A dangerous chemical warehouse storage safety management system and method

Abstract

The application discloses a kind of based on dangerous chemical warehouse storage safety management system and method, it is related to dangerous chemical warehouse safety management technical field, including acquisition dangerous chemical warehouse three-dimensional point cloud data and carries out space denoising, generates the space point cloud set after purification;Identify high-density aggregation area in point cloud and extract warehouse internal entity partition contour, track profile trend change capture turnaround position and mark as space topological conversion node, connect adjacent node to form closed space segmentation polygon, define as independent storage management domain;Traversal each management domain generates netted inspection line with cross feature, sensor detection site is arranged in the geometric center of netted line, and the global safety management path scheme is output.The application realizes storage domain automatic division by three-dimensional point cloud processing and space topological segmentation, combines netted inspection line with center sensor arrangement optimization management path, adapts dangerous chemical warehouse space safety management demand.

Description

Technical Field

[0001] This invention belongs to the field of hazardous chemical storage safety management technology, specifically a hazardous chemical warehouse storage safety management system and method. Background Technology

[0002] Current hazardous chemical warehouse storage safety management largely relies on manual methods to delineate storage areas, using two-dimensional plan drawings to define the internal space. Inspection routes often employ simple, manually planned circular or straight paths. Sensor placement depends on experience, and 3D point cloud data is used only for basic warehouse space visualization, without in-depth processing and application tailored to the structural needs of hazardous chemical warehouses. When manually defining the internal partition structure, partition distribution can only be determined visually, failing to accurately extract partition outlines using 3D spatial data. Conventional techniques lack spatial noise reduction processing for 3D point cloud data, resulting in insufficient accuracy in identifying high-density spatial areas, inability to capture the zigzag characteristics of partition directions, and inability to use topological nodes to achieve closed-loop division of warehouse space.

[0003] Conventional management methods cannot achieve automated topological segmentation of warehouse space. The division of independent storage management domains suffers from subjective bias, patrol routes lack intersecting mesh spatial coverage, and sensor sites cannot form a geometrical matching relationship with patrol routes. This invention addresses the problems of existing technologies being unable to extract partition outlines from high-density areas through post-cleaning 3D point cloud identification, and unable to mark partition turning points to form topological transformation nodes and segment spatial management domains. It also solves the problems of not being able to generate intersecting mesh patrol routes within independent storage management domains and not being able to deploy sensor detection sites at the geometric center of patrol routes. This enables precise segmentation of hazardous chemical warehouse storage space and standardized deployment of safety management paths. Summary of the Invention

[0004] This invention aims to solve at least one of the technical problems existing in the prior art; Therefore, this invention proposes a method for safe management of hazardous chemical warehouse storage, including: Collect three-dimensional point cloud data of a hazardous chemical warehouse, perform spatial noise reduction processing on the three-dimensional point cloud data, and generate a purified spatial point cloud set; Identify high-density clustered areas in the purified spatial point cloud set, and extract the outlines of the internal solid partitions of the warehouse based on the high-density clustered areas. Track the changes in the outline of the physical partitions inside the warehouse, capture the turning points of the changes in the outline, and mark the turning points as spatial topology transformation nodes; Connect adjacent spatial topology transformation nodes to form several closed spatial partitioning polygons, and define the closed spatial partitioning polygons as independent storage management domains; Traverse the independent storage management domain, generate a mesh patrol route with intersecting features within the independent storage management domain, and arrange sensor detection points at the geometric center of the mesh patrol route to output a global security management path scheme.

[0005] Furthermore, the process of collecting the three-dimensional point cloud data of the hazardous chemical warehouse and performing spatial noise reduction processing on the three-dimensional point cloud data includes: The multi-angle lidar array is activated to transmit detection pulse signals into the interior of the hazardous chemical warehouse and to receive the intensity information of the reflected echoes. The echo intensity information is summarized to construct an initial three-dimensional point cloud data frame sequence; The spatial noise reduction process is completed by analyzing the point distribution density of each frame in the three-dimensional point cloud data frame sequence, removing isolated noise points and floating ghost points, and retaining the point data attached to the solid surface.

[0006] Further, high-density clustered areas are identified in the purified spatial point cloud set, and the outlines of internal warehouse partitions are extracted based on these high-density clustered areas, including: The purified spatial point cloud set is voxelized to discretize the continuous space into several regularly arranged data units. Calculate the number of point clouds in each data unit, filter out the data units whose number of point clouds exceeds a preset threshold, and classify them as the high-density clustering region; The edge points of the high-density clustered area are extracted, and a smooth curve is fitted to the edge points to obtain the continuous outline of the internal solid partition of the warehouse.

[0007] Furthermore, tracking the directional changes of the physical partition outlines within the warehouse, and capturing the reversal points of these directional changes, includes: Calculate the azimuth difference between adjacent sampling points along the tangent direction of the internal solid partition outline of the warehouse; When the absolute value of the azimuth difference exceeds a right angle, it is determined that the direction change has occurred at that location; Record the coordinate information of the turning point where the direction change occurs, and store the coordinate information in the node candidate list.

[0008] Furthermore, connecting adjacent spatial topology transformation nodes forms several closed spatial partitioning polygons, including: Read the turnaround position from the node candidate list and rename it to the spatial topology transformation node; Find the shortest connecting path between the spatial topology transformation nodes, and the shortest connecting path must avoid the high-density clustered area; Based on the shortest connected path, draw connecting lines, and form a closed spatial segmentation polygon by connecting multiple connecting lines end to end.

[0009] Furthermore, traversing the independent storage management domain, a mesh inspection route with intersecting characteristics is generated within the independent storage management domain, including: Measure the long side and short side dimensions of the independent storage management domain, and determine the grid generation mode based on the ratio of the long side to the short side dimensions; If the ratio is greater than the preset ratio, the main inspection trunk line is laid first along the length direction, and then the branch inspection trunk line is arranged perpendicular to the main inspection trunk line. If the ratio is less than or equal to a preset ratio, the mesh inspection route is generated in a honeycomb pattern to ensure that the mesh inspection route covers the entire area of ​​the independent storage management domain.

[0010] Furthermore, sensor detection points are arranged at the geometric center of the mesh inspection route, including: Calculate the midpoint coordinates of each line segment in the network patrol route, and use the midpoint coordinates as candidate deployment locations; Evaluate the distance between the candidate deployment locations and the boundary of the independent storage management domain, and eliminate candidate deployment locations whose distance is less than the safe distance limit; The remaining candidate deployment locations are identified as the sensor detection sites, and corresponding monitoring parameter types are assigned to the sensor detection sites.

[0011] Furthermore, it also includes: Obtain the list of hazardous chemicals stored in the independent storage management domain, and parse the hazard level identifiers in the list of hazardous chemicals; Based on the hazard level markings, adjust the density of the mesh patrol routes, and increase the number of mesh patrol routes in high-hazard areas; The trigger sensitivity threshold of the sensor detection site is updated synchronously to adapt it to the hazardous characteristics of the hazardous chemical material list.

[0012] Furthermore, after the step of outputting the global security management path scheme, the method further includes: Receive real-time location feedback information from mobile inspection equipment and compare the deviation between the real-time location feedback information and the overall security management path scheme; When the deviation exceeds the allowable error range, the temporary travel route of the mobile inspection equipment is replanned, and the temporary travel route must bypass dynamic obstacles; Record the location of the dynamic obstacle and mark it on the cleaned spatial point cloud set to update the outline of the internal solid partition of the warehouse.

[0013] Furthermore, the present invention also includes a hazardous chemical warehouse storage safety management system, the system including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein when the processor executes the computer program, it implements the steps of the hazardous chemical warehouse storage safety management method described above.

[0014] Compared with the prior art, the beneficial effects of the present invention are: Spatial noise reduction processing is performed on the 3D point cloud data of hazardous chemical warehouses to remove redundant spatial noise information and form a clean spatial point cloud set. The purified point cloud set can accurately identify high-density clustered areas inside the warehouse. Based on the high-density areas, the real outline of the physical partitions can be directly extracted. By tracking the direction and changes of the physical partition outlines, the extension trajectory of the outlines in space can be completely captured. The turning points of the outline direction can be accurately locked and marked as spatial topology transformation nodes. Adjacent spatial topology transformation nodes are connected to form closed spatial segmentation polygons that fit the actual spatial structure of the warehouse. The independent storage management domain defined by this perfectly matches the spatial distribution of physical partitions inside the warehouse.

[0015] Within each independent storage management domain, a mesh-like inspection route with intersecting features is generated, allowing the inspection path to cover the entire spatial area of ​​the independent storage management domain. The geometric center of the mesh-like inspection route is the optimal placement location for spatial detection. Placing sensor detection points at this location maximizes the spatial coverage of sensor detection. After traversing all independent storage management domains to complete the generation of inspection routes and the placement of sensor points, a safety management path scheme adapted to the entire spatial structure of the warehouse can be output. This allows the inspection route and sensor detection points to form a spatial collaborative relationship, achieving precise adaptation of paths and locations for the safety management of hazardous chemical warehouse storage space. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the steps of a method for safe storage management of hazardous chemicals in a warehouse, as described in this invention. Figure 2 A flowchart for acquiring 3D point cloud data and performing spatial noise reduction processing; Figure 3 A flowchart illustrating how to form a closed spatial segmentation polygon by connecting adjacent spatial topology transformation nodes; Figure 4 A bar chart comparing sensor response times in areas with different hazard levels; Figure 5 A comparison of the noise reduction effects on 3D point cloud data of hazardous chemical warehouses. Detailed Implementation

[0017] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] See Figure 1 This invention provides a method for safe storage management of hazardous chemical warehouses. The method includes: collecting three-dimensional point cloud data of the hazardous chemical warehouse; performing spatial noise reduction processing on the three-dimensional point cloud data to generate a purified spatial point cloud set; identifying high-density clustered areas in the purified spatial point cloud set, where the high-density clustered areas correspond to physical structures such as walls and shelves within the warehouse; extracting the outlines of internal physical partitions based on the high-density clustered areas; tracking the directional changes of the internal physical partition outlines; capturing the turning points of the directional changes at locations where the outline direction changes significantly, and marking these turning points as spatial topology transformation nodes; connecting adjacent spatial topology transformation nodes to form several closed spatial segmentation polygons, which divide the warehouse space into non-overlapping areas, and defining the closed spatial segmentation polygons as independent storage management domains. Each of the independent storage management domains obtained by traversing the division is used to generate a mesh patrol route with intersecting features within the independent storage management domain. The mesh patrol route is used to plan the movement path of the inspection robot or the line of sight coverage of fixed monitoring points. Sensor detection points are arranged at the geometric center of the mesh patrol route. Finally, the schemes of all management domains are integrated to output the overall safety management path scheme.

[0019] In one embodiment of the present invention, see [reference] Figure 2 A multi-angle lidar array is activated to emit detection pulse signals into the hazardous chemical warehouse, receiving echo intensity information reflected from various physical surfaces within the warehouse. The echo intensity information collected by lidar from different angles is aggregated, aligned and fused according to timestamps and spatial coordinates to construct an initial three-dimensional point cloud data frame sequence. The point distribution density of each frame in the three-dimensional point cloud data frame sequence is analyzed. Using statistical filtering or radius filtering algorithms, isolated points far from the main point cloud group are identified as isolated noise points, and false points caused by dust in the air or multiple reflections are identified as floating ghost points. These noise points and ghost points are removed from the data frames, retaining the point data stably attached to the surfaces of walls, shelves, containers, etc., completing the spatial noise reduction process and obtaining a purified spatial point cloud set.

[0020] In the specific implementation, a 3D point cloud data acquisition and spatial noise reduction process was carried out for a hazardous chemical warehouse with dimensions of 50 meters (length), 30 meters (width), and 10 meters (height). A multi-angle lidar array deployed at the four corners and the center of the top of the warehouse was activated. The lidar array emitted detection pulse signals with a wavelength of 905 nanometers into the warehouse, receiving echo intensity information and time-of-flight data reflected from the surfaces of shelves, walls, pipes, and storage containers. The echo intensity information and spatial coordinates collected from the lidar at five different locations were aggregated, and a unified coordinate system was established with one corner of the warehouse as the origin. Spatiotemporal registration and fusion were performed on all points to construct an initial 3D point cloud data frame sequence of 10 frames per second.

[0021] In some embodiments, the point distribution density of a frame in a 3D point cloud data frame sequence is analyzed. This frame contains 1.2 million original point cloud data points. The average distance between each point and its 50 nearest neighbors is calculated, and the global mean and standard deviation of the average distances of all points are calculated. A distance threshold is set as the global mean plus three times the standard deviation. Points with an average distance exceeding this threshold are identified and removed; these points are classified as isolated noise points. The determination of floating ghost points is based on echo intensity information. When the echo intensity of a point is lower than a preset intensity threshold of -10dB, and its spatial location is within the non-physical occlusion area of ​​the line connecting the LiDAR and the nearest physical surface, the point is determined to be a floating ghost point and removed from the data. After the above processing, the point data attached to the physical surface is retained, and spatial noise reduction is completed, resulting in a purified spatial point cloud set containing 900,000 points in one frame.

[0022] Optionally, in another embodiment, spatial noise reduction can be performed using voxel mesh downsampling combined with statistical filtering. First, each frame of the initial 3D point cloud data frame sequence is voxelized, with each voxel having a side length of 0.01 meters. Within each voxel, the centroid of all its points is used as an approximation, achieving initial downsampling and smoothing. Statistical outlier filtering is then applied to the downsampled point cloud, as expressed by the following formula:

[0023] Where: symbol Indicating the first point cloud A point and its nearest The average Euclidean distance between the nearest neighbors, denoted by . and They represent the first The point and the first The three-dimensional coordinate vector of a point, symbol This indicates the calculation of the Euclidean distance between two points. It calculates the Euclidean distance for all points. mean and standard deviation Set the judgment threshold as follows: ,parameter The noise level is set to 2.5 based on the scene noise level. For any point, if its corresponding... If the value exceeds this threshold, the point is considered an outlier and is removed. This method effectively filters out discrete noise points caused by measurement errors or environmental dust, retaining the point data attached to the surface of the object, thus completing spatial noise reduction.

[0024] In one embodiment of the present invention, the purified spatial point cloud set is voxelized, discretizing the continuous warehouse space into several regularly arranged cubic data units. The number of point clouds contained in each data unit is calculated, a point cloud number threshold is set, and data units with a point cloud number exceeding the preset threshold are filtered out. These dense data units are classified as high-density clustering regions. The outer points of the high-density clustering regions are extracted and used as edge points. A smooth curve is fitted to the edge points to obtain a continuous and smooth outline of the internal physical partition of the warehouse. Sampling points are selected at fixed intervals along the tangent direction of the internal physical partition outline, and the azimuth difference between adjacent sampling points is calculated. When the calculation finds that the absolute value of the azimuth difference between adjacent sampling points exceeds a 90-degree right angle, it is determined that the orientation change has occurred at that location. The coordinate information of the location where the orientation change occurs is recorded and stored in a node candidate list. (See also...) Figure 3 The return positions of all records are read from the candidate node list and uniformly renamed as spatial topology transformation nodes. The shortest connecting path between any two adjacent spatial topology transformation nodes is retrieved, and the shortest connecting path planning must avoid high-density clustered areas identified as walls or shelves. Connecting lines are drawn based on the calculated shortest connecting paths, and multiple connecting lines are connected end-to-end to form a closed spatial segmentation polygon.

[0025] In the specific implementation, a point cloud set of a cleaned-up hazardous chemical warehouse (50 meters long, 30 meters wide, and 8 meters high) containing 900,000 points was processed using voxelization. This process discretized the continuous three-dimensional space into cubic data units with sides of 0.05 meters, generating a total of 2.4 million data units. The number of points within each data unit was calculated, and a preset threshold of 10 was set. Data units with more than 10 points were identified and categorized as high-density clustered areas. In this implementation, high-density clustered areas correspond to the walls and fixed shelving structures within the warehouse. Edge points of these high-density clustered areas were extracted; these edge points are those located outside the high-density clustered areas and adjacent to low-density data units. B-spline-based smooth curve fitting was applied to these edge points to obtain continuous internal warehouse partition outlines, which were represented by an ordered sequence of two-dimensional planar coordinate points.

[0026] In some embodiments, the contour line is resampled at 0.1-meter intervals along the tangent direction of the internal partition contour of the warehouse to obtain a series of sampling points. The azimuth difference between adjacent sampling points is calculated. The azimuth difference is obtained by calculating the change in the angle between the direction of the line connecting adjacent sampling points and the reference direction. When the absolute value of the azimuth difference exceeds a 90-degree right angle, it is determined that the position of the sampling point has changed direction. The coordinates of the position where the direction has changed are recorded and stored in the node candidate list. The node candidate list stores the X, Y, and Z coordinates of all turning points in array form. In some embodiments, for a straight wall contour with a length of 15 meters, the calculated azimuth difference between adjacent sampling points is always within ±5 degrees and does not exceed a right angle, so the turning point is not marked. However, for an interior corner of the warehouse, the azimuth difference changes abruptly from 0 degrees to 120 degrees, and the absolute value exceeds a right angle, so the vertex of the interior corner is captured as the turning point. Optionally, all turnaround positions in the candidate node list are read and renamed as spatial topology transformation nodes. These nodes represent key turning points in warehouse spatial partitioning. The shortest connected path between these nodes is retrieved, calculated using the A* pathfinding algorithm. The shortest connected path must avoid the 3D space occupied by high-density clustered areas. Optionally, connecting lines are drawn based on the calculated shortest connected paths. These lines can be straight segments or polylines, and multiple connecting lines joined end-to-end form a closed spatial partitioning polygon. This closed spatial partitioning polygon is represented as a simple, non-self-intersecting polygon on a 2D plane. It can be understood that calculating the number of point clouds in the voxelization process is a crucial step. The choice of the point cloud number threshold affects the recognition accuracy of high-density clustered areas. The smooth curve fitting process uses an iterative nearest-point algorithm to optimize the arrangement of edge points, ensuring the smoothness and accuracy of the warehouse's internal solid partition contours. The calculation of the azimuth difference involves vector operations, expressed by the following formula:

[0027] Where: symbol Indicates the first The azimuth difference at each sampling point, sign and Indicates the first The x and y coordinates of each sampling point on a two-dimensional plane, with symbols... This represents the arctangent function in the four quadrants, used to calculate the azimuth angle of a vector. When... The absolute value is greater than When calculating radians, the direction of change at the sampling point is determined. It can be understood that for contours in 3D space, azimuth calculation can be projected onto a horizontal plane to simplify processing. Coordinate information in the node candidate list is deduplicated, merging spatially similar foldback positions to avoid generating redundant spatial topology transformation nodes. The drawing of the shortest connected path must consider path feasibility. Connecting lines may have height information in 3D space, but ultimately projected onto a 2D plane to form a closed spatial segmentation polygon.

[0028] In one embodiment of the present invention, the outer contour of the independent storage management domain is measured to obtain its long side dimension and short side dimension, and the ratio of the long side dimension to the short side dimension is calculated. Based on this ratio, a grid generation pattern is determined. If the ratio is greater than a preset ratio value, it indicates that the management domain is a long strip area. In this case, a main inspection trunk line is preferentially laid along the length direction, and then multiple branch inspection trunk lines are arranged perpendicular to the main inspection trunk line at even intervals to form a grid-like network. If the ratio is less than or equal to the preset ratio value, it indicates that the management domain is relatively square. In this case, the mesh inspection line is generated in a honeycomb arrangement to ensure that the lines are evenly distributed within the domain. The midpoint coordinates of each line segment in the mesh inspection line are calculated, and the midpoint coordinates of all line segments are used as candidate deployment locations. The shortest distance between each candidate deployment location and the boundary of the independent storage management domain is evaluated, and candidate deployment locations with distances less than the safe distance limit are eliminated to avoid sensors getting too close to obstacles or boundaries. The remaining candidate deployment locations after screening are finally confirmed as the sensor detection sites, and each sensor detection site is assigned a corresponding monitoring parameter type, such as temperature, gas concentration or image monitoring.

[0029] In practical implementation, for a defined rectangular independent storage management domain with a long side of 20 meters and a short side of 8 meters, the long side and short side dimensions of the independent storage management domain are measured, and the ratio R of the long side to the short side dimension is calculated to be 2.5. The preset ratio value is set to 1.5. Since the ratio R=2.5 is greater than the preset ratio value of 1.5, a grid-like network generation mode is adopted. A main inspection trunk line is laid along the length direction, running through the long side of the management domain. The main inspection trunk line is parallel to the long side of the management domain and located on the center line of the width direction of the management domain. Then, seven branch inspection trunk lines are evenly arranged perpendicular to the main inspection trunk line at intervals of 2.5 meters. The branch inspection trunk lines are parallel to the short side of the management domain. The main inspection trunk line and the branch inspection trunk lines intersect to form a mesh inspection line with intersection characteristics. The mesh inspection line covers the entire area of ​​the independent storage management domain.

[0030] In some embodiments, the selection of the preset ratio value is based on statistical analysis of various warehouse layouts. Setting the preset ratio value to 1.5 can effectively distinguish between elongated areas and approximately square areas. For an independent storage management domain with a long side dimension of 12 meters and a short side dimension of 10 meters, the ratio of its long side dimension to its short side dimension is 1.2. Since the ratio of 1.2 is less than the preset ratio value of 1.5, a honeycomb arrangement is used to generate a mesh inspection route. The honeycomb arrangement uses regular hexagons as the basic unit, with a side length of 1.5 meters. The basic units are arranged from the geometric center of the management domain towards the boundary, ensuring that the edge lines between the units form a continuous mesh inspection route. The honeycomb arrangement makes the mesh inspection route evenly distributed within the management domain, without large-area monitoring blind spots. For the selection of the mesh generation mode for management domains of different sizes, please refer to Table 1.

[0031] Table 1: Independent Storage Management Domain Mesh Generation Mode Selection Table Management Domain Number Long side dimension L (meters) Short side dimension W (meters) ratio L / W Preset ratio value Selected mesh generation mode A 20.0 8.0 2.50 1.5 Grid network (backbone + branches) B 12.0 10.0 1.20 1.5 honeycomb arrangement C 15.0 12.0 1.25 1.5 honeycomb arrangement Optionally, in the grid-like network generation mode, the interval distance I of the branch inspection line can be dynamically calculated based on the short side size W of the management domain, as expressed by the following formula:

[0032] Where: symbol Indicates the interval distance of the branch inspection line, symbol Indicates the shorter side dimension of an independent storage management domain, symbol This represents the preset number of branch lines, where N is a positive integer. Adjusting the value of N changes the density of the mesh patrol lines. Optionally, in the honeycomb arrangement, the side length E of the regular hexagonal basic unit is a configurable parameter. The value of side length E ranges from 1 meter to 3 meters. The smaller the value of side length E, the longer the total length of the generated mesh patrol lines and the higher the monitoring coverage density.

[0033] The midpoint coordinates of each line segment in the mesh inspection route are calculated, and these midpoint coordinates are used as candidate deployment locations. For the aforementioned grid network, a total of fifteen line segments are generated, and the planar coordinates of fifteen candidate deployment locations are calculated. The shortest Euclidean distance between each candidate deployment location and the boundary of the independent storage management domain is evaluated. A safety distance limit is set at 1.2 meters, and candidate deployment locations less than 1.2 meters from the boundary of the independent storage management domain are eliminated. After screening, three candidate deployment locations located near the edge of the management domain are eliminated. The remaining twelve candidate deployment locations are confirmed as sensor detection sites, which will be used to install physical sensors.

[0034] In some embodiments, corresponding monitoring parameter types are assigned to sensor detection sites. These monitoring parameter types include temperature, specific gas concentration, smoke concentration, and video images. The assignment principle is based on the location of the sensor detection site within the network patrol route. Sensor detection sites located at the intersection of main lines and branch lines are assigned video surveillance and combined gas monitoring, while sensor detection sites located only on branch lines are assigned temperature and smoke monitoring. It can be understood that the evaluation function Γ for the candidate deployment location and boundary distance can be expressed as:

[0035] Where: symbol Indicates coordinates as The shortest distance from the candidate deployment location to the boundary of the independent storage management domain, symbol This represents the coordinates of any point on the boundary line of an independent storage management domain. The calculation process traverses a series of sampling points on the boundary line to find the minimum value. It can be understood that when... When the calculated result is less than the set safe distance limit, the candidate deployment location is eliminated to ensure that the sensor detection site has sufficient installation and maintenance space and avoids being too close to the wall or shelf, which may affect the monitoring effect or make it easy to be bumped.

[0036] In one embodiment of the present invention, a list of hazardous chemicals actually stored within the independent storage management domain is obtained, and the hazard level identifier of each item in the list is parsed. Based on the parsed hazard level identifier, the density of the mesh patrol route is dynamically adjusted. For areas storing high-hazard level materials, the number of mesh patrol routes is increased, and the spacing between routes is shortened. Simultaneously, the trigger sensitivity threshold of the sensor detection points in that area is updated to adapt to the hazardous characteristics of the materials recorded in the hazardous chemicals list; for example, for flammable gases, the response speed of the concentration alarm threshold is improved.

[0037] In practice, for an independent storage management area numbered Z-05, a list of hazardous chemicals stored within that area is obtained. This list is stored in spreadsheet format and includes fields for material name, storage location, quantity, and hazard level identifier. The hazard level identifier field in the list is analyzed. Hazard levels are categorized into "high hazard," "medium hazard," and "low hazard" according to national standards. In this case, independent storage management area Z-05 stores ammonium nitrate and sodium chloride. Through analysis of the list, the overall hazard level of this management area is determined based on the stored ammonium nitrate, classifying it as a "high hazard" area.

[0038] Based on the hazard level identification obtained from the analysis, the density of the pre-generated mesh patrol routes within the independent storage management domain Z-05 is dynamically adjusted. For areas with a hazard level identification of "high hazard level," the number of mesh patrol routes is increased, specifically by shortening the average interval distance between path segments in the mesh patrol routes. If the base interval distance is D0, the adjustment formula for high-hazard level areas is:

[0039] Where: symbol This indicates the adjusted track gauge distance for high-risk areas, indicated by the symbol. Indicates the basic line spacing, symbol Density coefficient, representing the hazard level It is a real number greater than 1, and its value is directly related to the hazard level label. Simultaneously, for high-hazard areas, the trigger sensitivity thresholds at the sensor detection sites deployed in those areas are updated to match the hazardous characteristics of ammonium nitrate in the hazardous materials list. For example, for temperature sensor detection sites, the alarm threshold is lowered from the basic 60 degrees Celsius to 45 degrees Celsius; for smoke sensor detection sites, the alarm response threshold is adjusted from 15% light reduction per meter to 10% light reduction per meter.

[0040] In some embodiments, the correspondence between hazard level indicators, line density adjustments, and sensitivity adjustments can be defined using a preset configuration mapping table. Refer to Table 2, which illustrates the parameter adjustment strategies under different hazard level indicators.

[0041] Table 2: Mapping Table of Hazard Level Identification and Safety Parameter Adjustment Hazard level label Hazard level description Line density coefficient λ Temperature alarm threshold adjustment Smoke alarm threshold adjustment (light reduction rate / meter) High risk level Flammable, explosive, strong oxidizer, highly toxic 1.8 Lower by 15°C 10% Medium risk level Flammable, corrosive 1.3 Lower by 7°C 12% Low risk level Flame-retardant and low-toxicity 1.0 Base threshold 15% Optionally, the density enhancement operation can be achieved by re-executing the mesh inspection line generation algorithm and passing in the adjusted parameters. When generating the cellular mesh inspection line, the side length E of the regular hexagonal basic unit is scaled according to the hazard level label, using the following formula:

[0042] in: The adjusted side length, Based on the side length, This refers to the line density coefficient obtained from the table above. For grid-like networks, the interval distance I between branch inspection lines is adjusted. The new interval distance... The rearrangement of the mesh patrol routes ensures denser path coverage in high-risk areas, thereby supporting more frequent inspections or more detailed monitoring.

[0043] In some embodiments, the hazardous chemical inventory may contain materials with various hazard level labels, and these materials are stored in partitions within an independent storage management domain. In this case, after parsing the hazardous chemical inventory, the independent storage management domain can be further virtually divided into several sub-regions based on the actual storage location of the materials, and each sub-region is assigned an independent hazard level label. The layout density of the mesh patrol route and the trigger sensitivity threshold of the sensor detection points are adjusted differentially based on the hazard level of the sub-regions. It can be understood that the update of the trigger sensitivity threshold is accomplished by sending configuration instructions to the sensors deployed at the corresponding sensor detection points, and these configuration instructions contain new threshold parameters. It can be understood that by driving the dynamic adjustment of safety parameters through the hazardous chemical inventory, the safety management path scheme can adapt to the actual risk of the stored materials, implementing enhanced monitoring in high-risk areas and using basic configuration in low-risk areas, which helps optimize the allocation efficiency of monitoring resources.

[0044] See Figure 4 This is a bar chart comparing sensor response times in areas with different hazard levels. Temperature sensors have the fastest basic response (0.8s in low-risk areas), with the most significant improvement in response speed under high-risk scenarios. Gas sensors have the slowest basic response (1.2s in low-risk areas), but still maintain a fast response of 0.4s under high-risk scenarios, suitable for monitoring flammable and explosive gases. Smoke sensors have a response time between the two, consistent with the monitoring logic for warehouse fire early warning. Adjusting sensor sensitivity for areas with different hazard levels avoids resource waste in low-risk areas while ensuring no blind spots in high-risk areas. This intuitively verifies the effectiveness of the differentiated safety configuration strategy: enhanced monitoring in high-risk areas and basic configuration in low-risk areas, optimizing the allocation of monitoring resources. The effect of sensor sensitivity adjustment is quantified, providing data support for parameter configuration of the hazardous chemical warehouse safety management system.

[0045] In one embodiment of the present invention, real-time location feedback information transmitted by a mobile inspection device is received. This real-time location feedback information is compared with a preset global security management path scheme to calculate the location deviation. When the deviation exceeds the system's allowable error range, a temporary travel route for the mobile inspection device is replanned based on its current coordinates and the target point. This temporary travel route must avoid any dynamically identified obstacles, such as temporarily stored items or personnel. The location coordinates and outlines of the dynamic obstacles are recorded and marked on the cleaned spatial point cloud set. The point cloud set is temporarily supplemented, and the temporary model of the warehouse's internal solid partition outline is updated accordingly for subsequent path planning reference.

[0046] In practice, the mobile inspection device performs inspection tasks within the independent storage management domain (A-03) along a pre-defined global security management path. The global security management path specifies that the mobile inspection device should reach coordinates (12.5, 8.2, 0.0) at timestamp T1, while the real-time location feedback information transmitted back by the mobile inspection device via its onboard UWB positioning module is (12.1, 7.9, 0.0). The system receives the real-time location feedback information from the mobile inspection device, compares it with the planned position in the global security management path at the same time, and calculates the Euclidean distance deviation Δd between the two. The calculation formula is as follows:

[0047] Where: symbol This represents the three-dimensional spatial deviation between the real-time location feedback information and the planned location, with the symbol... The three-dimensional coordinates representing real-time position feedback information, with symbols... This represents the planned three-dimensional coordinates at the corresponding moment in the overall safety management path scheme. Substituting the specific values, the calculated deviation Δd is approximately 0.424 meters. The system's preset allowable error range is 0.5 meters. Since the calculated deviation of 0.424 meters does not exceed the allowable error range of 0.5 meters, the system does not trigger replanning.

[0048] In some embodiments, at another time T2, the real-time location feedback information of the mobile inspection device is (18.3, 5.1, 0.0), while the planned location of the global security management path scheme is (19.0, 5.0, 0.0). The calculated deviation Δd is approximately 0.7 meters. This deviation exceeds the allowable error range of 0.5 meters, and the system determines that the mobile inspection device has deviated from the predetermined path. The system then replans the temporary travel route of the mobile inspection device based on its current coordinates (18.3, 5.1, 0.0) and the coordinates of the next global path point (20.5, 5.0, 0.0). The lidar detects a dynamic obstacle near coordinates (19.5, 5.3, 0.0) in real time. This obstacle may be a temporarily placed cargo container. The replanned temporary route must bypass this obstacle. The coordinate sequence of the newly planned temporary route is: (18.3, 5.1) -> (18.8, 4.7) -> (20.0, 4.7) -> (20.5, 5.0). This path bypasses the area where the dynamic obstacle is located on the two-dimensional plane. Optionally, when the deviation exceeds the allowable error range and triggers replanning, the replanning algorithm adopts an improved D*Lite algorithm. The algorithm takes the current position of the mobile inspection device as the starting point, the next critical waypoint of the original path as the temporary endpoint, and the updated environmental occupancy grid map as the search space. The grid where the dynamic obstacle is located is marked as impassable, thus searching for a temporary route that avoids the dynamic obstacle. Record the location coordinates of the dynamic obstacle (19.5, 5.3, 0.0) and annotate its estimated outline with a bounding box (1.2m x 0.8m). Add this bounding box annotation to the cleaned spatial point cloud dataset as temporary obstacle point cloud data. Optionally, updating the outline of internal warehouse partitions does not permanently modify the original wall outlines. Instead, within the current operating cycle, the bounding box of the dynamic obstacle is treated as a temporary, impassable entity for path planning calculations. In the next system perception cycle, if the dynamic obstacle disappears, its corresponding temporary point cloud data and outline annotation will be removed.

[0049] In some embodiments, the allowable error range can be dynamically set according to the operating speed of the mobile inspection device and the hazard level of the area. In areas with higher operating speeds or higher hazard levels within an independent storage management area, the allowable error range can be set to a more stringent value, such as 0.3 meters. It is understood that the comparison between real-time location feedback information and the planned location is performed periodically, at a frequency of 10 times per second. After each comparison, the real-time Δd value is calculated. After replanning the temporary travel route, the mobile inspection device will receive new path instructions and track the new path. Recording the location of dynamic obstacles includes not only their center coordinates but also the size, height, and timestamp of their point cloud clusters. This information is annotated on the cleaned spatial point cloud set, forming a time-attributed obstacle layer for short-term path planning and conflict analysis. It is understood that the updated warehouse interior solid partition outline is a comprehensive representation integrating static structures and dynamic temporary obstacles. This representation is used to provide an environmental model for subsequent real-time path planning by the mobile inspection device or other devices until the dynamic obstacle is detected and removed by the system.

[0050] See Figure 5 This is a comparison chart showing the denoising effects of 3D point cloud data from a hazardous chemicals warehouse. The original data exhibits drastic fluctuations, with peak values ​​exceeding 110 points / m², and its overall density is significantly higher than the denoised data. This is because the original data contains a large amount of invalid information such as floating noise points and isolated ghost images. The noise level feature remains consistently low, within the 0-25 point / m² range, representing the baseline intensity of environmental noise within the warehouse and serving as the core target for the denoising algorithm. The denoised data has a lower overall feature density than the original data, with smoother fluctuations. It accurately eliminates the false high-density peaks caused by noise, retaining only the point cloud data of the actual warehouse structure, such as partitions and shelves. The overall trend of the green curve after denoising is highly consistent with the original blue curve, indicating that while removing noise, the true outline information of the physical partitions within the warehouse is fully preserved.

[0051] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A method for safe management of hazardous chemical warehouse storage, characterized in that, The method includes: Collect three-dimensional point cloud data of a hazardous chemical warehouse, perform spatial noise reduction processing on the three-dimensional point cloud data, and generate a purified spatial point cloud set; Identifying high-density clustered regions in the purified spatial point cloud set, and extracting the outline of the internal physical partitions of the warehouse based on the high-density clustered regions, includes: performing voxelization processing on the purified spatial point cloud set to discretize the continuous space into several regularly arranged data units; calculating the number of point clouds in each data unit, filtering out the data units with a number of point clouds exceeding a preset threshold, and classifying them as the high-density clustered regions; extracting the edge points of the high-density clustered regions, and performing smooth curve fitting on the edge points to obtain the continuous outline of the internal physical partitions of the warehouse. Track the changes in the outline of the physical partitions inside the warehouse, capture the turning points of the changes in the outline, and mark the turning points as spatial topology transformation nodes; Connect adjacent spatial topology transformation nodes to form several closed spatial partitioning polygons, and define the closed spatial partitioning polygons as independent storage management domains; Traverse the independent storage management domain, generate a mesh patrol route with intersecting features within the independent storage management domain, and arrange sensor detection points at the geometric center of the mesh patrol route to output a global security management path scheme. The process of collecting 3D point cloud data from the hazardous chemical warehouse and performing spatial noise reduction on the 3D point cloud data includes: activating a multi-angle lidar array to emit detection pulse signals into the interior of the hazardous chemical warehouse and receiving the reflected echo intensity information; summarizing the echo intensity information to construct an initial 3D point cloud data frame sequence; analyzing the point distribution density of each frame in the 3D point cloud data frame sequence, removing isolated noise points and floating ghost points, and retaining the point data attached to the solid surface to complete the spatial noise reduction process. Tracking the change in the orientation of the physical partition outline inside the warehouse and capturing the turning point of the change in orientation includes: calculating the azimuth difference between adjacent sampling points along the tangent direction of the physical partition outline inside the warehouse; determining that the orientation change has occurred at the location when the absolute value of the azimuth difference exceeds a right angle; recording the coordinate information of the turning point where the orientation change occurs, and storing the coordinate information in the node candidate list.

2. The method for safe management of hazardous chemical warehouse storage as described in claim 1, characterized in that, Connecting adjacent spatial topology transformation nodes forms several closed spatial partitioning polygons, including: Read the turnaround position from the node candidate list and rename it to the spatial topology transformation node; Find the shortest connecting path between the spatial topology transformation nodes, and the shortest connecting path must avoid the high-density clustered area; Based on the shortest connected path, draw connecting lines, and form a closed spatial segmentation polygon by connecting multiple connecting lines end to end.

3. The method for safe management of hazardous chemical warehouse storage as described in claim 2, characterized in that, Traversing the independent storage management domain, generating a mesh inspection path with intersecting characteristics within the independent storage management domain, including: Measure the long side and short side dimensions of the independent storage management domain, and determine the grid generation mode based on the ratio of the long side to the short side dimensions; If the ratio is greater than the preset ratio, the main inspection trunk line is laid first along the length direction, and then the branch inspection trunk line is arranged perpendicular to the main inspection trunk line. If the ratio is less than or equal to a preset ratio, the mesh inspection route is generated in a honeycomb pattern to ensure that the mesh inspection route covers the entire area of ​​the independent storage management domain.

4. The method for safe management of hazardous chemical warehouse storage as described in claim 3, characterized in that, Sensor detection points are arranged at the geometric center of the mesh inspection route, including: Calculate the midpoint coordinates of each line segment in the network patrol route, and use the midpoint coordinates as candidate deployment locations; Evaluate the distance between the candidate deployment locations and the boundary of the independent storage management domain, and eliminate candidate deployment locations whose distance is less than the safe distance limit; The remaining candidate deployment locations are identified as the sensor detection sites, and corresponding monitoring parameter types are assigned to the sensor detection sites.

5. The method for safe management of hazardous chemical warehouse storage as described in claim 4, characterized in that, Also includes: Obtain the list of hazardous chemicals stored in the independent storage management domain, and parse the hazard level identifiers in the list of hazardous chemicals; Based on the hazard level markings, adjust the density of the mesh patrol routes, and increase the number of mesh patrol routes in high-hazard areas; The trigger sensitivity threshold of the sensor detection site is updated synchronously to adapt it to the hazardous characteristics of the hazardous chemical material list.

6. The method for safe management of hazardous chemical warehouse storage as described in claim 5, characterized in that, Following the step of outputting the global security management path scheme, the following is also included: Receive real-time location feedback information from mobile inspection equipment and compare the deviation between the real-time location feedback information and the overall security management path scheme; When the deviation exceeds the allowable error range, the temporary travel route of the mobile inspection equipment is replanned, and the temporary travel route must bypass dynamic obstacles; Record the location of the dynamic obstacle and mark it on the cleaned spatial point cloud set to update the outline of the internal solid partition of the warehouse.

7. A safety management system for hazardous chemical warehouse storage, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method for safe management of hazardous chemical warehouse storage as described in any one of claims 1 to 6.

Images