Construction site intelligent security control method based on video recognition and display platform

Abstract

The present invention discloses an intelligent security control method and display platform for construction sites based on video recognition, which relates to the field of construction site intelligent security technology. The method constructs a digital twin scenario of the construction site, deploys a deep learning model library and a rule engine, intelligently analyzes the real-time obtained construction site video data, identifies security risks in personnel behavior, equipment status, material storage, operation compliance and environmental elements, and generates risk event information; according to the type of risk event, automatically creates a digital control process instance in the digital twin scenario, and drives collaborative disposal and response based on the process instance, triggering the linkage control with the security subsystem; finally, aggregates the risk event information and digital control process data for multi-dimensional correlation analysis and predictive judgment. It solves the problems that the existing methods cannot achieve full-process closed-loop control and cannot achieve systematic security control covering all security elements, and improves the intelligent level of security control.

Description

Technical Field

[0001] This application relates to the field of intelligent security technology for construction sites, and in particular to an intelligent security management and control method and demonstration platform for construction sites based on video recognition. Background Technology

[0002] Currently, safety management at construction sites mainly relies on traditional manual inspections and fixed monitoring, which suffers from problems such as low efficiency, slow response, and inability to provide 24 / 7 coverage.

[0003] Existing AI video analytics-based solutions can automatically detect and alarm in specific scenarios such as helmet wearing and area intrusion. However, these solutions typically treat the video analytics module as an independent subsystem, and the alarm information is only presented in the form of sound, light, or lists. They fail to deeply integrate with the physical location of the construction site, management processes, responsible personnel, and historical data. This results in isolated early warning information and offline handling processes, creating a situation where there are early warnings but no closed loop, thus limiting the improvement of management efficiency.

[0004] While existing intelligent security management platforms integrate subsystems such as video surveillance, environmental monitoring, and tower crane monitoring, their integration methods are mostly limited to the parallel display of data, lacking entity logical association and linkage based on digital twin structures. This results in the system remaining in the stage of passive recording and post-event traceability, unable to achieve visualized closed-loop management of the entire process from pre-event prevention, in-event intervention, to post-event analysis, and especially unable to cover the systematic security management of all safety elements such as personnel behavior, equipment status, material management, work permits, and environmental compliance. Summary of the Invention

[0005] To address the technical problems of the existing technologies mentioned above, this application provides a video recognition-based intelligent security management and control method and demonstration platform for construction sites. By using AI models to deeply analyze video data, it achieves intelligent identification of all safety elements, including personnel behavior, equipment status, material management, work permits, and environmental compliance. By constructing a digital twin closed-loop management and control system driven by safety risk events, it covers the entire process before, during, and after an event, achieving a fundamental transformation from discrete alarms to process automation, and from passive monitoring to proactive prevention and intelligent linkage.

[0006] This application provides a method for intelligent security management and control of construction sites based on video recognition, including:

[0007] Step S10: Construct a digital twin scene of the construction site and digitally map and register the personnel, equipment, materials, work area entities and their associated video acquisition devices and IoT sensors within the construction site. Step S20: Build and deploy a deep learning model library and rule engine to intelligently analyze the video data acquired in real time by the video acquisition device, identify construction safety risks in personnel behavior, equipment status, material storage, operation compliance and environmental factors, and generate structured risk event information. Step S30: Based on the type of risk event information, automatically create a digital management process instance with a lifecycle in the digital twin scenario. This instance is bound to a spatial location in the digital twin scenario and has a dynamically changing visual identifier. Step S40: Based on the digital management and control process example, drive collaborative handling and response, update the visualization features corresponding to the management and control process in the digital twin scenario in real time, and trigger linkage control with the security subsystem in the construction site. Step S50: Aggregate risk event information from all elements with digital management and control process data, and conduct multi-dimensional correlation analysis and predictive judgment.

[0008] Furthermore, the building information model of the construction site and a high-precision map are obtained as the digital base map. The spatial coordinates of all video acquisition devices and physical sensors in the construction site, as well as key personnel, equipment, materials and work area entities are calibrated, registered and dynamically associated on the digital base map to build a digital twin scene that combines static and dynamic entities. Furthermore, a deep learning model library and rule engine optimized for all construction scenarios and elements are built and deployed to perform real-time analysis of video data and identify the following three major categories of risks: Personnel and behavioral safety: Personnel identification, by comparing with the real-name system to identify blacklisted and unauthorized personnel; personnel gathering detection to warn of stampede risks; standardized operation identification, to check whether the safety belts of the workers at height are worn and whether the hoisting operation supervisors are on duty; Equipment and material safety: Status monitoring of tower cranes and construction elevators, early warning of equipment collision zones; identification of whether hazardous chemicals are stored in compliance with regulations, whether building materials are stacked at excessive height and pose a risk of overturning, and whether scaffolding wall ties are missing; Environmental and process compliance and safety: Identify whether electronic permits are posted at hazardous work sites such as hot work, temporary power supply, and confined space operations, and whether safety measures are in place; environmental monitoring includes identification of dump truck covers, identification of cement tanker leak prevention, and visual aids for assessment of construction noise and dust.

[0009] Furthermore, establish an automatic mapping between construction site safety risk events and digital management processes, and achieve visualization-driven operation, including: Standard control process templates are predefined for each type of risk event. When a security risk event is generated, the process template library is matched according to the event type to automatically generate a digital control process instance, which serves as a process status entity generated in the digital twin scenario and bound to the event location. The visual form of this entity changes in real time with its life cycle status. Furthermore, based on examples of digital management and control processes, a proactive prevention, collaborative handling, and intelligent linkage mechanism for construction site safety risks is established, including: Prevention: Electronic permit management for hazardous operations is implemented in the digital twin platform. After approval, specific AI recognition rules and virtual electronic fences are automatically activated in the corresponding areas to achieve proactive defense. In-process handling: Work orders for handling security risk events are intelligently assigned according to preset strategies; handling personnel receive and provide feedback on the progress and data of each stage of the handling process through mobile terminals, and managers review them online; Cross-system linkage: When a risk event is generated or the work order status changes, predefined linkage actions are automatically triggered, including activating on-site audible and visual alarms, adjusting broadcast content, and controlling lighting and access control.

[0010] Furthermore, we continuously aggregate all-element safety risk events, digital management and control process work orders, and multi-source perception data, and conduct in-depth analysis based on spatiotemporal correlation to generate multi-dimensional statistical reports, safety risk heat maps, and contractor safety profiles; and utilize machine learning models, combined with construction plans, to achieve predictive early warning of risks and root cause tracing.

[0011] This application also provides a video recognition-based intelligent security management and control platform for construction sites, including: Digital Twin Building Module: Used to build digital twin scenarios of construction sites and to digitally map and register personnel, equipment, materials, work area entities and their associated video acquisition devices and IoT sensors within the construction site; Risk event identification module: used to build and deploy a deep learning model library and rule engine, intelligently analyze video data acquired in real time by video acquisition equipment, identify construction safety risks in personnel behavior, equipment status, material storage, operation compliance and environmental factors, and generate structured risk event information; Control process mapping module: It is used to automatically create a digital control process instance with a lifecycle in the digital twin scenario based on the type of risk event information. This instance is bound to the spatial location in the digital twin scenario and has a dynamically changing visual identifier. Linkage control module: Used to drive collaborative handling and response based on digital management and control process instances, update the visual features corresponding to the management and control process in the digital twin scenario in real time, and trigger linkage control with the security subsystem within the construction site; Big Data Analytics Module: Used to aggregate risk event information and digital management process data from all elements, and to conduct multi-dimensional correlation analysis and predictive judgment.

[0012] This application discloses the following technical effects: This application provides a video recognition-based intelligent security management and control method and display platform for construction sites. By constructing a risk identification network covering all safety elements of the construction site and proposing an automatic mapping mechanism between safety risk events and management processes, it solves the problem that existing methods can only identify single risk points. It automatically transforms different types of risk events into standard, traceable management processes and forms a visual entity in the digital twin space. This breaks down the barriers between different risk types and different management departments, forming an integrated closed-loop management and control system, and achieving full coverage and traceability of safety responsibilities.

[0013] Traditional security management and control models rely on manual detection and response, which is inefficient. This application achieves proactive defense in advance through electronic work permits, and through a built-in intelligent linkage control engine, it automatically triggers a series of responses when a risk event occurs, such as audible and visual alarms, directional broadcasts, and interlocking control equipment. This automatic linkage shortens the response time, improves the emergency response capability and reliability of sudden events, and enables effective intervention during the event.

[0014] This application transforms unstructured video and IoT sensor data into quantifiable and analyzable structured data, performs multi-dimensional correlation analysis, and uses machine learning models combined with construction plans to predict the probability of risks in different stages and areas. This allows for precise allocation of safety investment, inspection frequency, and training priorities based on data analysis results, significantly improving the efficiency of management resource utilization and reducing ineffective costs. Attached Figure Description

[0015] Figure 1 This is a flowchart illustrating the intelligent security management method for construction sites based on video recognition provided in this application embodiment.

[0016] Figure 2 This is a structural diagram of a video recognition-based intelligent security management and control platform for construction sites provided in an embodiment of this application. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description of this application will be provided in conjunction with the accompanying drawings. The described embodiments should not be considered as limitations on this application. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0018] Example 1: This application provides a method for intelligent security management and control of construction sites based on video recognition, such as... Figure 1 As shown, the method includes: Step S10: Construct a digital twin scene of the construction site and digitally map and register the personnel, equipment, materials, work area entities and their associated video acquisition devices and IoT sensors within the construction site.

[0019] In this embodiment, constructing a digital twin scenario of a construction site includes the following detailed steps: Step S11: Collect the BIM model from the design phase, the as-built BIM model, and the 3D geographic information model generated by on-site surveying of the construction site as the static scene base map of the digital twin; and access the data of various sensors deployed on the construction site in real time through the Internet of Things gateway.

[0020] Step S12: In the digital twin scenario, create a digital entity for each video acquisition device, recording the installation location, installation angle, field of view parameters, and associated network video data address; create a digital entity for each sensor, associating it with its device ID, type, spatial location, data unit, and alarm threshold; and establish dynamic entities for personnel, equipment, materials, and work areas. This step involves modeling dynamic entities within the digital twin scenario, encompassing personnel, equipment, materials, and work areas within the construction site, covering all safety elements of the site, specifically including: Personnel: It connects with the real-name management system to create a basic digital profile for each person entering the site; when they wear a location tag, the system automatically binds their location data to the personnel digital profile and presents it as a movable icon in the virtual reality scene; Equipment: 3D models are created for major construction machinery such as tower cranes, excavators, and concrete pump trucks, and bound to their real-time status data (location and operation data); for fixed equipment, location information is statically bound. Materials and work areas: Key areas such as steel bar processing area, hazardous materials warehouse, and foundation pit A area are automatically delineated through drawing analysis, and semantic labels (area type, responsible unit, risk level) are added to them.

[0021] Step S13: Establish a unified construction site spatial coordinate system, convert and calibrate all data to this unified coordinate system; based on the registration of video acquisition devices, IoT sensors and dynamic entities, construct the relationship between entities according to the device coverage, spatial location relationship and personnel arrangement; In step S14, personnel location, equipment status, and sensor data dynamically flow into the digital twin scene, driving the real-time update of the status of the corresponding entity, and video data is played in real-time in the view window of the digital twin scene.

[0022] Step S15: Simplify the mesh and compress the texture of the complex model, and publish the complete, data-driven digital twin scene through a professional 3D visualization platform to generate a scene service address that can be called.

[0023] Step S20: Build and deploy a deep learning model library and rule engine to intelligently analyze the video data acquired in real time by the video acquisition equipment, identify construction safety risks in personnel behavior, equipment status, material storage, operation compliance and environmental factors, and generate structured risk event information.

[0024] In this embodiment, dedicated deep learning models are established for construction safety risks related to personnel behavior, equipment status, material storage, operational compliance, and environmental factors, including: Personnel behavior model: Based on the target detection model, customized training is conducted to recognize behaviors such as wearing safety helmets, reflective vests, seat belts, and smoking; a video understanding model is introduced to identify temporal dangerous behaviors such as climbing, falling, and fighting; Equipment status model: An instance segmentation model is used to identify tower crane hooks and wire ropes, and key point detection is used to monitor their attitude; an anomaly detection model is used to monitor abnormal noises, abnormal vibrations and smoke during equipment operation; Material and Operation Compliance Model: Uses a semantic segmentation model to identify obstructed fire lanes, uncovered bare soil, and missing scaffolding wall ties; uses an OCR model to identify work permit information and verifies it with spatiotemporal information. The training data for all the above deep learning models comes from a large number of videos and images collected from real construction site scenes, and is finely annotated to cover different lighting, weather and viewing angle conditions to ensure the robustness of the models.

[0025] A two-layer intelligent analysis engine based on deep learning models and business rule inference is used to process video data. The deep learning model is used for basic perception, processing video data and image data extracted from video frames. It deeply analyzes each pixel in the video and images, outputting the identification results of construction site safety risks. The identification results are then input into the business rule engine, which combines regional attributes, equipment status, and work permit information from the digital twin scenario. Through a built-in AI recognition rule library, it infers business-significant safety risk events from the identification results. Event types include: Personnel and behavioral safety: Personnel identification, by comparing with the real-name system to identify blacklisted and unauthorized personnel; personnel gathering detection to warn of stampede risks; standardized operation identification, checking whether the safety belts of the workers at height are worn and whether the hoisting operation supervisors are on duty, etc. Equipment and material safety: Status monitoring of tower cranes and construction elevators, early warning of equipment collision zones; identification of whether hazardous chemicals are stored in compliance with regulations, whether building materials are stacked at excessive height and pose a risk of overturning, and whether scaffold wall ties are missing. Environmental and process compliance and safety: Identify whether electronic permits are posted and whether safety measures are in place at high-risk work sites such as hot work, temporary power supply, and confined space operations; environmental monitoring includes identification of dump truck covers, identification of cement tanker leak prevention, and visual aids for assessment of construction noise and dust.

[0026] Based on the risk event identification results, structured risk event information is generated, including at least the following fields: Key identifiers for an event: Event ID, Event Type Code, Event Level; Spatiotemporal information: latitude and longitude coordinates, altitude, time of occurrence, duration; Evidence association: the associated video data ID, the keyframe image that triggered the alarm, the storage path of the video clip, and the list of model recognition results that triggered this alarm; Contextual information: associated site area ID, associated equipment ID, associated work permit number, personnel and vehicles involved.

[0027] Step S30: Based on the type of risk event information, automatically create a digital management process instance with a lifecycle in the digital twin scenario. This instance is bound to a spatial location in the digital twin scenario and has a dynamically changing visual identifier.

[0028] In this embodiment, an automatic mapping is established between construction site safety risk events and digital management processes, and visualization-driven implementation is achieved, including: For each type of risk event, a standard control process template is predefined. Each process template defines the standard handling steps for this type of event, the responsible roles for each step, the processing time limit, and the acceptance criteria. When a security risk event is generated, a matching process template library is performed based on the event type to automatically generate a digital management process instance. This instance serves as a process state entity generated in the digital twin scenario and bound to the event location. The visual form (color and icon) of this entity is driven by a visual state machine and changes in real time with the lifecycle state of the process instance (pending dispatch, in progress, pending review, and closed). The pending assignment status is represented by a slowly rotating red semi-transparent pyramid with an exclamation mark animation at the top; the handling status changes to a blue pulsed light column, connecting the location of the risk event with the location icon of the handling personnel; the pending review status is displayed as a yellow flashing cube; and the closed status changes to a stable green marker that gradually fades.

[0029] Step S40: Based on the digital management and control process example, drive collaborative handling and response, update the visualization features corresponding to the management and control process in the digital twin scenario in real time, and trigger linkage control with the security subsystem within the construction site.

[0030] In this embodiment, based on a digital management and control process example, a proactive prevention, collaborative handling, and intelligent linkage mechanism for construction site safety risks is established, including: Prevention: Hazardous operations are managed electronically in the digital twin platform. After approval, a virtual electronic fence process status entity is automatically generated in the corresponding area, and a series of specific AI recognition rules and linkage plans are automatically bound and activated. This means that before the hazardous operation begins, a targeted intelligent monitoring and defense network has been deployed and is ready, realizing the transformation from alarm after the incident to prevention before the operation.

[0031] In-process handling: Based on preset strategies, work orders for handling security risk events are intelligently assigned. Personnel handling the events receive and provide feedback on the progress and data of each stage of the handling process via mobile terminals. Management personnel review the data online. The entire handling process is visualized and traceable within a digital twin scenario, specifically including: The real-time status of all digital management process instances for different types of safety risk events at construction sites is synchronized across multiple terminals, including digital twin scenarios, command center screens, and mobile terminals of project managers and safety officers. Operations on any terminal (such as dispatching orders from the command center or on-site feedback from safety officers) immediately update the global status, ensuring information transparency and consistent collaboration. The assignment of digital management process instances is intelligently recommended based on the location, workload, and skills of the responsible person. After assignment, a dynamic connection is generated in the digital twin scenario, pointing from the process entity to the location icon of the responsible person. The responsible person can then click on the process entity in the digital twin scenario to perform the operation. In a digital twin scenario, managers can click on any process entity to view associated live video feeds, real-time tracking of personnel involved, uploaded on-site photos, and communication records.

[0032] Cross-system linkage: When a risk event is generated or the work order status changes, predefined linkage actions are automatically triggered, including activating on-site audible and visual alarms, adjusting broadcast content, and controlling lighting and access control.

[0033] Step S50: Aggregate risk event information from all elements with digital management and control process data, and conduct multi-dimensional correlation analysis and predictive judgment.

[0034] In this embodiment, a security data lake is constructed to continuously aggregate structured risk event information and work order data for the entire lifecycle of all elements. Simultaneously, it aggregates multi-source sensing data from IoT sensors, construction plans and progress data, and static attribute data from the BIM model. All data is standardized according to a unified data model to ensure semantic consistency.

[0035] Perform multi-dimensional correlation analysis on the standardized data, and visualize the results to generate the following information: Dynamic security risk heat map: Based on the spatial density of historical security risk events and the severity level of real-time risk events, a multi-level risk heat map is dynamically generated and refreshed on the base map of the digital twin scenario using a kernel density estimation algorithm. Contractor Safety Profile: A quantitative digital profile is built for each contractor. The profile indicators include at least the total number of risk events and the incidence rate per thousand people, the proportion of various risk events, the average rectification response time, the problem recurrence rate, and the compliance rate of hazardous operation permits. The profile is presented in the form of a radar chart to provide accurate data for subcontracting management. Multi-dimensional statistical reports: Automatically generate daily, weekly, monthly, and special reports as needed. The report content includes not only statistics on the number and type of security risk events, but also conclusions of multi-dimensional correlation analysis.

[0036] In addition, static features (region type, contractor), dynamic features (real-time personnel density, equipment load rate, weather), time series features (number of similar events in the past N hours), and planning features (operation type and region in the construction plan for the next 24 hours) related to risk prediction are extracted from the data lake. Based on the time-series prediction infrastructure, multiple specialized risk prediction models are trained, including at least a strike accident prediction model and a fire prediction model. The models take various extracted risk prediction-related features as input and the probability of a certain type of risk occurring within a certain time window in the future as output. When the probability of occurrence exceeds a preset threshold, a predictive early warning work order is automatically created and sent to the safety responsible person in the corresponding area in advance. It is specially marked in the digital twin scenario to drive proactive inspection and deployment of preventive measures. For high-frequency risk events, association rule mining and causal inference methods are used to discover frequent item sets and potential causal paths that lead to the events, generate root cause analysis reports, automatically generate optimization suggestions, and transform the suggestions into new control rules, updating the AI ​​recognition rule base and control process template base.

[0037] Example 2: The intelligent security management and control platform for construction sites based on video recognition provided in this embodiment of the invention can execute the intelligent security management and control method for construction sites based on video recognition provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the method, such as... Figure 2 As shown, it has the following modules: Digital Twin Building Module: Used to build digital twin scenarios of construction sites and to digitally map and register personnel, equipment, materials, work area entities and their associated video acquisition devices and IoT sensors within the construction site; Risk event identification module: used to build and deploy a deep learning model library and rule engine, intelligently analyze video data acquired in real time by video acquisition equipment, identify construction safety risks in personnel behavior, equipment status, material storage, operation compliance and environmental factors, and generate structured risk event information; Control process mapping module: It is used to automatically create a digital control process instance with a lifecycle in the digital twin scenario based on the type of risk event information. This instance is bound to the spatial location in the digital twin scenario and has a dynamically changing visual identifier. Linkage control module: Used to drive collaborative handling and response based on digital management and control process instances, update the visual features corresponding to the management and control process in the digital twin scenario in real time, and trigger linkage control with the security subsystem within the construction site; Big Data Analytics Module: Used to aggregate risk event information and digital management process data from all elements, and to conduct multi-dimensional correlation analysis and predictive judgment.

[0038] Although this application makes various references to certain modules in the system according to the embodiments of this application, any number of different modules can be used and run on user terminals and / or servers. The various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of each functional unit are only for easy distinction between each other and are not used to limit the scope of protection of this invention.

[0039] The specific embodiments described above do not constitute a limitation on the scope of protection of this application. Those skilled in the art should understand that various modifications, combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the scope of protection of this application. In some cases, the actions or steps described in this application can be performed in a different order than that shown in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

Claims

1. A method for intelligent security management and control of construction sites based on video recognition, characterized in that, The method includes: Step S10: Construct a digital twin scene of the construction site and digitally map and register the personnel, equipment, materials, work area entities and their associated video acquisition devices and IoT sensors within the construction site. Step S20: Build and deploy a deep learning model library and rule engine to intelligently analyze the video data acquired in real time by the video acquisition device, identify construction safety risks in personnel behavior, equipment status, material storage, operation compliance and environmental factors, and generate structured risk event information. Step S30: Based on the type of risk event information, automatically create a digital management process instance with a lifecycle in the digital twin scenario, which is bound to a spatial location in the digital twin scenario; Step S40: Based on the digital management and control process example, drive collaborative handling and response, update the visual representation in the digital twin scenario in real time, and trigger linkage control with the security subsystem within the construction site; Step S50: Aggregate risk event information from all elements with digital management and control process data, and conduct multi-dimensional correlation analysis and predictive judgment.

2. The intelligent security management and control method for construction sites based on video recognition as described in claim 1, characterized in that, Step S10 involves constructing a digital twin scenario of the construction site, including the following detailed steps: Step S11: Collect the BIM model from the design phase, the as-built BIM model, and the 3D geographic information model generated by on-site surveying of the construction site as the static scene base map for the digital twin. Through the Internet of Things (IoT) gateway, data from various sensors deployed on the construction site can be accessed in real time; Step S12: In the digital twin scenario, create a digital entity for each video acquisition device, recording the installation location, installation angle, field of view parameters, and associated network video data address; create a digital entity for each sensor, associating it with its device ID, type, spatial location, data unit, and alarm threshold, and establish dynamic entities for personnel, equipment, materials, and work areas; Step S13: Establish a unified construction site spatial coordinate system, convert and calibrate all data to this unified coordinate system; based on the registration of video acquisition devices, IoT sensors and dynamic entities, construct the relationship between entities according to the device coverage, spatial location relationship and personnel arrangement; In step S14, personnel location, equipment status, and sensor data dynamically flow into the digital twin scene, driving the real-time update of the status of the corresponding entity, and video data is played in real-time in the view window of the digital twin scene. Step S15: Simplify the mesh and compress the texture of the complex model, and publish the complete, data-driven digital twin scene through a professional 3D visualization platform to generate a scene service address that can be called.

3. The intelligent security management and control method for construction sites based on video recognition as described in claim 1, characterized in that, In step S20, a two-layer intelligent analysis engine based on deep learning models and business rule reasoning is used to process the video data: Deep learning models are used for basic perception, processing video data and image data extracted from video frames, deeply analyzing each pixel in the video and images, outputting the identification results of safety risks at construction sites, and inputting the identification results into the business rule engine. Combining regional attributes, equipment status and work permit information from the digital twin scenario, and through the built-in AI recognition rule library, the engine infers business-significant safety risk events from the identification results.

4. The intelligent security management and control method for construction sites based on video recognition as described in claim 3, characterized in that, The deep learning model includes: Personnel behavior model: Based on the target detection model, customized training is conducted to recognize behaviors such as wearing safety helmets, reflective vests, seat belts, and smoking; a video understanding model is introduced to identify temporal dangerous behaviors such as climbing, falling, and fighting; Equipment status model: An instance segmentation model is used to identify tower crane hooks and wire ropes, and key point detection is used to monitor their attitude; an anomaly detection model is used to monitor abnormal noises, abnormal vibrations and smoke during equipment operation; Material and Operation Compliance Model: Uses semantic segmentation model to identify obstructed fire lanes, uncovered bare soil, and missing scaffolding wall ties; uses OCR model to identify work permit information and verifies it with spatiotemporal information; The training data for all deep learning models comes from videos and images collected from real construction sites, and is labeled to cover different lighting, weather, and viewing conditions to ensure the robustness of the models.

5. The intelligent security management and control method for construction sites based on video recognition as described in claim 3, characterized in that, The types of security risk events include: Personnel and behavioral safety: Personnel identification, by comparing with the real-name system to identify blacklisted and unauthorized personnel; personnel gathering detection to warn of stampede risks; standardized operation identification, to check whether the safety belts of the workers at height are worn and whether the hoisting operation supervisors are on duty; Equipment and material safety: Status monitoring of tower cranes and construction elevators, early warning of equipment collision zones; identification of whether hazardous chemicals are stored in compliance with regulations, whether there is a risk of overturning when building materials are stacked, and whether scaffolding wall ties are missing; Environmental and process compliance and safety: Identify whether electronic permits are posted at hot work sites, temporary power supply sites, and confined space work sites, and whether safety measures are in place; environmental monitoring includes identification of dump truck covers, identification of cement tanker leak prevention, and visual aids for assessment of construction noise and dust.

6. The intelligent security management and control method for construction sites based on video recognition as described in claim 1, characterized in that, In step S30, an automatic mapping is established between construction site safety risk events and digital management processes, and visualization-driven implementation is achieved, including: For each type of risk event, a standard control process template is predefined. Each process template defines the standard handling steps for this type of event, the responsible roles for each step, the processing time limit, and the acceptance criteria. When a security risk event is generated, a matching process template library is performed based on the event type to automatically generate a digital management process instance, which serves as a process status entity bound to the event location in the digital twin scenario. The visual form of this entity changes in real time with the lifecycle status of the process instance, which includes pending dispatch, in progress, pending review, and closed.

7. The intelligent security management and control method for construction sites based on video recognition as described in claim 1, characterized in that, In step S40, based on a digital management and control process example, a proactive prevention, collaborative handling, and intelligent linkage mechanism for construction site safety risks is established, including: Prevention: Electronic permit management for hazardous operations is implemented in the digital twin platform. Once the operation is approved, specific AI recognition rules and virtual electronic fences are automatically activated in the corresponding area to achieve proactive defense. In-process handling: Based on preset strategies, work orders for handling security risk events are intelligently assigned. Personnel handling the events receive and provide feedback on the progress and data of each stage of the handling process via mobile terminals, and managers review the data online. Cross-system linkage: When a risk event is generated or the work order status changes, predefined linkage actions are automatically triggered, including activating on-site audible and visual alarms, adjusting broadcast content, and controlling lighting and access control.

8. The intelligent security management and control method for construction sites based on video recognition as described in claim 1, characterized in that, In step S50, the results of the multi-dimensional correlation analysis are visualized to generate the following information: Dynamic security risk heat map: Based on the spatial density of historical security risk events and the severity level of real-time risk events, a multi-level risk heat map is dynamically generated and refreshed on the base map of the digital twin scenario using a kernel density estimation algorithm. Contractor Safety Profile: A quantitative digital profile is built for each contractor. The profile indicators include at least the total number of risk events and the incidence rate per thousand people, the proportion of various risk events, the average rectification response time, the problem recurrence rate, and the compliance rate of hazardous operation permits. The profile is presented in the form of a radar chart to provide accurate data for subcontracting management. Multi-dimensional statistical reports: Automatically generate daily, weekly, monthly, and special reports as needed. The report content includes not only statistics on the number and type of security risk events, but also conclusions of multi-dimensional correlation analysis.

9. The intelligent security management and control method for construction sites based on video recognition as described in claim 1, characterized in that, In step S50, the predictive assessment trains multiple specialized risk prediction models based on the time-series prediction infrastructure, including at least a strike accident prediction model and a fire prediction model. The model analyzes the multidimensional features of the construction site scenario and outputs the probability of a certain type of safety risk event occurring within a certain time window in the future. When the probability of occurrence exceeds the preset threshold, a predictive early warning work order is automatically created and sent to the safety responsible person in the corresponding area in advance. It is specially marked in the digital twin scenario to drive proactive inspection and deployment of preventive measures.

10. A construction site intelligent security management and display platform based on video recognition, characterized in that: The demonstration platform is used to implement the intelligent security management and control method for construction sites based on video recognition as described in any one of claims 1-9, and the demonstration platform includes: Digital Twin Building Module: Used to build digital twin scenarios of construction sites and to digitally map and register personnel, equipment, materials, work area entities and their associated video acquisition devices and IoT sensors within the construction site; Risk event identification module: used to build and deploy a deep learning model library and rule engine, intelligently analyze video data acquired in real time by video acquisition equipment, identify construction safety risks in personnel behavior, equipment status, material storage, operation compliance and environmental factors, and generate structured risk event information; Control process mapping module: It is used to automatically create a digital control process instance with a lifecycle in the digital twin scenario based on the type of risk event information. This instance is bound to the spatial location in the digital twin scenario and has a dynamically changing visual identifier. Linkage control module: Used to drive collaborative handling and response based on digital management and control process instances, update the visualization features corresponding to the management and control process in the digital twin scenario in real time, and trigger linkage control with the security subsystem within the construction site; Big Data Analytics Module: Used to aggregate risk event information and digital management process data from all elements, and to conduct multi-dimensional correlation analysis and predictive judgment.

Images