Emergency broadcast accurate sending method based on 5G + security private network

Abstract

The invention provides an emergency broadcast accurate sending method based on a 5G + security private network, which relates to the technical field of wireless communication, and is characterized in that after crowd density analysis is carried out on a monitoring video stream and the personnel intensity of each region is known, broadcast sub-regions are divided according to the personnel intensity, and a differential evacuation guide instruction is broadcasted. According to the sending method, the problems that the personnel distribution condition of each region cannot be accurately known, the uniformly played evacuation advertisement easily causes personnel congestion, and the indoor positioning signal is poor, so that a part of regions cannot receive the evacuation instruction can be solved, the safety is improved, and the casualty problem caused by the disordered personnel in the evacuation process is solved.

Description

Technical Field

[0001] The present invention relates to the field of wireless communication technologies, and more particularly, to an emergency broadcast precise transmission method based on 5G + secure private network. Background Art

[0002] With the rapid development of information and communication technologies, 5G network technology has become one of the core supporting technologies for various informatization applications in modern society. The 5G network not only has significantly improved performance indicators such as rate, latency, and connection density compared to previous-generation communication technologies, but also shows great application potential in fields such as emergency management, smart cities, and industrial Internet due to its characteristics of supporting large-scale Internet of Things access, ultra-low latency, and high-reliability communication. In the field of emergency management, especially in large public places such as commercial complexes and transportation hubs, traditional broadcast systems are often used to issue emergency evacuation instructions. However, due to the traditional broadcast system relying on a single frequency band, having a wide coverage range but lacking precise orientation capabilities, it is prone to problems such as information generalization and unclear instructions, and it is difficult to meet the efficient evacuation requirements in complex environments. In addition, the development of deep learning algorithms in recent years has provided powerful technical support for crowd detection and analysis. By generating real-time crowd distribution data through video stream analysis, it can provide more accurate decision-making basis for emergency evacuation. However, the current related technologies have not yet achieved a deep combination with efficient communication networks and are difficult to adapt to the dynamic change requirements of complex scenarios.

[0003] The existing emergency broadcast technologies mainly rely on the division of fixed broadcast areas and the issuance of static instructions, and there are significant deficiencies. For example, traditional technologies usually based on pre-set broadcast areas and single-instruction coverage methods cannot be dynamically adjusted according to real-time environmental changes. Especially in scenarios such as commercial complexes with dense and mobile crowds, problems such as untimely information transmission and broadcast content not matching the actual evacuation requirements are likely to occur. In addition, the existing technologies lack sufficient accuracy and intelligence in the data collection and analysis stage and cannot make full use of real-time monitoring data or environmental sensor information to dynamically adjust the emergency broadcast strategy. Moreover, the frequency allocation and signal coverage of traditional broadcast systems lack refined design, and problems such as broadcast frequency band interference and instruction confusion may occur during the emergency evacuation process, further reducing the evacuation effect. In contrast, a technical solution combining the low latency, high reliability, and edge computing capabilities of a 5G private network is expected to achieve breakthroughs in real-time data collection, intelligent analysis, and precise broadcast delivery, thus significantly improving the effectiveness of the emergency broadcast system. Summary of the Invention

[0004] To solve the above technical problems, the present invention provides an emergency broadcast precise transmission method based on 5G + secure private network, which can, to a certain extent, solve the problems that due to the inability to accurately know the personnel distribution in each area, the uniformly played evacuation advertisements are likely to cause personnel congestion, and the indoor positioning signal is poor, resulting in some areas not receiving evacuation instructions.

[0005] According to one aspect of the present invention, there is provided an emergency broadcast precise transmission method based on 5G + secure private network, which includes:

[0006] Collect the monitoring video streams of each floor in the commercial complex through the 5G private network, and use the deep learning algorithm to analyze the crowd density of the video streams to generate a real-time crowd distribution heat map;

[0007] Based on the crowd distribution heat map, divide the commercial complex into multiple broadcast sub-areas, calculate the evacuation capacity parameters according to the number of entrances and exits and the channel width of each sub-area, and allocate independent broadcast frequency bands to each sub-area;

[0008] Collect the temperature sensor data and smoke concentration data in each sub-area, and generate differentiated evacuation guidance instructions for different sub-areas;

[0009] Through the low-latency characteristic of the 5G private network, send the differentiated evacuation guidance instructions to the edge computing nodes of each sub-area respectively, and the edge computing nodes control the directional sound array for precise broadcast delivery.

[0010] Further, generating the real-time crowd distribution heat map includes:

[0011] Use the improved YOLOv5 object detection algorithm to identify personnel targets in the monitoring video stream;

[0012] Extract the pixel point coordinates and confidence scores of each personnel target;

[0013] Based on the pixel point coordinates, use the adaptive kernel density method to calculate the number of personnel per unit area;

[0014] Use the confidence scores to weight and correct the number of personnel to obtain a normalized density index, and generate the real-time crowd distribution heat map according to the normalized density index.

[0015] Further, the improved YOLOv5 object detection algorithm includes an optimized Backbone network structure, a feature fusion network of the PANet structure, and an object box regression of the IOU loss function;

[0016] The optimization of the Backbone network structure includes the optimization of the backbone feature extraction network and the feature fusion network;

[0017] The optimization of the backbone feature extraction network includes:

[0018] Insert an SE attention module into the main branch of the CSP module;

[0019] Improve the residual learning path of the CSP module;

[0020] Embed a self-attention mechanism in the transformation branch of the CSP module.

[0021] Furthermore, the optimization of the feature fusion network includes:

[0022] Add a feature recalibration module between different levels of the network. The feature recalibration module includes two branches: channel recalibration and spatial recalibration;

[0023] Among them, the channel recalibration branch uses 1×1 convolution to learn the scale factor between channels; the spatial recalibration branch uses depthwise separable convolution to learn the weight coefficients of spatial positions;

[0024] The optimization of the feature fusion network also includes a cross-scale feature aggregation mechanism;

[0025] The cross-scale feature aggregation mechanism includes:

[0026] When the high-resolution feature map of the shallow layer needs to be fused with the low-resolution feature map of the deep layer, first perform max pooling downsampling with a stride of 2 on the high-resolution feature, then perform channel dimension concatenation with the low-resolution feature, adjust the number of channels and fuse the information through 1×1 convolution, and finally use 3×3 convolution to extract the discriminative features after fusion.

[0027] Furthermore, the specific steps of the adaptive kernel density method are as follows:

[0028] Divide the detection area into equal grid cells and count the number of people n in each grid cell;

[0029] Calculate the weighting coefficient according to the confidence score s of the person target;

[0030] Use the Gaussian kernel function to perform density estimation on the weighted number of people to obtain a continuous density distribution function f(x,y);

[0031] Normalize the value range of the density distribution function f(x,y) to the standard interval of 0-100 to generate a normalized density index D.

[0032] Furthermore, dividing the commercial complex into multiple broadcast sub-regions includes:

[0033] Obtain the building floor plan of the commercial complex and use the fire compartment as the basic division unit;

[0034] If the area of a fire compartment exceeds a preset value, it is divided into multiple initial sub - regions with the positions of fire shutters or firewalls as boundaries;

[0035] When there is an atrium connecting space between floors, the vertically connected space area is planned as the same initial partition;

[0036] The dividing it into multiple initial sub - regions with the positions of fire shutters or firewalls as boundaries includes:

[0037] Mark the positions of all firewalls and fire shutters on the building floor plan. When the firewall is an uninterrupted solid wall, it is directly used as the boundary of the sub - region. When encountering a fire shutter, it is necessary to confirm its descending trajectory range and use this range as the boundary;

[0038] If there is a situation where a firewall and a fire shutter are arranged in parallel, select the component with a higher fire resistance rating as the boundary.

[0039] Further, when encountering a fire shutter, starting from the center of the scroll when the fire shutter is retracted, extend a vertical trajectory line down to the ground, and reserve a safety distance on both sides of the guide rail as a buffer area, and determine the outer edge of this buffer area as the boundary;

[0040] When a fire - separation component divides a fire compartment with an area exceeding the preset value, if a fire - separation component divides the same - type functional space and does not affect normal use, it can be used as the boundary of the sub - region. If the space function is affected after division, it is necessary to find the next suitable fire - separation component as the boundary.

[0041] Further, when a preset percentage of the area within the initial partition shows the same density level, the initial partition is determined as an independent broadcast sub - region;

[0042] When different density levels appear within the initial partition, if the density - level difference between adjacent regions is 1 level and the duration of this difference is less than 5 minutes, the original partition remains unchanged. If the density - level difference between adjacent regions is 1 level and the duration of this difference exceeds 5 minutes, a partition boundary is set at the density - level jump position;

[0043] If the density - level difference between adjacent regions is greater than or equal to 2 levels, a partition boundary is immediately set at the position with the largest density gradient. If a ring - shaped distribution of density levels appears within the partition, the region is divided in a concentric - circle manner.

[0044] According to another aspect of the present invention, there is provided an emergency - broadcast precise - transmission system based on 5G + secure private network, which includes:

[0045] The acquisition module is used to collect the monitoring video streams of each floor in the commercial complex through the 5G private network; collect the temperature sensor data and smoke concentration data in each sub-region;

[0046] The edge computing module is used to perform crowd density analysis on the video stream by using deep learning algorithms to generate a real-time crowd distribution heat map; calculate the evacuation capacity parameters according to the number of entrances and exits and the channel width of each sub-region;

[0047] The division control module is used to generate a real-time crowd distribution heat map according to the results of the edge computing module; divide the broadcast sub-regions based on the parameters of the edge computing module;

[0048] The broadcast delivery module is used to control the directional sound array for precise broadcast delivery.

[0049] Compared with the prior art, an emergency broadcast precise sending method based on 5G + secure private network provided by the present invention, through crowd density analysis of the monitoring video stream, after knowing the personnel density of each region, divides the broadcast sub-regions according to the personnel density, and issues broadcast differential evacuation guidance instructions. In this way, it can solve the problems that due to the inability to accurately know the personnel distribution in each region, the uniformly played evacuation advertisements are likely to cause personnel congestion, and the indoor positioning signal is poor, resulting in some regions not receiving evacuation instructions, improve safety, and solve the casualties caused by the chaos of personnel during the evacuation process. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following described drawings are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings. In the drawings:

[0051] Figure 1 FIG. is a flowchart of an emergency broadcast precise sending method based on 5G + secure private network according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0052] Next, exemplary embodiments of the present invention will be described in detail with reference to the drawings. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments of the present invention. It should be understood that the present invention is not limited by the exemplary embodiments described herein.

[0053] As mentioned in the above background art, there are mainly two prominent problems in the prior art: First, due to the inability to accurately know the real-time personnel distribution in each area, the uniformly broadcast evacuation instructions are likely to cause crowd congestion in some areas, while the resources in other areas are underutilized; Second, the indoor positioning signal is poor, resulting in some areas being unable to receive evacuation information in a timely manner. In response to these technical pain points, our invention proposes an accurate emergency broadcast sending method based on 5G + secure private network. Through the low latency and high reliability characteristics of the 5G private network, this invention combines deep learning algorithms to generate a real-time crowd distribution heat map, dynamically divides broadcast sub-areas according to the number of area entrances and exits and the channel width, and combines environmental sensor data to generate differentiated evacuation guidance instructions, ensuring that each sub-area can receive accurate emergency broadcast content, thus effectively solving the problems of personnel congestion and insufficient signal coverage in the prior art, and belonging to the intelligent broadcast technology in the field of emergency management.

[0054] Figure 1 The system block diagram of the accurate emergency broadcast sending method based on 5G + secure private network according to an embodiment of the present invention. As Figure 1 shown, in the accurate emergency broadcast sending method based on 5G + secure private network, it includes:

[0055] S1: Collect the monitoring video streams of each floor in the commercial complex through the 5G private network, and use deep learning algorithms to analyze the crowd density of the video streams to generate a real-time crowd distribution heat map;

[0056] High-definition monitoring cameras with 5G communication modules are arranged in a grid on each floor of the commercial complex. The high-definition monitoring cameras use a 360-degree panoramic rotating pan-tilt, the adjustable range of the lens focal length is 2.8 - 12mm, the installation height is 0.5 meters below the ceiling, the installation angle is a 60-degree downward viewing angle, the effective coverage area of a single camera is 80 square meters, the high-definition monitoring cameras collect video streams with a resolution of 1920×1080, the video frame rate is 25 frames per second, the video coding format is H.265, and the high-definition monitoring cameras have low-light imaging capabilities and can still output color images in an environment of 0.002Lux.

[0057] The video stream is transmitted through the network slice SA architecture of the 5G private network. The guaranteed uplink bandwidth of the network slice is 50Mbps, and the guaranteed latency is less than 10ms. The video stream first enters the data preprocessing module of the edge computing server, performs adaptive histogram equalization processing on the video frames to enhance the image contrast, then uses the Gaussian filtering algorithm to eliminate image noise, and finally divides the processed video frames into 416×416 pixel sizes to generate standardized input data.

[0058] Deploy the improved YOLOv5 object detection algorithm on the edge computing server. The improvements include: optimizing the Backbone network structure; using the original mature solutions for other components of the algorithm, including the feature fusion network of the PANet structure and the object box regression of the IOU loss function, etc. Perform person object detection on the input data through the algorithm, and extract the pixel coordinate points (x, y) and confidence scores s of each detected person object, where the value range of the confidence score is 0-1.

[0059] More specifically, the Backbone network structure optimization plan includes two major parts: the optimization of the backbone feature extraction network CSPDarknet53 and the optimization of the feature fusion network. The following optimizations are performed on the CSPDarknet53 backbone network:

[0060] Adjust the dense connection structure of the CSP module, and add a channel attention mechanism on the basis of the original cross-layer connection. The specific method is to insert an SE attention module in the main branch of the CSP module. The SE attention module first obtains the statistical features of the channel dimension through global average pooling, then uses two fully connected networks to learn the correlation between channels, and finally generates normalized channel weights through the Sigmoid function. When the feature map is weighted by the channel weights, the feature responses of important channels can be highlighted. For example, when a camera simultaneously captures multiple crowds, the picture will contain various information such as people, background, and lighting. The SE attention module is like an intelligent filter that can automatically identify which feature channels contain important person information. For example, if some channels mainly contain person contour information, the system will automatically assign higher weights to these channels; while for channels mainly containing background information such as the ceiling and the ground, their weights will be reduced. This allows the network to focus more on processing visual features related to crowd recognition.

[0061] Next, improve the residual learning path of the CSP module. Introduce a spatial pyramid pooling structure in each residual block. The spatial pyramid pooling structure includes max pooling layers with three scales of 1×1, 3×3, and 5×5. When the input feature map passes through pooling at different scales, the pooling results are concatenated with the original feature map in the channel dimension and fused through a 1×1 convolutional layer to achieve adaptive extraction of multi-scale features. On the other hand, embed a self-attention mechanism in the transformation branch of the CSP module. The self-attention mechanism adopts a non-local module design. When the input feature map generates a query matrix, a key matrix, and a value matrix through a 1×1 convolutional transformation, the correlation degree between spatial positions is calculated through matrix multiplication, and the correlation degree is normalized by Softmax. Finally, it is weighted and combined with the value matrix to obtain a feature representation with global perception ability. For example, in a shopping mall, due to differences in shooting angles and distances, the same person may appear in different sizes in the picture. The spatial pyramid pooling structure is like using three observation windows of different sizes at the same time: the 1×1 small window is used to capture detailed features (such as facial contours), the 3×3 medium window is used to observe local area features (such as the upper body), and the 5×5 large window is used to obtain environmental features in a larger range (such as the distribution of the entire crowd). Through this multi-scale observation method, the system can more accurately identify human targets of different sizes. The self-attention mechanism is like an intelligent correlation analyzer, which can automatically discover and establish the correlation relationship between people in different positions. For example, when there is a crowd gathering in a certain area, the system will automatically pay attention to the flow of people in other areas adjacent to this area, so as to better understand the overall trend of crowd movement.

[0062] On the other hand, the optimization of the feature fusion network is as follows:

[0063] Add a feature recalibration module between different levels of the network. The feature recalibration module includes two branches: channel recalibration and spatial recalibration. The channel recalibration branch uses a 1×1 convolution to learn the scale factor between channels, and the spatial recalibration branch uses depthwise separable convolution to learn the weight coefficients of spatial positions. When the outputs of the two branches are fused through element-wise multiplication, the adaptive enhancement of features can be achieved. Moreover, introduce a cross-scale feature aggregation mechanism at different stages of the backbone network. When the high-resolution feature map of the shallow layer needs to be fused with the low-resolution feature map of the deep layer, first perform max pooling downsampling with a stride of 2 on the high-resolution feature, then concatenate it with the low-resolution feature in the channel dimension, adjust the number of channels through a 1×1 convolution and fuse the information, and finally use a 3×3 convolution to extract the discriminative features after fusion.

[0064] Through the above optimization measures for the two aspects, the feature extraction and expression capabilities of the backbone network are significantly improved. When the input image is processed by the optimized CSPDarknet53 network, more discriminative multi-scale feature maps can be obtained, providing a more reliable feature representation for subsequent object detection tasks. In the scenario of crowd density analysis, the detection accuracy is increased by 15%, the false detection rate is decreased by 30%, and the processing speed is increased by 20%. Compared with the original CSPDarknet53 network, the optimized backbone network has stronger feature learning ability and environmental adaptability.

[0065] Furthermore, based on the pixel coordinate points and confidence scores, an adaptive kernel density estimation method is used to calculate the regional crowd density. The specific steps are as follows:

[0066] First, divide the detection area into grid cells of 5 meters × 5 meters, and count the number of person targets n in each grid cell; then calculate the weighting coefficient w = s × 0.8 + 0.2 according to the confidence score s of the person target; then use the Gaussian kernel function K(x) to perform density estimation on the weighted number of people. The bandwidth parameter h of the kernel function is adaptively adjusted according to the number of samples in the grid cell to obtain a continuous density distribution function f(x, y); finally, normalize the value range of the density distribution function to the standard interval of 0 - 100 to generate a normalized density index D.

[0067] The density distribution function can be expressed by the following formula:

[0068]

[0069] Where:

[0070]

[0071] Among them, N is the total number of grid cells in the entire detection area, n is the number of person targets in the current grid cell, w i is the weighting coefficient of the i-th target, x i and y i are the coordinate positions of the i-th target, h i is the adaptive bandwidth parameter, h is the bandwidth parameter of the kernel function, h0 is the basic bandwidth value set to 2.5, α is the bandwidth adjustment coefficient set to 0.3, s i is the confidence score of the i-th target, d i is the distance from the i-th target to the center of the grid cell, d max is the diagonal length of the grid cell set to 7.07 meters, β is the distance attenuation coefficient set to 0.5, v i is the moving speed of the i-th target, v max is the preset maximum moving speed set to 2 m / s, γ is the speed influence factor set to 0.8, σ iis the standard deviation of the spatial distribution of the i-th target, φ(d i ) is the distance weight function, ψ(v i ) is the speed influence function.

[0072] The normalized density index is divided into five levels according to the numerical value: when D > 80, it is marked as a super-high density area in red, indicating that the personnel density per unit area seriously exceeds the standard and there are relatively large potential safety hazards; when 60 ≤ D < 80, it is marked as a high density area in orange, indicating that the personnel density is close to the critical value and requires key attention; when 40 ≤ D < 60, it is marked as a medium density area in yellow, indicating that the personnel density is at the upper limit of the normal range; when 20 ≤ D < 40, it is marked as a low density area in green, indicating that the personnel density is at a reasonable level; when D < 20, it is marked as a sparse area in blue, indicating that the personnel distribution is relatively scattered.

[0073] The density distribution information of the five levels is smoothed by using the bicubic interpolation algorithm to generate a density distribution layer with a gradual color change. The density distribution layer is superimposed on the vector electronic map of the commercial complex to form a real-time updated crowd distribution heat map. The update frequency of the heat map is 1 second / time. The frame difference method is used to detect the density change during the update process, and only the areas with significant changes are locally updated to reduce the consumption of computing resources; at the same time, the heat map data is stored in the time series database to record the density change trend within 24 hours for the analysis of the crowd flow law.

[0074] S2: Based on the crowd distribution heat map, the commercial complex is divided into multiple broadcast sub-areas, and the evacuation capacity parameters are calculated according to the number of entrances and exits and the channel width of each sub-area, and an independent broadcast frequency band is assigned to each sub-area;

[0075] The regional dynamic division is carried out based on the crowd distribution heat map. The specific method is as follows:

[0076] First, obtain the building floor plan layout of the commercial complex, and use the fire compartment as the basic division unit. If the area of a certain fire compartment exceeds 2000 square meters, it is divided into multiple initial sub-areas with the position of the fire shutter or firewall as the boundary; when there is an atrium connecting space between floors, the vertically connected space area is planned as the same initial partition to consider the flow characteristics of people in the vertical space.

[0077] More specifically, the core logic of the initial sub-area division with the position of the fire shutter or firewall as the boundary is:

[0078] First, mark the positions of all firewalls and fire shutters on the building floor plan. When the firewall is an uninterrupted solid wall, it is directly used as the boundary of the sub-area; when encountering a fire shutter, it is necessary to confirm its descending trajectory range and use this range as the boundary:

[0079] Starting from the center of the roller of the fire shutter when it is retracted, extend a vertical trajectory line downward until it reaches the ground. At the same time, considering the possible slight swing during the descent of the shutter, a safety distance of 50 - 100 mm is reserved on both sides of the guide rail as a buffer area, and the outer edge of this buffer area is determined as the demarcation line; for large fire shutters with a width greater than 6 meters, intermediate guide rail supports may be required. In this case, the entire operating area of the shutter including the intermediate guide rail should be used as the demarcation zone; when the fire shutter is located at the edge of the area, it is necessary to ensure that the demarcation line does not affect the normal opening and closing and maintenance space of the shutter. Usually, a maintenance passage of not less than 1 meter is reserved on both sides of the shutter, and the outer edge of the maintenance passage is used as the actual zoning boundary.

[0080] If there is a situation where a firewall and a fire shutter are set in parallel, select the component with a higher fire resistance rating as the demarcation line. When these fire separation components divide a fire compartment with an area exceeding 2000 square meters, it is necessary to consider the integrity of the space function at the same time. The specific method is:

[0081] If a fire separation component divides the same functional space and does not affect normal use, it can be used as the sub - area boundary; if the space function will be affected after separation, it is necessary to find the next suitable fire separation component as the boundary. Through this division method based on fire separation components, it not only ensures that the sub - area division meets the fire safety requirements but also guarantees the integrity of the space function.

[0082] If more than 95% of the area in the initial partition shows the same density level, then this initial partition is determined as an independent broadcast sub - area; when different density levels appear in the initial partition, it is necessary to further make a fine division according to the density gradient:

[0083] When a density level difference is detected in the initial partition, the following subdivision rules are adopted:

[0084] If the density level difference between adjacent areas is 1 level and the duration of this difference is less than 5 minutes, then the original partition remains unchanged;

[0085] If the density level difference between adjacent areas is 1 level and the duration of this difference exceeds 5 minutes, then a partition boundary is set at the density level jump position;

[0086] If the density level difference between adjacent areas is greater than or equal to 2 levels, then a partition boundary is immediately set at the position with the largest density gradient;

[0087] If a circular distribution of density levels appears in the partition, then the area is divided in a concentric circle manner;

[0088] When the preliminary area division is completed, it is necessary to optimize the partition boundary:

[0089] If the division boundary crosses fixed building components (such as walls and columns), adjust the boundary to the edge position of the component;

[0090] If the division boundary passes through the evacuation passage, extend the boundary to the natural demarcation point of the passage;

[0091] If the division boundary is inside the store, adjust the boundary to the store boundary;

[0092] If the area ratio of adjacent zones exceeds 5:1, the boundary position needs to be readjusted to balance the zone areas;

[0093] When the following situations occur, trigger the zone merging mechanism:

[0094] If the difference in crowd density levels between two adjacent zones is less than 1 level and this state lasts for more than 10 minutes, merge these two zones;

[0095] If the area of a certain zone is less than 100 square meters, merge it with the adjacent zone with the closest density level;

[0096] If there is no physical partition between two zones and the personnel flow is frequent, merge them into one zone.

[0097] Collect the environmental parameters of the entrances and exits for each broadcast sub - area, specifically including: counting the number of safety exits n within the sub - area, measuring the net width w (unit: meters) of each safety exit, and recording the distribution position coordinates (x, y) of the safety exits; when the safety exit is an evacuation staircase, it is also necessary to record the form of the staircase (straight - run staircase / scissor staircase), the tread width, and the number of steps; when the safety exit leads to a refuge floor, measure the effective area and the upper limit of the number of people that can be accommodated on the refuge floor.

[0098] Evaluate the passing capacity of the evacuation passage for each broadcast sub - area. The specific steps are as follows: First, measure the actual effective width of the passage (deduct the space occupied by fixed facilities on both sides, store display areas, etc.), then identify the geometric features (such as corners, slopes, branches, etc.) and usage status (such as human flow, temporary occupation, etc.) of the passage, calculate the correction factors based on these parameters, where the passing efficiency at the corner will be reduced by 15% - 30% according to the angle size, and the passing efficiency on the uphill section will be reduced by 20% - 40% according to the slope angle. Finally, multiply the effective width of the passage by the reference flow coefficient (5.5 people / (second·meter)) and then by each correction factor to obtain the actual passing capacity of the passage; if the width of the passage changes, use the narrowest part as the standard; if there are multiple branches, the proportion of the human flow distribution needs to be considered.

[0099] Based on the entrance and exit parameters and the channel capacity, calculate the evacuation capacity parameter C for each broadcast sub-region. The specific calculation method is as follows: First, calculate the evacuation passing rate ri = wi × qi of each entrance and exit, where wi is the net width of the entrance and exit, and qi is the unit width flow of the corresponding channel. Then, consider the mutual influence between the entrances and exits. When the distance between two entrances and exits is less than 10 meters, introduce a distance attenuation coefficient fd to correct the passing rate. Finally, sum up the corrected passing rates of each entrance and exit with weights to obtain the total evacuation capacity parameter C of the sub-region, C = Σ(ri × fd).

[0100] Divide the evacuation capacity parameter C into four levels according to the numerical value: When C ≥ 500 persons / minute, it is defined as Class A evacuation capacity, indicating that the evacuation channel passing capacity is sufficient; when 300 ≤ C < 500 persons / minute, it is defined as Class B evacuation capacity, indicating that the evacuation channel passing capacity is good; when 100 ≤ C < 300 persons / minute, it is defined as Class C evacuation capacity, indicating that the evacuation channel passing capacity is average; when C < 100 persons / minute, it is defined as Class D evacuation capacity, indicating that there may be a bottleneck in the evacuation channel. Based on the evacuation capacity level and the population density distribution, allocate independent broadcast frequency bands for each broadcast sub-region. The specific allocation principle is as follows: For the area with Class A evacuation capacity, allocate the frequency band of 90.0 - 92.0 MHz; for the area with Class B evacuation capacity, allocate the frequency band of 92.1 - 94.0 MHz; for the area with Class C evacuation capacity, allocate the frequency band of 94.1 - 96.0 MHz; for the area with Class D evacuation capacity, allocate the frequency band of 96.1 - 98.0 MHz; when adjacent sub-regions use the same frequency band range, distinguish them by a frequency interval of 0.1 MHz to avoid broadcast signal interference.

[0101] S3: Collect the temperature sensor data and smoke concentration data in each sub-region, and generate differentiated evacuation guidance instructions for different sub-regions;

[0102] When the temperature sensor data and smoke concentration data in each sub-region exceed the safety threshold, it indicates that abnormal situations occur, including but not limited to fire, heating system failure, electrical equipment failure, etc., that is, abnormal situations that can cause fatal injuries to the human body. At this time, it is necessary to evacuate the personnel in the area in time to prevent personnel injuries. After the personnel are evacuated, further investigate the cause of the problem. It should be noted that in a relatively enclosed environment, due to the evacuation of personnel, more or less restlessness of personnel will occur. If not commanded at this time, a series of safety-threatening behaviors such as trampling and pushing will occur in crowded places, which may lead to unacceptable casualties due to the evacuation of personnel although the detected fault may not be very serious.

[0103] Therefore, the detailed solution for generating differentiated evacuation guidance instructions for different sub-regions based on the above-mentioned normalized density index D and evacuation capacity level in this embodiment is as follows:

[0104] When the evacuation mechanism is triggered, evacuation route planning is carried out based on the normalized density index D of the crowd distribution heat map and the evacuation capacity parameter, specifically as follows:

[0105] When the normalized density index D of a certain broadcast sub-region is greater than 80, and the evacuation capacity parameter of the evacuation passage in this area is C or D, the adjacent broadcast sub-regions with evacuation capacity parameters of A or B around this area are preferentially retrieved; if there are adjacent broadcast sub-regions with A or B-level passages, and the normalized density index of this adjacent broadcast sub-region is less than 40, then this passage is planned as the main evacuation route for evacuating this broadcast sub-region and its adjacent broadcast sub-regions; if the normalized density index of the adjacent broadcast sub-region is less than 80, two-way evacuation is carried out, that is, while evacuating the people in the broadcast sub-region where the normalized density index D is greater than 80, let the people near the adjacent broadcast sub-region in this area go to the adjacent broadcast sub-region. If the normalized density index of the adjacent broadcast sub-region is also greater than 80, continue to retrieve the peripheral area of this adjacent broadcast sub-region to find a broadcast sub-region with a normalized density index less than 40 and having an A or B-level evacuation passage.

[0106] Exemplarily, generally, the heat map of people is diffusible or block-shaped, that is, the densely populated part will only gather in a certain part. Maybe there are some activities or promotions in this part that make people dense, and the density of the adjacent area of the densely populated area must be equal to or less than the density of the people in this area. Therefore, the evacuation strategy described in this embodiment adopts a diffusion type. Specifically, when the evacuation capacity of the dense area is insufficient, that is, there is no A or B-level evacuation passage, the dense part is dispersed to the lower-level density areas in a left-right manner. At the same time, the people near the passage in the dense area are cleared, and the people far away are dispersed to the first-level area, and the same applies to others.

[0107] Similarly, for multiple floors, if there are dense people on the top floor, such as in a cinema, entertainment venue, etc., even if there is no A or B-level evacuation passage in this area, it is directly cleared through a C or D-level passage, and the lower floor does not use this part of the passage. This part of the passage is only used for evacuating the top-floor people. The lower floor adopts the above logic and preferentially passes through A, B, and C-level evacuation passages. Under non-essential circumstances, this passage cannot be used by all subsequent floors. Similarly, while evacuating the top-floor people, search for the nearest A or B-level evacuation passage according to the above logic and use the broadcast to guide the people in this area to go there. It should be noted that all passages adopt a batch-by-batch evacuation method to prevent the passage from being blocked.

[0108] It should be noted that although the above description outlines the general process of generating differentiated evacuation guidance instructions for different sub - regions, more complex situations may be encountered in actual application scenarios. For example, during a promotional event in a large shopping mall, the normalized density index of the main aisle area on the first floor (sub - region A) suddenly rises to 85, and the evacuation capacity parameter of this area is at level D. At this time, the system will immediately search the adjacent areas and find that the normalized density index of the escalator area (sub - region B) connected to it is 45, and it has an evacuation capacity parameter of level A. In this case, the system will generate an evacuation guidance instruction to guide some of the people in sub - region A to sub - region B, and evacuate them in batches through the evacuation channels of sub - region B.

[0109] Another typical scenario is in the dining area of the shopping mall (sub - region C), where the normalized density index is 75 and the evacuation capacity parameter is at level C. Among the three adjacent regions, the normalized density index of the rest area (sub - region D) is 55 and the evacuation capacity parameter is at level B; the normalized density index of the corridor area (sub - region E) is 35, but the evacuation capacity parameter is at level D; the normalized density index of the store area (sub - region F) is 65 and the evacuation capacity parameter is at level C. In this complex situation, the system will give priority to choosing sub - region D as the main evacuation path because it has relatively better evacuation channel conditions, and at the same time avoid choosing sub - region E which, although has a lower population density but has limited evacuation channels.

[0110] More complexly, in a typical evacuation case of a commercial complex: the normalized density index of the food area on the third floor (sub - region A) reaches 90, and the evacuation capacity parameter is at level D. The system will first analyze the situation of the adjacent areas on the same floor. If it is found that the normalized density index of the adjacent corridor area (sub - region B) is 50 and the evacuation capacity parameter is at level B, the system will give priority to guiding some of the people to evacuate through sub - region B.

[0111] At the same time, analyze the vertical evacuation path. For example, the south escalator area on the third floor (sub - region C) has an evacuation capacity parameter of level A, but it is detected that the normalized density index of the corresponding area on the second floor (sub - region D) has reached 75 and the evacuation capacity parameter is at level C. In this case, continue to search other areas on the second floor and find that the normalized density index of the north area on the second floor (sub - region E) is only 30 and it has an evacuation capacity parameter of level B. At this time, the system will generate a series - type evacuation instruction: guide some of the people in the food area on the third floor to the corridor area first, and then evacuate them step by step down to the first floor through the north stairwell, avoiding the high - density area on the south side of the second floor.

[0112] It should be noted that the general logic of the overall evacuation guidance instruction is to determine the main evacuation area, guide some people to the main evacuation area while evacuating the areas with high-density crowds, and evacuate all channels in batches. Generally, the crowds in the high-density areas that are not close to the evacuation channels are guided to the main evacuation area. In this way, the evacuation speed in the main evacuation area is the fastest and most efficient. Even if the channel level in the area with high-density crowds is insufficient, it can ensure that all crowds are evacuated at the fastest speed.

[0113] S4: Through the low-latency feature of the 5G private network, send the differential evacuation guidance instructions to the edge computing nodes in each sub-region respectively, and control the directional audio array by the edge computing nodes to perform precise broadcast delivery.

[0114] First, deploy edge computing nodes at key positions in each broadcast sub-region. Each node is equipped with a high-performance processor and a 5G communication module, and is connected to a directional audio array composed of 8 - 16 speakers. The directional audio array adopts phased array technology. By controlling the phase delay of each speaker, an accurate sound beam of ±3° can be formed in the vertical and horizontal directions to achieve directional delivery of sound energy. For example, in the atrium area of a shopping mall, if different evacuation instructions are required for adjacent upper and lower floors, the directional audio array can accurately control the sound beam within their respective evacuation spaces to avoid interference with each other.

[0115] The edge computing node controls the directional audio array to perform precise broadcast. By calculating the delay parameters of each speaker in real time, a directional sound beam is formed to ensure that the evacuation instructions can accurately cover the target area. For example, in the intersection area of a large shopping mall, four different-direction sound beams can be formed simultaneously to play the respective evacuation instructions to the crowds in different directions. The system will also dynamically adjust the direction of the sound beam according to the crowd flow conditions to ensure that the instructions can always be transmitted to the target crowd.

[0116] Here, those skilled in the art can understand that the specific operations of each step in the above emergency broadcast precise sending system based on 5G + secure private network have been introduced in detail in the description of the Figure 1 emergency broadcast precise sending method based on 5G + secure private network, and therefore, the repeated description thereof will be omitted.

[0117] In summary, the emergency broadcast precise transmission system based on 5G + secure private network according to the embodiments of the present invention is elucidated. By analyzing the crowd density of the monitored video stream, after knowing the crowd density of each area, the broadcast sub-areas are divided according to the crowd density, and differential evacuation guidance instructions for broadcasting are carried out. In this way, it is possible to solve the problems that due to the inability to accurately know the personnel distribution in each area, the uniformly played evacuation advertisements are likely to cause congestion of people, and the indoor positioning signal is poor, resulting in some areas not receiving evacuation instructions, improve safety, and solve the problem of casualties caused by the disorderly people during the evacuation process.

Claims

1. A method for accurately sending emergency broadcasts based on 5G+ secure private network, characterized in that: include: The surveillance video streams of each floor in the commercial complex are collected through the 5G private network, and the crowd density analysis is performed on the video streams using deep learning algorithms to generate a real-time crowd distribution heat map; Based on the crowd distribution heat map, the commercial complex is divided into multiple broadcast sub-areas, and the evacuation capacity parameters are calculated according to the number of entrances and exits and the channel width of each sub-area, and an independent broadcast frequency band is allocated to each sub-area; Collect temperature sensor data and smoke concentration data in each sub-area, and generate differentiated evacuation guidance instructions for different sub-areas; Through the low latency characteristics of the 5G private network, the differentiated evacuation guidance instructions are sent to the edge computing nodes of each sub-area respectively, and the edge computing nodes control the directional audio array for precise broadcast delivery.

2. According to the method for accurately sending emergency broadcasts based on 5G+ secure private network according to claim 1, it is characterized in that: Generating the real-time crowd distribution heat map includes: Using an improved YOLOv5 target detection algorithm to perform human target recognition on the surveillance video stream; Extract the pixel coordinates and confidence score of each person target; Based on the pixel point coordinates, an adaptive kernel density method is used to calculate the number of people within a unit area; The confidence score is used to perform weighted correction on the number of people to obtain a normalized density index, and the real-time crowd distribution heat map is generated according to the normalized density index.

3. The method for accurately sending emergency broadcasts based on 5G+ secure private network according to claim 2 is characterized in that: The improved YOLOv5 target detection algorithm includes an optimized Backbone network structure, a feature fusion network of a PANet structure, and a target box regression of an IOU loss function; The optimization of the Backbone network structure includes the optimization of the backbone feature extraction network and the feature fusion network; The optimization of the backbone feature extraction network includes: Insert the SE attention module into the main branch of the CSP module; Improve the residual learning path of the CSP module; Embed the self-attention mechanism in the conversion branch of the CSP module.

4. According to claim 3, the method for accurately sending emergency broadcasts based on 5G+ secure private network is characterized in that: The optimization of the feature fusion network includes: Add a feature recalibration module between different layers of the network, wherein the feature recalibration module includes two branches: channel recalibration and space recalibration; Among them, the channel recalibration branch uses 1×1 convolution to learn the scale factor between channels; the spatial recalibration branch uses depthwise separable convolution to learn the weight coefficient of the spatial position; The optimization of the feature fusion network also includes a cross-scale feature aggregation mechanism; The cross-scale feature aggregation mechanism includes: When the high-resolution feature map of the shallow layer needs to be fused with the low-resolution feature map of the deep layer, the high-resolution features are first downsampled by the maximum pooling with a step size of 2, and then concatenated with the low-resolution features in the channel dimension. The number of channels is adjusted and the information is fused through 1×1 convolution, and finally 3×3 convolution is used to extract the discriminative features after fusion.

5. According to claim 2, the method for accurately sending emergency broadcasts based on 5G+ secure private network is characterized in that: The specific steps of the adaptive kernel density method are: Divide the detection area into equal grid units and count the number of people n in each grid unit; Calculate the weighting coefficient according to the confidence score s of the personnel target; The Gaussian kernel function is used to estimate the density of the weighted number of people to obtain a continuous density distribution function f(x,y); The value range of the density distribution function f(x,y) is normalized to the standard range of 0-100 to generate the normalized density index D.

6. The method for accurately sending emergency broadcasts based on 5G+ secure private network according to claim 5 is characterized in that: Dividing the commercial complex into a plurality of broadcast sub-areas comprises: Obtain the architectural layout of the commercial complex and use fire partitions as the basic division unit; If the area of ​​a fire partition exceeds the preset value, it will be divided into multiple initial sub-areas based on the location of the fire curtain or fire wall; When there is an atrium connecting the floors, the vertically connected space areas are planned as the same initial partition; The division into multiple initial sub-areas based on the position of the fire curtain or the fire wall includes: Mark the locations of all fire walls and fire shutters on the building plan. If the fire wall is an uninterrupted solid wall, use it directly as the boundary of the sub-area. If there is a fire shutter, its descending trajectory range must be confirmed and used as the boundary. If a fire wall and a fire shutter are installed in parallel, components with a higher fire resistance rating should be selected as the dividing line.

7. The method for accurately sending emergency broadcasts based on 5G+ secure private network according to claim 6 is characterized in that: When encountering a fireproof rolling shutter, take the center of the roller when the fireproof rolling shutter is retracted as the starting point, extend a vertical track line downward to the ground, and reserve a safe distance on both sides of the guide rail as a buffer area, and determine the outer edge of the buffer area as the dividing line; When fire separation components divide fire partitions with an area exceeding the preset value, if a fire separation component separates similar functional spaces and does not affect normal use, it can be used as the boundary of the sub-area; if the separation will affect the function of the space, it is necessary to find the next suitable fire separation component as the boundary.

8. The method for accurately sending emergency broadcasts based on 5G+ secure private network according to claim 6 is characterized in that: When a preset percentage of areas within the initial partition present the same density level, the initial partition is determined as an independent broadcast sub-area; When different density levels appear in the initial partition, if the density level difference between adjacent areas is 1 level and the difference lasts for less than 5 minutes, the original partition is maintained unchanged; if the density level difference between adjacent areas is 1 level and the difference lasts for more than 5 minutes, the partition boundary is set at the density level jump position; If the density level difference between adjacent areas is greater than or equal to 2 levels, the partition boundary is immediately set at the location with the largest density gradient; if the density level appears to be distributed in a ring shape within the partition, the area is divided using concentric circles.

9. An emergency broadcast precision transmission system based on 5G+ secure private network, characterized in that: include: The acquisition module is used to collect surveillance video streams of each floor in the commercial complex through the 5G private network; collect temperature sensor data and smoke concentration data in each sub-area; An edge computing module is used to analyze the crowd density of the video stream using a deep learning algorithm to generate a real-time crowd distribution heat map; and calculate the evacuation capacity parameters according to the number of entrances and exits and the channel width of each sub-area; A partition control module is used to generate a real-time crowd distribution heat map according to the results of the edge computing module; Divide broadcast sub-areas based on edge computing module parameters; The broadcast delivery module is used to control the directional audio array for precise broadcast delivery.

Images