Overhead deployment construction safety management method based on ground oblique photography

Abstract

The invention relates to the field of building construction, and discloses an upward deployment construction safety management method based on ground oblique photography, which comprises the following steps: acquiring multi-angle ground oblique photography image data of a construction site, and constructing a high-density point cloud model in combination with a visual angle optimization strategy and a region shielding compensation mechanism; performing spatial semantic segmentation on the high-density point cloud model, extracting key component information of an upper deployment area and a lower operation area in the high-density point cloud model by using a structure recognition algorithm, and constructing a spatial distribution map reflecting a component spatial relationship; based on the spatial distribution map, a multi-level safety evaluation algorithm is adopted to analyze the deployment and operation area with spatial intersection, and a potential interference area is identified; introducing a spatio-temporal evolution modeling method, and carrying out dynamic analysis on continuously updated oblique photography images in the construction process; and carrying out dynamic risk grade division on the current deployment scheme based on the safety early warning event set. The method has the advantage of improving the safety management level.

Description

Technical Field

[0001] The present invention relates to the field of construction, and specifically to a safety management method for above - below allocation construction based on ground oblique photography. Background Art

[0002] With the continuous improvement of the intelligent level of construction site management, construction monitoring means based on image recognition and 3D modeling have been gradually widely used. Among them, ground oblique photography technology has played an important role in spatial surveying and mapping of construction sites, construction progress tracking, and safety risk assessment due to its advantages such as convenient deployment, high imaging accuracy, and fast modeling efficiency. This technology sets multi - angle oblique photography equipment on the ground to collect multi - perspective image data of the operation area, and combines image registration and structure reconstruction algorithms to generate a 3D model with real scale to assist on - site managers in construction organization and safety analysis. However, in the actual application process, the existing construction management methods based on ground oblique photography still have certain limitations when dealing with above - below allocation operations. Especially in multi - layer structures or high - formwork areas, due to factors such as building occlusion, shooting angle limitations, and insufficient local lighting, the collected image data often has spatial coverage blind spots, making it impossible to accurately model some key areas. Specifically, when the upward viewing angle of the oblique photography system is limited, it is difficult to clearly capture the spatial relationship between the high - altitude working surface and the lower working area, resulting in the inability to accurately evaluate the potential safety interference to the lower working area when arranging above - below allocation tasks. For example, when a temporary hoisting path needs to cross a densely form - worked area or intersect with a personnel passage area, the existing system is difficult to timely detect the existing spatial conflicts, thus affecting the timeliness and accuracy of construction safety control. Therefore, it is necessary to design a safety management method for above - below allocation construction based on ground oblique photography to improve the safety management level. Summary of the Invention

[0003] Aiming at the deficiencies of the prior art, the present invention provides a safety management method for above - below allocation construction based on ground oblique photography, which has the advantage of improving the safety management level and solves the problems in the above - mentioned background art.

[0004] To achieve the above - mentioned purpose of improving the safety management level, the present invention provides the following technical solution: A safety management method for above - below allocation construction based on ground oblique photography, including the following steps: Obtain ground oblique photography image data from multiple angles at the construction site, and combine the perspective optimization strategy and the regional occlusion compensation mechanism to construct a high - density point cloud model; Perform spatial semantic segmentation on the high - density point cloud model, and use the structure recognition algorithm to extract the key component information of the above - below allocation area and the lower working area in the high - density point cloud model, and construct a spatial distribution map reflecting the spatial relationship of the components; Based on the spatial distribution map, a multi-level security assessment algorithm is used to analyze the allocation and operation areas with spatial intersections, and potential interference areas are identified; Introduce a spatio-temporal evolution modeling method to dynamically analyze the continuously updated oblique photography images during the construction process. Based on the potential interference areas, through the interference event triggering mechanism, the actual safety conflict points during the construction process are identified, and a safety warning event set is formed; Based on the safety warning event set, a dynamic risk level division of the current allocation plan is carried out. Combining the allocation path planning and the operation time window, an allocation safety management report is output.

[0005] Preferably, the process of constructing a high-density point cloud model by combining the perspective optimization strategy and the regional occlusion compensation mechanism is as follows: Perform multi-angle registration on the ground oblique photography images, and optimize the perspective layout of the reconstructed model through a multi-perspective fusion algorithm; Predict the boundaries and reconstruct and compensate the structure missing areas caused by the occlusion of adjacent components in the images; Based on the reconstruction accuracy evaluation function, dynamically adjust the fusion parameters to optimize the consistency and density distribution of the point cloud data reconstruction; After the fusion is completed, a refined interpolation method is used to supplement the point cloud void areas, and finally a high-density point cloud model with high spatial resolution and complete structural information is formed.

[0006] Preferably, the process of performing spatial semantic segmentation on the high-density point cloud model is as follows: Use a multi-scale three-dimensional convolutional neural network to extract features from the point cloud data and predict spatial labels, and label different types of structural components; For the areas in the point cloud data where the construction boundaries are not clearly defined, combined with the context structure relationship of the structural components in the construction scene, refine the labels of the boundary fuzzy areas; Group the point cloud areas with similar semantic labels through a clustering algorithm, and optimize the boundary morphology using the shape preservation constraint, and output the point cloud component set after semantic segmentation.

[0007] Preferably, the process of using a structure recognition algorithm to extract the key component information of the upper allocation area and the lower operation area in the high-density point cloud model is as follows: Through the point cloud morphology analysis algorithm, identify the structural components with characteristic dimensions and spatial direction characteristics, and divide them into components in the upper allocation area and components in the lower operation area according to the spatial distribution; Combined with the construction design drawings and the component information database, compare and confirm the geometric parameters of the extracted components; Adopt a topological relationship analysis algorithm to establish a relative spatial relationship model between components, and label whether there are crossing, overlapping or attachment relationships between components; Finally, a set of component information with spatial attribute tags is generated.

[0008] Preferably, the process of constructing a spatial distribution map reflecting the spatial relationships of components is as follows: Perform three-dimensional mapping of the key components for upper-level allocation and lower-level operation according to their spatial coordinates and height differences, and establish a spatial layout model; Based on the relative position relationships and movement path rules among components, construct a spatial interference judgment matrix to record the spatial ranges affected or influenced by each component; Introduce a dynamic weight factor for components, and combine the movement frequency, task priority, and operation time period of components to adjust the influence levels of each component in the spatial map, forming a spatial distribution map.

[0009] Preferably, the said.

[0010] Preferably, the process of analyzing the allocation and operation areas with spatial intersections using a multi-level safety assessment algorithm is as follows: Based on the spatial attributes and interaction relationships of components marked in the spatial distribution map, conduct a first-order spatial intersection detection for all pairs of components to identify the set of pairs of components that overlap or are close to intersecting during the construction period; For the identified pairs of components, combine the construction task schedule and operation plan to construct a task overlap matrix in the time dimension, forming a second-order spatio-temporal intersection relationship graph; On the basis of the second-order relationship graph, introduce the task importance coefficient and construction stage risk coefficient to establish a third-order safety risk assessment model; According to the calculation results of the third-order model, assign corresponding risk level tags to each pair of components with potential conflicts, and output the priority ranking of component conflicts.

[0011] Preferably, the process of identifying potential interference areas is as follows: Based on the risk level tags and the priority ranking of component conflicts output in the multi-level safety assessment algorithm, extract the spatial ranges of the areas where pairs of components with high risk levels are located as a preliminary candidate set of potential interference areas; Conduct an expansion analysis of the areas in the candidate set, and combine the component activity trajectories and dynamic offset models to evaluate the influence radius of the allocation equipment under different operation error conditions; After external variables, dynamically correct the area boundaries to form potential interference areas.

[0012] Preferably, the process of introducing a spatio-temporal evolution modeling method to dynamically analyze oblique photography images is as follows: Collect ground oblique photography image data at the construction site, register and compare it with historical images to generate an image sequence within a continuous time period; Based on the morphological and spatial position changes of components in the image sequence, construct a temporal model of component states to track the spatial trajectories and state transition processes of each component; Use spatio-temporal graph convolutional networks to model the evolution trend of component states and extract key node changes and abnormal spatio-temporal behavior patterns; Combined with the set of potential interference regions, set the triggering conditions for interference events, monitor the construction site images in real time and compare with the triggering conditions; Once the triggering conditions are met, automatically mark the conflict positions and related components in the current image frame, generate the corresponding interference events, and record them as early warning event candidates.

[0013] Preferably, the process of forming the safety early warning event set is as follows: For the identified early warning event candidates, calculate the weight values of the early warning events by combining the importance of components, the urgency of deployment tasks, and historical conflict data; According to the weight values of the early warning events and the event occurrence frequencies, classify and screen the events, and retain the high-risk and medium-risk events as formal safety early warning events; Record the detailed attributes of the formal early warning events; Integrate multi-source data, establish a unified early warning event management database, index and classify the events and update them dynamically to form a safety early warning event set.

[0014] Compared with the prior art, the present invention provides a safety management method for overhead deployment construction based on ground oblique photography, which has the following beneficial effects: 1. Through ground oblique photography, comprehensively collect multi-angle data of the construction site, introduce a perspective optimization strategy to improve the modeling accuracy of the edge area, and effectively repair the missing point clouds caused by component occlusion in combination with the occlusion compensation mechanism, and finally generate a three-dimensional point cloud model with strong structural continuity and high resolution, providing a complete and reliable data basis for subsequent semantic analysis and spatial modeling.

[0015] 2. Through deep learning and structure recognition algorithms, intelligently analyze the point cloud data, accurately label the component types and distinguish the components in the overhead deployment and lower operation areas, which helps to construct a three-dimensional map reflecting the spatial positions, directions and association relationships of the components, providing structured and computable spatial information support for subsequent conflict detection and risk assessment.

[0016] 3. Through safety assessment algorithms, conduct multi-level analysis on the possible spatial interference between the deployed components and the operating components, not only can identify static overlaps, but also can evaluate the dynamic intersection risks in combination with the task plan, discover high-risk areas in advance, and realize the proactive safety interference prediction before construction.

[0017] 4. By introducing spatio-temporal evolution modeling, dynamically analyze the spatial trajectories and state changes of components at different construction stages, and combined with the interference event triggering mechanism, it is possible to perceive conflict risks in real time during construction, accurately capture and record potential dangerous events, and provide intelligent and dynamic early warning capabilities for construction monitoring.

[0018] 5. By integrating safety warning data and deployment task planning, using a multi-objective optimization algorithm to conduct risk assessment and grading for different deployment plans, it is possible to rank the pros and cons of the plans, recommend path adjustment suggestions and operation windows, so as to assist construction management personnel in formulating safer and more reasonable deployment plans and comprehensively improving the risk control level during the construction deployment stage. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a schematic diagram of the method of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0020] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0021] Embodiment 1 Please refer to Figure 1 As shown, a method for safety management of overhead deployment construction based on ground oblique photography according to an embodiment of the present invention includes the following steps: S1: Obtain ground oblique photography image data from multiple angles at the construction site, and combine the perspective optimization strategy and the regional occlusion compensation mechanism to construct a high-density point cloud model.

[0022] The process of constructing a high-density point cloud model by combining the perspective optimization strategy and the regional occlusion compensation mechanism in S1 is as follows: Perform multi-angle registration on the ground oblique photography images, optimize the perspective layout of the reconstructed model through a multi-view fusion algorithm, and improve the reduction accuracy of edge components; use the image feature point matching algorithm to register the overlapping areas in the ground oblique photography images, and globally optimize the perspective poses in combination with the bundle adjustment algorithm; perform pixel-level depth estimation on the registered multi-angle images through the multi-view geometric consistency principle; in the depth map fusion stage, adopt a weight weighted fusion strategy to improve the credibility of depth calculation in the image edge area; automatically adjust the perspective shooting density according to the geometric complexity of the components, and generate a higher density of perspective coverage in the edge detail area to improve the three-dimensional reduction accuracy of the edge components; Perform boundary prediction and reconstruction compensation for the structural missing areas in the image caused by the occlusion of adjacent components; identify the component boundary information in the occluded edge area based on the semantic segmentation model; for the occluded area, use the shape completion network to geometrically infer the missing structure; combine the scene context information and the known component template library to perform boundary shape matching and selection of completion candidates; evaluate the errors of multiple reconstruction candidate schemes, and select the result with the most consistent geometric relationship with the surrounding reconstructed components as the final compensation data to input into the point cloud reconstruction. Based on the reconstruction accuracy evaluation function, dynamically adjust the fusion parameters to optimize the consistency and density distribution of the point cloud data reconstruction; construct a composite accuracy evaluation function with depth error, geometric consistency error, and point cloud density uniformity as the core indicators; during the point cloud reconstruction process, periodically evaluate the accuracy index values of the currently generated point cloud and compare them with the expected accuracy threshold; if there is a decrease in accuracy or uneven density in a local area, perform dynamic correction by adjusting the pixel confidence weight, view angle weighting coefficient, and occlusion penalty factor during the fusion process; use the adaptive fusion window adjustment algorithm to enable a smaller fusion window in geometrically complex areas to capture details, and expand the fusion window in flat areas to improve the reconstruction efficiency; after the fusion is completed, use a refined interpolation method to supplement the void areas in the point cloud, and finally form a high-density point cloud model with high spatial resolution and complete structural information; perform void detection on the sparse areas in the point cloud to determine whether the voids are caused by occlusion or view blind spots; based on the void position and surrounding geometric information, estimate the missing point positions using the interpolation algorithm based on local plane fitting; introduce an interpolation strategy based on triangular mesh reconstruction in complex structural areas, reconstruct the point cloud into a surface mesh and then perform mesh refinement and anti-sampling; fuse the supplemented points generated after interpolation with the original point cloud to ensure density continuity and component shape consistency, and output a complete high-density point cloud model for subsequent spatial analysis.

[0023] S2: Perform spatial semantic segmentation on the high-density point cloud model, and use the structure recognition algorithm to extract the key component information of the upper deployment area and the lower operation area in the high-density point cloud model, and construct a spatial distribution map reflecting the spatial relationship of the components.

[0024] The process of performing spatial semantic segmentation on the high-density point cloud model in S2 is as follows: Use a multi-scale three-dimensional convolutional neural network to extract features and predict spatial labels from the point cloud data, and label different types of structural components, such as steel beams, brackets, lifting devices, etc. For the regions with unclear boundary transitions in the point cloud data, combined with the context structure relationship of structural components in the construction scenario, refine the labels for the boundary-blurred regions; construct a point cloud adjacency graph and calculate the context neighborhood of each point in space; use the scene structure atlas to identify the spatial arrangement and combination patterns of common components (for example, steel beams usually span between two columns, and lifting devices are generally suspended in the upper area); for the regions with low prediction confidence or fuzzy category boundaries, adjust the votes based on the dominant labels in the context semantic neighborhood. Group the point cloud regions with similar semantic labels through a clustering algorithm, and optimize the boundary morphology using shape-preserving constraints; adopt the DBSCAN algorithm to perform spatial clustering in the point cloud subset with consistent semantic labels to identify the point cloud clusters corresponding to individual components; perform fitting analysis on the boundaries of each point cloud cluster to identify the positions of geometric shape mutations; for the boundary regions where multiple components overlap, preferentially use the dissection method based on component template matching for separation processing to ensure the accuracy of the classification granularity. Output the point cloud component set after semantic segmentation to provide the input data basis for subsequent structure recognition and spatial relationship modeling; attach the semantic annotation and clustering segmentation results as attributes to the point cloud, and each point contains spatial coordinates, semantic category, and component attribution number; the output format includes standardized *.ply, *.pcd, or custom BIM structural point cloud data format.

[0025] The process of using the structure recognition algorithm to extract the key component information of the upper deployment area and the lower operation area in the high-density point cloud model in S2 is as follows: Through the point cloud morphology analysis algorithm, identify the structural components with characteristic dimensions and spatial direction features, and divide them into components in the upper deployment area and components in the lower operation area according to their spatial distribution; perform voxelization processing on the high-density point cloud model, and use principal component analysis to extract the main direction, size distribution, and geometric shape features of the local area; identify the components with long, plate-like, or columnar morphological features, and preliminarily screen the candidate point sets of the components; by calculating the Z-axis height of the centroid coordinates of each component, the angle between the main axis direction and the ground, judge the spatial height level where the parameters are located; according to the construction site zoning standard, set the Z-axis threshold or spatial bounding box, and divide the components higher than the critical height into components in the upper deployment area, and those lower than the threshold into components in the lower operation area. Combined with the construction design drawings and component information database, the extracted component geometric parameters are compared and confirmed to ensure the accuracy of recognition; the geometric templates of known components are extracted from the construction BIM model, including length, width, height, tolerance range, material information, etc.; bounding box fitting is performed on the identified point cloud components to extract their geometric properties and spatial posture; the geometric matching algorithm is used to calculate the similarity between the point cloud components and the drawing components; the similarity threshold is set and manual or intelligent auxiliary review is performed to confirm the component category and attribute labeling, eliminate recognition errors or redundant objects, and improve recognition accuracy; Using topological relationship analysis algorithms, a relative spatial relationship model between components is established to mark whether there is a crossing, overlapping or dependent relationship between components; based on the positional relationship between component bounding boxes, a spatial adjacency graph is constructed to record the relative distance, direction and contact area between components; for intersecting or adjacent component pairs, Boolean operations are used to detect whether there is a spatial crossing, partial overlap or surface contact phenomenon; the structural connection mode between components is defined, and the relationship at the boundary is semantically interpreted; the topological analysis results are attached to the component information and recorded in the form of a logical structure tree or a graph structure; Finally, a component information set with spatial attribute labels is generated, which provides the core elements for constructing a spatial distribution map. The identified component set is organized into a standard data structure, including component ID, category label, spatial position, geometric parameters, region and topological relationship. A structured point cloud component database is constructed to support query and indexing. The visual component layer is output and can be superimposed on the three-dimensional point cloud for spatial distribution map display.

[0026] The process of introducing the timing marking mechanism in S2 to mark the occurrence time of key deployment actions is as follows: The lifting, moving and installation time information of each key component is obtained through the construction task scheduling system, and associated with the deployment path in the map; Use construction site video surveillance or continuous oblique photography image sequences to perform action recognition and time matching of the actual operation process; A timing marking mechanism is introduced to mark key actions such as lifting start, path conversion, and installation as independent time nodes.

[0027] The process of constructing the spatial distribution map reflecting the spatial relationship of components in S2 is as follows: Three-dimensionally map the key components of the upper deployment and the lower operation according to their spatial coordinates and height differences to establish a spatial layout model; extract the spatial information of the extracted key components, including the centroid coordinates, geometric dimensions, and main axis directions. By statistically calculating the Z-axis heights of all components and the spatial dimensions of the component bodies, calculate their absolute height levels relative to the ground, and combine the construction area division criteria to divide the components into components in the upper deployment area and components in the lower operation area. Map these components to a unified three-dimensional space coordinate system and generate a spatial layout model that includes component numbers, spatial positions, dimensions, and area attributes; Based on the relative position relationship and movement path rules between components, construct a spatial interference judgment matrix to record the spatial range that each component may affect or be affected; generate a bounding box model for each component and expand the dynamic movement path area to determine whether the area intersects with the static volume or path of other components. If there is an intersection, it is considered that there is a potential interference relationship between the two components. Record the interference relationships between all components in pairs in matrix form to form a spatial interference judgment matrix, which is used to represent possible occlusion, collision, or interference between components; Introduce a component dynamic weight factor, and combine the component movement frequency, task priority, and operation period to adjust the influence level of each component in the spatial map; the weight factor consists of three parts: the movement frequency of the component, the task priority of the component in the overall construction process, and the operation period when the component is active. By setting reasonable weighting coefficients to comprehensively calculate the above factors, obtain the dynamic influence values of each component, and accordingly adjust the influence level of the component in the spatial map to more realistically reflect the spatio-temporal interaction relationship in the construction site; Form a spatial distribution map that includes component spatial attributes, interaction relationships, and dynamic parameters, providing data support for subsequent safety assessment and spatio-temporal modeling.

[0028] Embodiment 2 As Figure 1 shown, a safety management method for upper deployment construction based on ground oblique photography further includes the following steps: S3: Based on the spatial distribution map, use a multi-level safety assessment algorithm to analyze the deployment and operation areas with spatial intersections to identify potential interference areas.

[0029] The process of using the multi-level safety assessment algorithm to analyze the deployment and operation areas with spatial intersections in the above S3 is as follows: Based on the spatial attributes and interaction relationships of components marked in the spatial distribution map, first perform a first-order spatial intersection detection on all component pairs to identify the set of component pairs that overlap or are close to intersection during the construction period; based on the constructed spatial distribution map, extract the spatial attributes of each component, including its three-dimensional coordinates, geometric dimensions, movement paths, and other information; use the AABB collision test algorithm to perform geometric overlap on each component pair one by one to identify the component pairs that have spatial intersection or intersection risk within the construction time window, which is recorded as the first-order spatial conflict set , where i and j are component numbers. If the bounding box of component i and the bounding box of component j meet any of the following conditions, it is determined to have a spatial intersection risk: indicates the existence of geometric overlap; indicates that the geometric distance is less than the set distance; For the identified component pairs, further combine the construction task schedule and operation plan to construct a task overlap matrix in the time dimension to form a second-order spatio-temporal intersection relationship graph; after obtaining the set of spatial intersection component pairs, further combine the construction task schedule and operation sequence data to extract the specific construction task time range participated by the components , for each component pair , judge whether there is an overlap in the task time interval: > and > ; if satisfied, the component pair has a task intersection risk in space-time, and is included in the second-order relationship graph , where the node V is the component set, and the edge is the second-order spatio-temporal conflict relationship; Based on the second-order relationship graph, introduce the task importance coefficient and the construction stage risk coefficient to establish a third-order safety risk assessment model; based on the second-order spatio-temporal conflict graph, introduce two weight factors for each component pair: Task importance coefficient : Represents the importance degree of the task where the component is located, and is assigned according to the task type; Construction stage risk coefficient : Represents the overall safety sensitivity of the current construction stage, such as the high-altitude operation period, the intensive operation period, etc.; Combined with the spatial overlap intensity , , construct a third-order risk assessment function, and the formula is: In the formula, is the comprehensive safety risk degree of the component pair (i,j); is the task importance coefficient corresponding to component j; According to the calculation results of the third-order model, assign corresponding risk level labels to each pair of components with potential conflicts, and output the priority ranking of component conflicts, providing a basis for subsequent interference area identification and interference event modeling; set risk level labels, for example, high risk is greater than 0.7, medium risk is 0.4 - 0.7, and low risk is less than 0.4. Label all pairs of components with conflict risks as high risk, medium risk, or low risk according to the level, and generate a conflict risk priority queue.

[0030] The process of identifying potential interference areas in S3 is as follows: Based on the risk level labels and component conflict priority ranking output in the multi-level security assessment algorithm, extract the spatial range of the area where high-risk level component pairs are located as the preliminary potential interference area candidate set; Conduct an expansion analysis on the areas in the candidate set, and combine the component activity trajectories and dynamic offset models to evaluate the influence radius of the deployment equipment under different operation error conditions; After considering external variables such as operation errors, weather factors, and personnel collaboration effects, dynamically correct the area boundaries to form a complete set of potential interference areas; Mark the potential interference areas in the three-dimensional construction scene model, providing a basic area framework for dynamic spatio-temporal evolution modeling and early warning event extraction.

[0031] S4: Introduce a spatio-temporal evolution modeling method to dynamically analyze the continuously updated oblique photography images during the construction process. Based on the potential interference areas, through the interference event trigger mechanism, identify the actual safety conflict points during the construction process and form a safety early warning event set.

[0032] The process of introducing a spatio-temporal evolution modeling method to dynamically analyze the oblique photography images in S4 is as follows: Regularly collect the latest ground oblique photography image data at the construction site, register and compare it with historical images to generate an image sequence within a continuous time period; Based on the changes in component morphology and spatial position in the image sequence, construct a component state time series model to track the spatial trajectories and state transition processes of each component; perform three-dimensional reconstruction and component segmentation on each frame of the image, extract the component set, and record the geometric state and spatial coordinates of the component set at time to construct a component state time series model; Use a spatio-temporal graph convolutional network to model the component state evolution trend, extract key node changes and abnormal spatio-temporal behavior patterns; construct a spatio-temporal graph structure, input the graph structure and component state vectors into the ST-GCN model for learning; the model outputs the evolution prediction trend of the component, including potential movement trajectories, state transition probabilities, and abnormal behavior pattern recognition; Combined with the aforementioned set of potential interference regions, set the triggering conditions for interference events, such as component position approaching, movement direction convergence, operation time overlap, etc., and monitor the construction site images in real time and compare with the triggering conditions; Once the triggering condition is met, automatically mark the conflict positions and related components in the current image frame, generate the corresponding interference event, and record it as a candidate warning event.

[0033] The process of forming the safety warning event set in S4 is as follows: For the identified candidate warning events, calculate the weight values of the warning events in combination with component importance, deployment task urgency, and historical conflict data; According to the weight values of the warning events and the event occurrence frequency, classify and screen the events, and retain the high-risk and medium-risk events as formal safety warning events; Record the detailed attributes of the formal warning events, including the occurrence time, involved components, warning level, expected duration, and possible impact range; Integrate multi-source data, establish a unified warning event management database, and index, classify, and dynamically update the events; Form a structured safety warning event set, providing input for the risk response and plan optimization module.

[0034] S5: Based on the safety warning event set, dynamically divide the risk levels of the current deployment plan, and combine the deployment path planning and operation time window to output a deployment safety management report.

[0035] The process of dynamically dividing the risk levels of the current deployment plan in S5 is as follows: Receive the safety warning event set and analyze whether the identified high-risk event nodes are included in the current deployment path; Based on the number, level, and impact range of the warning events, evaluate each segment of the deployment path one by one, and use a dynamic risk scoring function to calculate the risk value of each segment. The formula is: In the formula, is the risk level weight of the event, is the overlapping ratio of the event impact area and the path segment, is the overlapping degree of the event action duration and the path passing period; Based on the comprehensive risk score, combined with the timing arrangement of the deployment task and the resources in the operation window, judge the feasibility and substitutability of each path segment; Introduce a multi-objective optimization algorithm to find a balance among risk minimization, optimal resource utilization, and schedule guarantee, output multiple optional deployment plans and mark the risk levels; construct a multi-objective optimization model, and the objective function formula is: In the formula, is the risk value of the path segment, is the resource conflict, is the total optimized deployment duration, is the originally planned deployment duration; Finally, based on the result of the plan sorting, a deployment safety management report is generated, which includes the risk level assessment result of the current deployment task, the recommended adjusted path and time window, and the suggestions for dealing with key warning events.

[0036] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article or device.

[0037] Although the embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention, and the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A safety management method for upper deployment construction based on ground tilt photography, characterized in that, It includes the following steps: Obtain the ground tilt photography image data from multiple angles at the construction site, and combine the perspective optimization strategy and the regional occlusion compensation mechanism to construct a high-density point cloud model; Perform spatial semantic segmentation on the high-density point cloud model, and use the structure recognition algorithm to extract the key component information of the upper deployment area and the lower operation area in the high-density point cloud model, and construct a spatial distribution map reflecting the spatial relationship of the components; Based on the spatial distribution map, adopt a multi-level safety assessment algorithm to analyze the deployment and operation areas with spatial intersections, and identify potential interference areas; Introduce the spatio-temporal evolution modeling method to dynamically analyze the continuously updated tilt photography images during the construction process. On the basis of the potential interference areas, through the interference event triggering mechanism, identify the actual safety conflict points during the construction process, and form a safety warning event set; Based on the safety warning event set, dynamically divide the risk level of the current deployment plan, and combine the deployment path planning and the operation time window to output the deployment safety management report.

2. The above deployment construction safety management method based on ground tilt photography according to claim 1, wherein The process of constructing a high-density point cloud model by combining the perspective optimization strategy and the regional occlusion compensation mechanism is as follows: Perform multi-angle registration on the ground tilt photography images, and optimize the perspective layout of the reconstructed model through the multi-view fusion algorithm; Perform boundary prediction and reconstruction compensation on the structural missing areas caused by the occlusion of adjacent components in the images; Based on the reconstruction accuracy evaluation function, dynamically adjust the fusion parameters to optimize the consistency and density distribution of the point cloud data reconstruction; After the fusion is completed, use the refined interpolation method to supplement the point cloud void areas, and finally form a high-density point cloud model with high spatial resolution and complete structural information.

3. The safety management method for above - deployment construction based on ground tilt photography according to claim 2, wherein, The process of performing spatial semantic segmentation on the high-density point cloud model is as follows: Use a multi-scale three-dimensional convolutional neural network to extract features and predict spatial labels for the point cloud data, and label different types of structural components; For the areas with unclear boundary transitions in the point cloud data, combine the context structure relationship of the structural components in the construction scene to refine the labels of the boundary fuzzy areas; Group the point cloud areas with similar semantic labels through the clustering algorithm, and optimize the boundary shape using the shape preservation constraint, and output the point cloud component set after semantic segmentation.

4. The above-mentioned construction safety management method for above-ground deployment based on ground tilt photography according to claim 3, characterized in that, The process of using the structure recognition algorithm to extract the key component information of the upper deployment area and the lower operation area in the high-density point cloud model is as follows: Through the point cloud morphology analysis algorithm, identify the structural components with characteristic dimensions and spatial direction characteristics, and divide them into components in the upper deployment area and components in the lower operation area according to the spatial distribution; Combine the construction design drawings and the component information database to compare and confirm the geometric parameters of the extracted components; Adopt the topological relationship analysis algorithm to establish the relative spatial relationship model between the components, and mark whether there are crossing, overlapping or attachment relationships between the components; Finally, generate a component information set with spatial attribute labels.

5. The above-mentioned construction safety management method for upper deployment based on ground oblique photography according to claim 4, characterized in that, The process of constructing a spatial distribution map reflecting the spatial relationship of the components is as follows: Perform three-dimensional mapping of the key components of the upper deployment and the lower operation according to their spatial coordinates and height differences, and establish a spatial layout model; Construct a spatial interference judgment matrix based on the relative position relationship and movement path rules between components, and record the spatial range affected or influenced by each component; Introduce a component dynamic weight factor, and combine the component movement frequency, task priority, and operation time period to adjust the influence level of each component in the spatial map, forming a spatial distribution map.

6. The above-mentioned construction safety management method for upper deployment based on ground oblique photography according to claim 5, characterized in that The process of analyzing the allocation and operation areas with spatial intersections using a multi-level safety assessment algorithm is as follows: Based on the component spatial attributes and interaction relationships marked in the spatial distribution map, perform a first-order spatial intersection detection on all component pairs, and identify the set of component pairs that overlap or are close to intersection during the construction period; For the identified component pairs, combine the construction task schedule and operation plan to construct a task overlap matrix in the time dimension, forming a second-order spatio-temporal intersection relationship diagram; On the basis of the second-order relationship diagram, introduce the task importance coefficient and construction stage risk coefficient to establish a third-order safety risk assessment model; According to the calculation results of the third-order model, assign corresponding risk level labels to each potentially conflicting component pair, and output the priority ranking of component conflicts.

7. A method for safety management of overhead allocation construction based on ground tilt photography according to claim 6, characterized in that The process of identifying potential interference areas is as follows: According to the risk level labels and component conflict priority ranking output in the multi-level safety assessment algorithm, extract the spatial range of the areas where the high-risk level component pairs are located as the preliminary potential interference area candidate set; Conduct an extended analysis of the areas in the candidate set, and combine the component activity trajectories and dynamic offset models to evaluate the influence radius of the allocation equipment under different operation error conditions; After external variables, dynamically correct the area boundary to form a potential interference area.

8. A method for safety management of above-ground deployment construction based on ground tilt photography according to claim 7, characterized in that, The process of introducing a spatio-temporal evolution modeling method to dynamically analyze oblique photography images is as follows: Collect the ground oblique photography image data of the construction site, register and compare it with historical images to generate an image sequence within a continuous time period; Based on the changes in component morphology and spatial position in the image sequence, construct a component state time series model to track the spatial trajectories and state transition processes of each component; Use a spatio-temporal graph convolutional network to model the evolution trend of component states, and extract key node changes and abnormal spatio-temporal behavior patterns; Combined with the set of potential interference areas, set the trigger conditions for interference events, and continuously monitor the construction site images and compare them with the trigger conditions; Once the trigger conditions are met, automatically mark the conflict positions and related components in the current image frame, generate corresponding interference events, and record them as early warning event candidates.

9. A method for safety management of above-ground deployment construction based on ground tilt photography according to claim 8, characterized in that, The process of forming a safety early warning event set is as follows: For the identified early warning event candidates, combine the component importance, allocation task urgency, and historical conflict data to calculate the weight value of the early warning events; According to the weight value of the early warning events and the event occurrence frequency, classify and screen the events, and retain the high-risk and medium-risk events as formal safety early warning events; Record the detailed attributes of the formal early warning events; Integrate multi-source data, establish a unified early warning event management database, index and classify the events, and dynamically update them to form a safety early warning event set.

10. A construction safety management method for upper deployment based on ground oblique photography according to claim 9, characterized in that The process of dynamically dividing the risk level of the current allocation plan is as follows: Receive the safety early warning event set, and analyze whether the current allocation path contains the identified high-risk event nodes; Based on the quantity, level and impact scope of early warning events, conduct a section-by-section assessment of the deployment path, and use a dynamic risk scoring function to calculate the risk values of each section; On the basis of the comprehensive risk score, combined with the time sequence arrangement of the deployment task and the resources of the operation window, judge the feasibility and substitutability of each path section; Introduce a multi-objective optimization algorithm to find a balance among risk minimization, optimal resource utilization and schedule guarantee, output multiple optional deployment plans and mark the risk levels.

Images