Image analysis-based post-earthquake evacuation method and system

Abstract

The invention discloses a post-earthquake evacuation method and system based on image analysis, belongs to the technical field of earthquake engineering, and aims to solve the problems of slow response and inaccurate guidance due to the fact that dynamic changes of secondary disasters such as aftershocks and fire disasters cannot be handled in real time by means of manual guidance and traditional emergency broadcast and static evacuation routes. Comprising the steps of obtaining high-definition image data through an unmanned aerial vehicle, a satellite, a camera, a LiDAR and a spectrum sensor in an earthquake influence area, performing multi-dimensional processing on the image data based on a deep learning model, and generating a risk thermodynamic diagram of a dangerous area; through deep analysis of different types of image data, multiple risk sources in the post-earthquake environment can be identified, the dangerous area thermodynamic diagram is generated according to the risk sources, the system can dynamically plan the optimal evacuation path based on the dangerous area thermodynamic diagram, it is ensured that personnel avoid high-risk areas in the evacuation process, and the personnel evacuation efficiency is improved. Therefore, the evacuation efficiency and the personnel safety are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of earthquake engineering, and in particular relates to a post-earthquake evacuation method and system based on image analysis. Background Art

[0002] Earthquakes are one of the most devastating natural disasters affecting daily life. Globally, an average of over 10,000 earthquakes occur daily. Since 2003, various countries have proposed urban seismic resilience plans and targets to minimize post-disaster losses. With the growing population, increasing building density, and the proliferation of complex bridges within cities, the transportation system, as one of the most critical infrastructure systems within a city, bears the brunt of the rescue and relief efforts following an earthquake. Following an earthquake, buildings are damaged, roads are blocked, and secondary disasters (such as gas leaks and fires) are frequent, making evacuation a critical emergency response.

[0003] Current post-earthquake evacuation plans mainly rely on manual guidance and traditional emergency broadcasts. Static evacuation routes cannot respond to the dynamic changes of secondary disasters such as aftershocks and fires in real time, and there are problems such as slow response and inaccurate guidance. With the rapid development of computer vision technology and image analysis technology, image-based emergency evacuation systems are expected to provide more efficient and accurate solutions for post-earthquake evacuation.

[0004] Therefore, a post-earthquake evacuation method and system based on image analysis is needed to solve the problems in the existing technology that rely on manual guidance and traditional emergency broadcasts, static evacuation routes cannot respond to the dynamic changes of secondary disasters such as aftershocks and fires in real time, and have slow response and inaccurate guidance. Summary of the Invention

[0005] The object of the present invention is to provide a post-earthquake evacuation method and system based on image analysis to solve the problems raised in the above background technology.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a post-earthquake evacuation method based on image analysis, comprising:

[0007] S1. Obtain high-definition image data using drones, satellites, cameras, LiDAR, and spectral sensors within the earthquake-affected area;

[0008] S2. Perform multi-dimensional processing on image data based on a deep learning model to generate a risk heat map of hazardous areas.

[0009] S3. Define the evacuation target area based on the risk heat map of the dangerous area;

[0010] S4. Plan the evacuation route based on the target evacuation area.

[0011] It should be noted in the plan that the high-definition image data includes but is not limited to building collapse, road damage, personnel distribution, fire, gas leakage and obstacle post-disaster risk points.

[0012] It is further worth noting that step S2 includes:

[0013] 201. Preprocessing the image data;

[0014] 202. Based on the YOLOv8 model, the input is four-channel data after the fusion of visible light image and infrared image. The output includes static obstacles and dynamic obstacles, and an obstacle distribution map is generated;

[0015] 203. Detect gas leaks through spectral analysis and generate leak thermograms;

[0016] 204. Based on the infrared threshold segmentation temperature > 60℃, combined with the flame flicker frequency > 8Hz to filter out false alarms, the fire heat source is located; the area with temperature > 60℃ in the infrared image is marked as a potential fire source, and its area ratio A is calculated. fire , analyze the flicker frequency f of the flame area by optical flow method, if f>8Hz, confirm the fire source, otherwise filter; fire Normalized to [0,1], we get A fire , generate a fire hazard source distribution map;

[0017] 205. Calculate the safety score of the building structure surface and output the building structure safety heat map;

[0018] 206. Based on the obstacle distribution map, leakage heat map, fire hazard source distribution map and building structure safety heat map, calculate the hazard index and generate the hazardous area risk heat map.

[0019] It should be further explained that step S203 includes:

[0020] Based on the image data, the methane characteristic band is extracted and the spectral index is calculated:

[0021]

[0022] Among them, R 3.3 represents the 3.3 μm infrared absorption peak of methane in the spectral data, R 3.9 Represents the 3.9μm infrared absorption peak of methane in the spectral data. When the spectral index I gas >Threshold 1.2, judged as leakage;

[0023] Based on weather wind speed and direction data, the Gaussian diffusion model is used to predict the leakage range and generate a leakage heat map.

[0024] As a preferred implementation, step S205 includes:

[0025] Use COLMAP to perform multi-view stereo vision reconstruction, generate dense point clouds, perform plane fitting on the point clouds, and extract wall and floor structural surfaces;

[0026] Based on the point cloud normal angle θ>30° and the curvature mutation Δκ, the structural surface normal is fitted by principal component analysis to calculate the angle with the vertical direction.

[0027] α=arccos(|n·z|)

[0028] Where α>5° is marked as tilt, n is the normal vector of the structural surface obtained by 3D reconstruction, indicating the vertical direction of the structural surface, and z is the vertical reference vector in the global coordinate system, usually defined as z = (0, 0, 1), that is, the vertical upward unit vector;

[0029] Compared with the BIM model before the earthquake, the displacement vector modulus Δd>10cm is considered dangerous;

[0030] Assign a weight to each structural surface, and the comprehensive score S is:

[0031]

[0032] Among them, w i is the weight coefficient, the weight of the load-bearing wall is 0.5, and the weight of the non-load-bearing wall is 0.3; Δd i is the displacement of the i-th structural surface; D max is the maximum allowable displacement threshold; α i is the inclination angle of the i-th structural surface; α max is the maximum allowable tilt angle threshold;

[0033] Based on the comprehensive score, a building structure safety heat map is output.

[0034] As a preferred implementation, step S206 includes:

[0035] Project the obstacle distribution map, leakage heat map, fire hazard source distribution map and building structure safety heat map data onto a two-dimensional grid;

[0036] The risk index D is calculated as:

[0037] D=w1·I gas +w2·I fire +w3·(1-S)

[0038] Among them, w1, w2, and w3 are weight coefficients, reflecting the contribution of different hazard sources to the overall risk;

[0039] The hazard index D is mapped to a two-dimensional grid to generate a risk heat map of the hazardous area.

[0040] As a preferred embodiment, step S4 includes:

[0041] 401. The roads, buildings, and shelters in the disaster area are considered as nodes in the graph;

[0042] 402. Define paths by the connectivity between adjacent nodes. If there is a road connection between two nodes, an edge is added to the graph. If there is a high-risk area between the two nodes, they are not connected.

[0043] 403. Convert the danger value of each node into edge weight according to the danger zone heat map;

[0044] 404. Set the starting point and the end point, calculate the safest path from the starting point to the end point, and obtain the initial path;

[0045] 405. When the heat map update causes the grid danger index D of a node to be greater than 0.7, relocation is triggered;

[0046] 406. Reset the starting point and end point and re-plan the evacuation route.

[0047] The present invention also provides an earthquake post-earthquake evacuation system based on image analysis, comprising:

[0048] The image acquisition module collects multimodal high-definition image data of post-earthquake scenes in real time through drones, ground cameras, LiDAR, and spectral sensors;

[0049] The data analysis unit processes and analyzes the image data collected by the image acquisition module, including a data preprocessing module, a spectral analysis module, a structural safety assessment module, an obstacle analysis module, a fire hazard source analysis module and a hazard index calculation module. The data preprocessing module receives the image data from the image acquisition module, denoises, aligns, registers and fuses the collected image or point cloud data, extracts key features, and provides standardized input for subsequent analysis. The spectral analysis module analyzes the characteristic bands in the spectral data, locates the gas leakage point, and predicts the leakage range in combination with the Gaussian diffusion model, marks it as a high-risk area, and generates a leakage heat map. The structural safety assessment module generates a dense The point cloud is compared with the BIM model to calculate the displacement vector and tilt angle, quantify the building structure safety score, mark it as "safe", "warning" or "dangerous" areas, and output the building structure safety heat map. The obstacle analysis module detects static and dynamic obstacles, uses the YOLOv8 model to identify the obstacle type and location, and outputs an obstacle distribution map and type label. The fire hazard source analysis module combines infrared images and visible light images to detect fire heat sources, predict fire spread trends, mark them as high-risk areas, and output a fire hazard source distribution map. The hazard index calculation module calculates the hazard index based on the obstacle distribution map, leakage heat map, fire hazard source distribution map, and building structure safety heat map.

[0050] The heat map generation module integrates the output of the data analysis unit to generate a risk heat map of the hazardous area;

[0051] The path planning unit is used to safely plan the evacuation path, including a model construction module and a path algorithm planning module. The model construction module constructs a dynamic navigation map based on the risk heat map of the dangerous area and the map. The nodes are safe areas, the edges are passable paths, and each edge is associated with a risk cost. The path algorithm planning module calculates the safest path from the starting point to the end point based on the set starting point and end point to obtain the initial path, and then adjusts the path in real time according to the dynamic navigation map. When the original end point fails, it automatically switches to the backup end point to obtain the optimal evacuation path.

[0052] Compared with the prior art, the present invention provides an image analysis-based post-earthquake evacuation method and system, which has at least the following beneficial effects:

[0053] By setting up data from multiple sensors, such as drones, satellites, cameras and deep learning technology, the system can quickly and accurately obtain and process post-earthquake site information, and can evaluate the post-disaster environment more real-time and objectively; through in-depth analysis of different types of image data, it can identify multiple risk sources in the post-earthquake environment and generate a heat map of dangerous areas based on these risk sources, helping users to clearly understand the risk situation at the post-disaster site and quickly determine which areas are safe. At the same time, based on the heat map of dangerous areas, the system can dynamically plan the optimal evacuation route to ensure that personnel avoid high-risk areas during evacuation, thereby improving evacuation efficiency and personnel safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 This is a flow chart of the post-earthquake evacuation method based on image analysis of the present invention;

[0055] Figure 2 Flowchart of the method for generating a risk heat map of a dangerous area in step S2 of the present invention

[0056] Figure 3 This is a structural block diagram of the post-earthquake evacuation system based on image analysis of the present invention. DETAILED DESCRIPTION

[0057] The present invention will be further described below with reference to the embodiments.

[0058] Reference Figure 1 As shown, the present invention provides an earthquake post-earthquake evacuation method based on image analysis, comprising:

[0059] S1. Obtain high-definition image data using drones, satellites, cameras, LiDAR, and spectral sensors within the earthquake-affected area. The high-definition image data includes, but is not limited to, collapsed buildings, road damage, personnel distribution, and post-disaster risk points such as fires, gas leaks, and obstacles.

[0060] S2. Perform multi-dimensional processing on image data based on a deep learning model to generate a risk heat map of hazardous areas.

[0061] S3. Define the evacuation target area based on the risk heat map of the dangerous area;

[0062] S4. Plan the evacuation route based on the target evacuation area.

[0063] Further, such as Figure 2 As shown, step S2 includes:

[0064] 201. Preprocess the image data, including integrating visible light, infrared, LiDAR point cloud and spectral data to generate a four-channel input; synchronize multi-sensor data using timestamps and align point cloud and image pixels through spatial coordinate transformation; dehaze and contrast enhance visible light images, and perform outlier filtering on LiDAR point clouds;

[0065] 202. Based on the YOLOv8 model, the input is four-channel data after the fusion of visible light image and infrared image. The output includes static obstacles and dynamic obstacles, and generates an obstacle distribution map; detects static and dynamic obstacles such as collapsed walls and vehicle pileups; generates a mask for each obstacle through Mask R-CNN and labels the type; combines optical flow method to analyze continuous frames and filter static false detections

[0066] 203. Detect gas leaks through spectral analysis and generate leak thermograms;

[0067] Based on the image data, the methane characteristic band is extracted and the spectral index is calculated:

[0068]

[0069] Among them, R 3.3 represents the 3.3 μm infrared absorption peak of methane in the spectral data, R 3.9 Represents the 3.9μm infrared absorption peak of methane in the spectral data. When the spectral index I gas >Threshold 1.2 is considered a leak;

[0070] Based on weather wind speed and direction data, the Gaussian diffusion model is used to predict the leakage range and generate a leakage heat map;

[0071] 204. Based on the infrared threshold segmentation temperature > 60℃, combined with the flame flicker frequency > 8Hz to filter out false alarms, the fire heat source is located; the area with temperature > 60℃ in the infrared image is marked as a potential fire source, and its area ratio A is calculated. fire , analyze the flicker frequency f of the flame area by optical flow method, if f>8Hz, confirm the fire source, otherwise filter; fire Normalized to [0,1], we get A fire , generate a fire hazard source distribution map;

[0072] 205. Calculate the safety score of the building structure surface and output the building structure safety heat map;

[0073] Use COLMAP to perform multi-view stereo vision reconstruction, generate dense point clouds, perform plane fitting on the point clouds, and extract wall and floor structural surfaces;

[0074] Based on the point cloud normal angle θ>30° and the curvature mutation Δκ, the structural surface normal is fitted by principal component analysis to calculate the angle with the vertical direction.

[0075] α=arccos(|n·z|)

[0076] Where α > 5° is marked as tilt, n is the normal vector of the structural surface (such as wall or floor) obtained by 3D reconstruction, indicating the vertical direction of the structural surface, and z is the vertical reference vector in the global coordinate system, usually defined as z = (0, 0, 1), that is, the vertical upward unit vector;

[0077] Compared with the BIM model before the earthquake, the displacement vector modulus Δd>10cm is considered dangerous;

[0078] Assign a weight to each structural surface, and the comprehensive score S is:

[0079]

[0080] Among them, w i is the weight coefficient, the weight of the load-bearing wall is 0.5, and the weight of the non-load-bearing wall is 0.3; Δd i is the displacement of the i-th structural surface; D max is the maximum allowable displacement threshold; α i is the inclination angle of the i-th structural surface; α max is the maximum allowable tilt angle threshold;

[0081] Based on the comprehensive score, output the building structure safety heat map;

[0082] 206. Based on the obstacle distribution map, leakage heat map, fire hazard source distribution map and building structure safety heat map, calculate the hazard index and generate the hazard area risk heat map;

[0083] Project the obstacle distribution map, leakage heat map, fire hazard source distribution map and building structure safety heat map data onto a two-dimensional grid;

[0084] Calculate the risk index

[0085] D=w1·I gas +w2·I fire +w3·(1-S)

[0086] Among them, w1, w2, and w3 are weight coefficients, reflecting the contribution of different hazard sources to the overall risk;

[0087] The hazard index D is mapped to a two-dimensional grid to generate a risk heat map of the hazardous area.

[0088] Furthermore, step S4 includes:

[0089] 401. The roads, buildings, and shelters in the disaster area are considered as nodes in the graph;

[0090] 402. Define paths by the connectivity between adjacent nodes. If there is a road connection between two nodes, an edge is added to the graph. If there is a high-risk area between the two nodes, they are not connected.

[0091] 403. Convert the danger value of each node into edge weight according to the danger zone heat map;

[0092] 404. Set the starting point and the end point, calculate the safest path from the starting point to the end point, and obtain the initial path;

[0093] 405. When the heat map update causes the grid danger index D of a node to be greater than 0.7, relocation is triggered;

[0094] 406. Reset the starting point and end point and re-plan the evacuation route.

[0095] According to the above working process, it can be seen that: through in-depth analysis of different types of image data, various risk sources in the post-earthquake environment can be identified, and a heat map of dangerous areas can be generated based on these risk sources, helping users to clearly understand the risk situation at the post-disaster site and quickly determine which areas are safe. At the same time, based on the heat map of dangerous areas, the system can dynamically plan the optimal evacuation route to ensure that personnel avoid high-risk areas during evacuation, thereby improving evacuation efficiency and personnel safety.

[0096] For further information, please refer to Figure 3 As shown in FIG, a post-earthquake evacuation method based on image analysis is proposed. This solution proposes a post-earthquake evacuation system based on image analysis, comprising:

[0097] The image acquisition module collects multimodal high-definition image data of post-earthquake scenes in real time through drones, ground cameras, LiDAR, and spectral sensors;

[0098] The data analysis unit processes and analyzes the image data collected by the image acquisition module, including a data preprocessing module, a spectral analysis module, a structural safety assessment module, an obstacle analysis module, a fire hazard source analysis module and a hazard index calculation module. The data preprocessing module receives the image data from the image acquisition module, denoises, aligns, registers and fuses the collected image or point cloud data, extracts key features, and provides standardized input for subsequent analysis. The spectral analysis module analyzes the characteristic bands in the spectral data, locates the gas leakage point, and predicts the leakage range in combination with the Gaussian diffusion model, marks it as a high-risk area, and generates a leakage heat map. The structural safety assessment module generates a dense The point cloud is compared with the BIM model to calculate the displacement vector and tilt angle, quantify the building structure safety score, mark it as "safe", "warning" or "dangerous" areas, and output the building structure safety heat map. The obstacle analysis module detects static and dynamic obstacles, uses the YOLOv8 model to identify the obstacle type and location, and outputs an obstacle distribution map and type label. The fire hazard source analysis module combines infrared images and visible light images to detect fire heat sources, predict fire spread trends, mark them as high-risk areas, and output a fire hazard source distribution map. The hazard index calculation module calculates the hazard index based on the obstacle distribution map, leakage heat map, fire hazard source distribution map, and building structure safety heat map.

[0099] The heat map generation module integrates the output of the data analysis unit to generate a risk heat map of the hazardous area;

[0100] The path planning unit is used to safely plan the evacuation path, including a model construction module and a path algorithm planning module. The model construction module constructs a dynamic navigation map based on the risk heat map of the dangerous area and the map. The nodes are safe areas, the edges are passable paths, and each edge is associated with a risk cost. The path algorithm planning module calculates the safest path from the starting point to the end point based on the set starting point and end point to obtain the initial path, and then adjusts the path in real time according to the dynamic navigation map. When the original end point fails, it automatically switches to the backup end point to obtain the optimal evacuation path.

[0101] The usage process of the above-mentioned post-earthquake evacuation system based on image analysis is as follows:

[0102] Step 1: Use image acquisition modules such as drones, satellites, and ground cameras to collect image data from the post-earthquake scene;

[0103] Step 2: The collected image data is processed and analyzed through the data preprocessing module to provide standardized input for subsequent analysis;

[0104] Step 3: The spectral analysis module analyzes the characteristic bands in the spectral data, locates the gas leak point, and uses the Gaussian diffusion model to predict the leakage range, marking it as a high-risk area and generating a leakage heat map. The structural safety assessment module generates a dense point cloud through COLMAP, compares it with the BIM model to calculate the displacement vector and tilt angle, quantifies the building structure safety score, marks it as "safe", "warning" or "dangerous" area, and outputs the building structure safety heat map. The obstacle analysis module detects static and dynamic obstacles, uses the YOLOv8 model to identify the obstacle type and location, and outputs an obstacle distribution map and type label. The fire hazard source analysis module combines infrared images and visible light images to detect fire heat sources, predict the fire spread trend, mark it as a high-risk area, and output a fire hazard source distribution map.

[0105] Step 4: The hazard index calculation module calculates the hazard index based on the obstacle distribution map, leakage heat map, fire hazard source distribution map, and building structure safety heat map;

[0106] Step 5: The heat map generation module integrates the leakage heat map, building structure safety heat map, obstacle distribution map and type labels, fire hazard source distribution map and hazard index to generate a risk heat map of the dangerous area;

[0107] Step 6: Model construction module, based on the risk heat map of dangerous areas and the map to build a dynamic navigation map;

[0108] Step 7: The path algorithm planning module calculates the safest path from the starting point to the end point based on the set starting point and end point, obtains the initial path, and then adjusts the path in real time according to the dynamic navigation map. When the original end point fails, it automatically switches to the backup end point to obtain the optimal evacuation path.

[0109] To sum up, the advantages of the present invention are: by setting up multiple sensor data, such as drones, satellites, cameras and deep learning technology, the system can quickly and accurately obtain and process post-earthquake site information, and can evaluate the post-disaster environment more in real time and objectively; by performing in-depth analysis of different types of image data, it can identify multiple risk sources in the post-earthquake environment, such as collapsed buildings, fires, gas leaks, etc., and generate a heat map of dangerous areas based on these risk sources, helping users to clearly understand the risk situation at the post-disaster site, and quickly determine which areas are safe and which areas need special attention. At the same time, based on the heat map of dangerous areas, the system can dynamically plan the optimal evacuation route to ensure that personnel avoid high-risk areas during evacuation, thereby improving evacuation efficiency and personnel safety.

[0110] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions merely illustrate the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and modifications are intended to fall within the scope of the present invention. The scope of protection claimed by the present invention is defined by the appended claims and their equivalents.

Claims

1. A post-earthquake evacuation method based on image analysis, characterized in that: include: S1. Obtain high-definition image data using drones, satellites, cameras, LiDAR, and spectral sensors within the earthquake-affected area; S2. Perform multi-dimensional processing on image data based on a deep learning model to generate a risk heat map of hazardous areas. S3. Define the evacuation target area based on the risk heat map of the dangerous area; S4. Plan the evacuation route based on the target evacuation area.

2. The post-earthquake evacuation method based on image analysis according to claim 1, characterized in that: The high-definition image data includes but is not limited to building collapse, road damage, personnel distribution, fire, gas leakage and obstacle post-disaster risk points.

3. The post-earthquake evacuation method based on image analysis according to claim 1, characterized in that: The step S2 comprises:

201. Preprocessing the image data; 202. Based on the YOLOv8 model, the input is four-channel data after the fusion of visible light image and infrared image. The output includes static obstacles and dynamic obstacles, and an obstacle distribution map is generated; 203. Detect gas leaks through spectral analysis and generate leak thermograms; 204. Based on the infrared threshold segmentation temperature > 60℃, combined with the flame flicker frequency > 8Hz to filter out false alarms, the fire heat source is located; the area with temperature > 60℃ in the infrared image is marked as a potential fire source, and its area ratio A is calculated. fire , analyze the flicker frequency f of the flame area by optical flow method, if f>8Hz, confirm the fire source, otherwise filter; fire Normalized to [0,1], we get A fire , generate a fire hazard source distribution map; 205. Calculate the safety score of the building structure surface and output the building structure safety heat map; 206. Based on the obstacle distribution map, leakage heat map, fire hazard source distribution map and building structure safety heat map, calculate the hazard index and generate the hazardous area risk heat map.

4. The post-earthquake evacuation method based on image analysis according to claim 3, characterized in that: Step S203 includes: Based on the image data, the methane characteristic band is extracted and the spectral index is calculated: Among them, R 3.3 represents the 3.3 μm infrared absorption peak of methane in the spectral data, R 3.9 Represents the 3.9μm infrared absorption peak of methane in the spectral data. When the spectral index I gas >Threshold 1.2, judged as leakage; Based on weather wind speed and direction data, the Gaussian diffusion model is used to predict the leakage range and generate a leakage heat map.

5. The post-earthquake evacuation method based on image analysis according to claim 3, characterized in that: Step S205 includes: Use COLMAP to perform multi-view stereo vision reconstruction, generate dense point clouds, perform plane fitting on the point clouds, and extract wall and floor structural surfaces; Based on the point cloud normal angle θ>30° and the curvature mutation Δκ, the structural surface normal is fitted by principal component analysis to calculate the angle with the vertical direction. α=arccos(|n·z|) Where α>5° is marked as tilt, n is the normal vector of the structural surface obtained by 3D reconstruction, indicating the vertical direction of the structural surface, and z is the vertical reference vector in the global coordinate system, usually defined as z = (0, 0, 1), that is, the vertical upward unit vector; Compared with the BIM model before the earthquake, the displacement vector modulus Δd>10cm is considered dangerous; Assign a weight to each structural surface, and the comprehensive score S is: Among them, w i is the weight coefficient, the weight of the load-bearing wall is 0.5, and the weight of the non-load-bearing wall is 0.3; Δd i is the displacement of the i-th structural surface; D max is the maximum allowable displacement threshold; α i is the inclination angle of the i-th structural surface; α max is the maximum allowable tilt angle threshold; Based on the comprehensive score, a building structure safety heat map is output.

6. The post-earthquake evacuation method based on image analysis according to claim 5, characterized in that: Step S206 includes: Project the obstacle distribution map, leakage heat map, fire hazard source distribution map and building structure safety heat map data onto a two-dimensional grid; Calculate the risk index D: D=w1·I gas +w2·I fire +w3·(1-S) Among them, w1, w2, and w3 are weight coefficients, reflecting the contribution of different hazard sources to the overall risk; The hazard index D is mapped to a two-dimensional grid to generate a risk heat map of the hazardous area.

7. The post-earthquake evacuation method based on image analysis according to claim 6, characterized in that: Step S4 includes:

401. The roads, buildings and shelter areas in the disaster area are considered as nodes in the graph; 402. Define paths by the connectivity between adjacent nodes. If there is a road connecting two nodes, an edge is added to the graph. If there is a high-risk area between the two nodes, they are not connected.

403. Convert the danger value of each node into edge weight according to the danger zone heat map; 404. Set the starting point and the end point, calculate the safest path from the starting point to the end point, and obtain the initial path; 405. When the heat map update causes the grid danger index D of a node to be greater than 0.7, relocation is triggered; 406. Reset the starting point and end point and re-plan the evacuation route.

8. An image analysis-based post-earthquake evacuation system, used to implement an image analysis-based post-earthquake evacuation method according to any one of claims 1 to 7, characterized in that: include The image acquisition module collects multimodal high-definition image data of post-earthquake scenes in real time through drones, ground cameras, LiDAR, and spectral sensors; The data analysis unit processes and analyzes the image data collected by the image acquisition module, including a data preprocessing module, a spectral analysis module, a structural safety assessment module, an obstacle analysis module, a fire hazard source analysis module and a hazard index calculation module. The data preprocessing module receives the image data from the image acquisition module, denoises, aligns, registers and fuses the collected image or point cloud data, extracts key features, and provides standardized input for subsequent analysis. The spectral analysis module analyzes the characteristic bands in the spectral data, locates the gas leakage point, and predicts the leakage range in combination with the Gaussian diffusion model, marks it as a high-risk area, and generates a leakage heat map. The structural safety assessment module generates a dense The point cloud is compared with the BIM model to calculate the displacement vector and tilt angle, quantify the building structure safety score, mark "safe," "warning," or "dangerous" areas, and output the building structure safety heat map. The obstacle analysis module detects static and dynamic obstacles, uses the YOLOv8 model to identify obstacle types and locations, and outputs an obstacle distribution map and type labels. The fire hazard source analysis module combines infrared images and visible light images to detect fire heat sources, predict fire spread trends, mark high-risk areas, and output a fire hazard source distribution map. The hazard index calculation module calculates the hazard index based on the obstacle distribution map, leakage heat map, fire hazard source distribution map, and building structure safety heat map. The heat map generation module integrates the output of the data analysis unit to generate a risk heat map of the hazardous area; The path planning unit is used to safely plan the evacuation path, including a model construction module and a path algorithm planning module. The model construction module constructs a dynamic navigation map based on the risk heat map of the dangerous area and the map. The nodes are safe areas, the edges are passable paths, and each edge is associated with a risk cost. The path algorithm planning module calculates the safest path from the starting point to the end point based on the set starting point and end point to obtain the initial path, and then adjusts the path in real time according to the dynamic navigation map. When the original end point fails, it automatically switches to the backup end point to obtain the optimal evacuation path.

Images