediction module is used to construct a visual inertial dynamics model by taking visual inertial intensity, real-time crowd density and environmental visibility as inputs. It uses the parameters in the current crowd motion state database as initial conditions and performs numerical iterations over the next 5-10 minutes to predict the distribution changes of the crowd along the evacuation path.

7. The system according to claim 6, characterized in that, The early warning generation module includes: The carrying capacity time series generation module is used to generate time series data of the real-time carrying capacity of each evacuation route area based on the real-time population density and real-time carrying capacity of each evacuation route area; The capacity threshold setting module is used to read the historical capacity threshold of each evacuation route area. The historical capacity threshold is the maximum carrying capacity when the route reaches full saturation, and the optimal capacity threshold is set to 85% of the historical capacity threshold data. The threshold achievement time calculation module is used to perform pattern matching between real-time carrying capacity time series data and historical capacity thresholds, establish a path saturation time prediction model, simulate and extrapolate the future load growth of each evacuation channel area based on the current real-time carrying capacity and the predicted distribution changes of the population on the evacuation path, and calculate the time point when the real-time carrying capacity of each evacuation channel area reaches the optimal capacity threshold. The graded early warning signal generation module is used to generate graded early warning signals based on the difference between the calculated time point when the real-time load rate reaches the optimal capacity threshold and the current time. The early warning level is divided according to the magnitude of the difference.

8. The system according to claim 7, characterized in that, The instruction transmission module includes: The path bifurcation node identification module is used to find the nearest path bifurcation node upstream of any evacuation route area when a graded early warning signal appears in any evacuation route area, combined with the distribution changes of people on the evacuation route, based on the building structure topology map, and determine the bifurcation node location coordinates as the bifurcation node location coordinates of the upstream path. The effective graded early warning signal judgment module is used to determine whether there are effective graded early warning signals in other evacuation channel areas guided by the location coordinates of the path branch node. If there are, the path branch node is used as a new starting point to continue searching for the next path branch node upstream along the building structure topology map, and the judgment process is repeated. If there are no, the path branch node is determined as the location coordinates of the branch node of the upstream path used for diversion. The diversion start-up time calculation module is used to calculate the optimal time window for diversion start-up by working backward from the time point when the real-time carrying rate in the graded early warning signal reaches the optimal capacity threshold, and taking into account the crowd response delay. The flow guidance intensity parameter calculation module is used to calculate the required flow guidance intensity parameters based on the difference between the real-time carrying capacity and the optimal capacity threshold, as well as the predicted population density level. The control command encapsulation and transmission module is used to encapsulate the bifurcation node position coordinates of the upstream path, the optimal time window for diversion initiation, and the diversion intensity parameters into a structured set of control commands, and transmit them to the instruction execution terminal located at the bifurcation node position coordinates.

Images