A collaborative perception and intelligent management and control method and system for external wall insulation hanging basket construction

Abstract

The application discloses a kind of collaborative perception and intelligent management and control methods and systems for external wall insulation hanging basket construction, and relates to building construction management and control technical field, technical scheme is, including obtaining construction site video image information and construction progress information, image data and the corresponding relationship of construction progress are established after pre-processing;According to the current construction stage, a construction identification task set is generated, and edge task scheduling and construction state recognition are carried out to obtain personnel safety recognition results and construction quality recognition results;Determine construction abnormal events based on the recognition results and handle them in stages, extract abnormal evidence information and build upload tasks, and perform upload scheduling in combination with network status;The upload processing result is transmitted to the cloud to form early warning, retention, review, rectification and traceability results, and is fed back to the edge task scheduling process.The beneficial effects of the present application are: it can improve the timeliness and reliability of external wall insulation hanging basket construction safety and quality anomaly identification, evidence uploading and closed-loop control.

Description

Technical Field

[0001] This invention relates to the field of building construction management and control technology, and in particular to a collaborative perception and intelligent management and control method and system for the construction of external wall insulation suspended scaffolding. Background Technology

[0002] External wall insulation construction is typically carried out continuously on the building facade using suspended platforms or work scaffolds. The process involves not only worker safety and personnel control within the suspended platform, but also quality control aspects such as the quality of insulation board plastering, anchor installation status, and the sequence of procedures. Due to the characteristics of suspended platform construction for external wall insulation—high working positions, dispersed work surfaces, frequent process transitions, and numerous on-site obstructions and interferences—safety and quality anomalies often occur locally, are short-lived, and are difficult to investigate and trace.

[0003] Current construction site management still relies heavily on manual inspections, spot checks, and post-event reviews. This is easily affected by personnel experience, inspection frequency, observation angle, and site environment, making it difficult to achieve continuous, stable, and objective monitoring results for multiple buildings, suspended platforms, and work surfaces. Although some technologies have begun to use video image recognition to assist construction management, most solutions only identify single objects or single anomaly types. There is a lack of unified data association and collaborative processing mechanisms between safety identification, quality identification, anomaly classification, evidence retention, and rectification feedback.

[0004] Furthermore, construction sites for external wall insulation suspended scaffolding typically face challenges such as concurrent access of multiple video feeds, limited edge computing power, and instability of shared wireless networks. Existing solutions lack mechanisms for task generation, scheduling, and upload priority control based on construction progress, target area type, and edge resource status, which can easily lead to problems such as repeated recognition of irrelevant images, untimely response to critical anomalies, and unstable uploading of abnormal evidence. Therefore, it is necessary to propose a collaborative perception and intelligent management method and system for external wall insulation suspended scaffolding construction. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a collaborative sensing and intelligent control method and system for the construction of external wall insulation suspended platforms.

[0006] The technical solution includes the following steps: S1. Acquire video image information and construction progress information of the construction site of the external wall insulation suspended platform, preprocess the video image information to obtain image data to be identified, and establish the correspondence between the image data to be identified and the construction progress information; S2. Based on the construction progress information and the image data to be identified, determine the identification task type corresponding to the current construction stage and generate a construction identification task set; S3. Perform edge task scheduling on the construction identification task set, and perform construction status identification according to the scheduling result to obtain collaborative perception output data including personnel safety identification results and construction quality identification results; S4. Based on the collaborative sensing output data, determine the construction anomaly events, classify the construction anomaly events for processing, and obtain the anomaly handling results. S5. Extract the corresponding abnormal evidence information based on the abnormal handling result, construct an upload task based on the abnormal handling result and the abnormal evidence information, schedule the upload task for upload, and obtain the upload processing result. S6. Transmit the uploaded processing results to the cloud and perform cloud management to form a cloud management result that includes early warning, retention, review, rectification and traceability, and feed back the review and rectification results to the edge task scheduling process.

[0007] Preferably, in step S2, generating the construction identification task set includes: The current construction stage is determined based on the construction progress information, and the recognition task type is determined based on the target area type corresponding to the current construction stage and the image data to be recognized. When the current construction stage is the basket preparation stage and the target area type is a suspended basket or work platform area, a construction identification task set is generated, which includes personnel safety status identification task and suspended basket personnel verification task. When the current construction stage is the plastering construction stage and the target area type is the plastering operation area, a construction identification task set containing plastering quality identification tasks is generated. When the current construction stage is the anchor installation stage and the target area type is the anchor installation area, a construction identification task set is generated, which includes anchor shape identification task and anchor assembly sequence identification task.

[0008] Preferably, in step S3, edge task scheduling for the construction identification task set includes: Extract the attribute information of each construction identification task in the construction identification task set. The attribute information includes task type, generation time, corresponding construction stage and task status, and obtain the resource status information of edge devices. Based on the attribute information and the edge device resource status information, the scheduling priority of each construction identification task is determined, and each construction identification task is assigned to a task queue of different levels according to the scheduling priority. Based on the task queue level and edge device resource status information, determine the computing power resource quota and model call order corresponding to each task queue, and perform construction status identification according to the computing power resource quota and model call order.

[0009] Preferably, in step S3, performing construction status identification includes: When the construction identification task set includes a plastering quality identification task, the image data to be identified corresponding to the plastering quality identification task is obtained, the image data to be identified is segmented into regions to obtain insulation board area information and plastering coverage area information, and the information of unplastered areas is determined based on the insulation board area information and plastering coverage area information. The overall plastering coverage status is determined based on the insulation board area information and the plastering coverage area information. When the overall grout coverage does not meet the preset requirements, a grout quality identification result representing an abnormal grout quality is generated. When the overall grouting coverage meets the preset requirements, the local grouting defect status is determined based on the distribution of the ungrouted area information in the key area of ​​the insulation board, and a grouting quality identification result is generated based on the overall grouting coverage and the local grouting defect status.

[0010] Preferably, in step S3, performing construction status identification includes: When the construction identification task set includes anchor morphology identification task and anchor assembly sequence identification task, the image data to be identified corresponding to the anchor construction area is obtained, the sleeve target and nail core target in the image data to be identified are identified, and the instance-level association relationship within the same anchor hole position is established according to the spatial matching relationship between the sleeve target and the nail core target. Based on the instance-level association, sleeve events and nail core events corresponding to the anchoring hole positions are generated, and the sleeve events and nail core events in consecutive image frames are stabilized and confirmed to obtain a stable event sequence. Based on the stable event sequence and the instance-level association, state transition judgment is performed to determine the anchor status of the corresponding anchor hole position. The anchor status includes the installation completed state, the component missing state, and the installation abnormal state, and anchor morphology recognition result and anchor assembly timing recognition result are generated.

[0011] Preferably, in step S4, the graded processing of the construction anomaly includes: The abnormality type, abnormal location, and trigger time of the construction abnormal event are determined based on the collaborative sensing output data, and the abnormality level of the construction abnormal event is determined based on the abnormality type. When the anomaly type is a personnel safety anomaly, the construction anomaly event is identified as a work stoppage-level anomaly event, and a work stoppage-level anomaly handling result is generated. When the anomaly type is a construction quality anomaly, the construction anomaly event is identified as a rectification-level anomaly event, and a rectification-level anomaly handling result is generated. An event identifier is generated for the construction anomaly, and the event identifier, anomaly type, anomaly level, anomaly location, trigger time, and anomaly handling result are associated to form hierarchical anomaly event information.

[0012] Preferably, in step S5, the uploading and scheduling of the anomaly handling result and anomaly evidence information includes: An upload task is constructed based on the abnormal handling results and abnormal evidence information, and the upload task is configured with the event identifier, abnormality level, data type, data volume, upload time limit and task status corresponding to the construction abnormal event; Obtain the current available network bandwidth and bandwidth fluctuation status, determine the corresponding fragment size based on the current available network bandwidth, bandwidth fluctuation status and the data volume of the upload task, and divide the upload task into fragments for data upload according to the fragment size; The upload order of each upload task is determined based on the anomaly level and upload time limit, and the segmented upload data is sent based on the upload order to obtain the upload processing result.

[0013] Preferably, in step S5, sending the fragmented upload data based on the upload order includes: Based on the anomaly level and the amount of data sent for each upload task, the actual service status of upload tasks with different anomaly levels within the current scheduling window is statistically analyzed, and the actual service status is compared with the preset service target to obtain the warning level service bias status. The warning level corresponds to the anomaly level and is used to distinguish the service guarantee level of different abnormal events during the upload scheduling process; Pre-configured service targets can be based on the target upload ratio, target sending opportunities, or target completion time limit for work stoppage-level and rectification-level abnormal events. The upload order of each upload task is adjusted according to the service bias status of the warning level, so that upload tasks with higher abnormal levels and insufficient service receive higher sending priority. Before the current segment is sent, check if there is an uploading task with a higher sending priority than the current uploading task. When the detection result meets the segment boundary preemption condition, switch to the uploading task to be sent at the segment boundary after the current segment is sent and mark the switched uploading task as suspended. When the switched upload task meets the sending conditions again, it resumes sending based on the amount of data remaining to be sent.

[0014] Preferably, in step S6, cloud-based management includes: After receiving the uploaded processing results, the cloud will associate and store the abnormal handling results, abnormal evidence information, review results and rectification status according to the event identifier corresponding to the construction abnormal event, forming a closed-loop record of the abnormal event; The establishment status of the construction anomaly is updated based on the review results, the handling status of the construction anomaly is updated based on the rectification status, and the updated establishment and handling status are fed back to the edge task scheduling process. The edge task scheduling process updates the task priority, task admission status, or identification strategy in subsequent scheduling cycles based on the established status and the processing status.

[0015] A collaborative sensing and intelligent control system for the construction of external wall insulation suspended scaffolding, the system comprising edge devices and a cloud control unit; The edge device includes a video process association module, a task generation module, an edge scheduling and identification module, an anomaly classification module, and an upload scheduling module; The video process association module is used to acquire video image information and construction process information of the external wall insulation suspended platform construction site, preprocess the video image information to obtain image data to be identified, and establish the correspondence between the image data to be identified and the construction process information. The task generation module is used to determine the recognition task type corresponding to the current construction stage based on the construction progress information and the image data to be recognized, and to generate a set of construction recognition tasks. The edge scheduling and identification module is used to schedule edge tasks for the construction identification task set and perform construction status identification according to the scheduling results to obtain collaborative perception output data that includes personnel safety identification results and construction quality identification results. The anomaly classification module is used to determine construction anomalies based on the collaborative sensing output data, classify the construction anomalies, and obtain anomaly handling results. The upload scheduling module is used to extract corresponding abnormal evidence information based on the abnormal handling result, construct an upload task based on the abnormal handling result and the abnormal evidence information, schedule the upload task, obtain the upload processing result, and send the upload processing result to the cloud management unit. The cloud management unit is used to receive the uploaded processing results and perform cloud management, forming a cloud management result that includes early warning, retention, review, rectification and traceability, and feeding back the review and rectification results to the edge scheduling and identification module.

[0016] The beneficial effects of the technical solution provided by the embodiments of the present invention are as follows: 1. This invention associates video image information of the construction site of the external wall insulation suspended platform with construction progress information. It can automatically generate corresponding construction recognition tasks according to different construction stages such as preparation for the platform, plastering, and anchor installation. This reduces irrelevant images and invalid tasks from entering the subsequent recognition process and improves the targeting and effectiveness of multi-channel video data processing.

[0017] 2. This invention integrates personnel safety status identification, suspended platform personnel verification, insulation board plastering quality identification, anchor morphology identification, and anchor assembly sequence identification into the same collaborative sensing process, which can simultaneously cover safety and quality risks in the construction of external wall insulation suspended platforms, improving the timeliness and completeness of construction anomaly detection.

[0018] 3. This invention uses an edge task scheduling mechanism to determine scheduling priority, task queue, computing power resource quota and model calling order based on the attribute information of construction identification tasks and the resource status of edge devices. This can prioritize the execution of key identification tasks when edge computing power is limited or multiple tasks are running concurrently, thereby reducing invalid inference and resource consumption.

[0019] 4. This invention classifies construction anomalies into different levels, classifying personnel safety anomalies as work stoppage-level anomalies and construction quality anomalies as rectification-level anomalies. It also generates event identifiers for construction anomalies, which facilitates a unified association between anomaly handling results, anomaly evidence information, uploaded tasks, and cloud records.

[0020] 5. This invention improves the timeliness and stability of uploading key abnormal evidence by using an abnormal evidence extraction and uploading scheduling mechanism to segment, sort, preempt, and resume uploading tasks based on the abnormality level, uploading time limit, available network bandwidth, and bandwidth fluctuation status under weak network or shared network conditions.

[0021] 6. This invention generates early warning, retention, review, rectification and traceability results through cloud-based management and feeds back the review and rectification results to the edge task scheduling process, enabling construction anomalies to form a closed-loop management from identification, handling, uploading, review to rectification and traceability. Attached Figure Description

[0022] Figure 1 This is an overall flowchart of the collaborative perception and intelligent control method for the construction of external wall insulation suspended scaffolding according to an embodiment of the present invention.

[0023] Figure 2 This is a structural block diagram of a collaborative perception and intelligent control system for the construction of external wall insulation suspended scaffolding, according to an embodiment of the present invention.

[0024] Figure 3 This is a flowchart illustrating the process of generating a set of construction identification tasks in an embodiment of the present invention.

[0025] Figure 4 This is a flowchart of edge task scheduling and construction status identification in an embodiment of the present invention.

[0026] Figure 5 This is a flowchart of the abnormal evidence extraction and uploading scheduling process in an embodiment of the present invention.

[0027] Figure 6 This is a flowchart illustrating the cloud-based control and feedback closed-loop process in an embodiment of the present invention. Detailed Implementation

[0028] Example 1 See Figure 1 This invention provides a collaborative sensing and intelligent control method for the construction of suspended scaffolds for exterior wall insulation. This method can be applied to continuous monitoring and collaborative control scenarios during the construction of exterior wall insulation panels, including monitoring the safety status of scaffold workers, the quality of insulation panel plastering, the installation status of anchors, and the handling of abnormal events. The method can be jointly executed by edge devices located at the construction site and a cloud-based control platform communicating with the edge devices. The edge devices are used to access video image information and construction progress information from the construction site, and complete task generation, edge scheduling, construction status identification, anomaly classification, and upload scheduling on the construction site side. The cloud-based control platform receives the uploaded processing results and performs early warning, retention, review, rectification, and traceability management.

[0029] It should be noted that the construction site of the external wall insulation suspended platform in this embodiment may include a suspended platform or work platform area, a plastering work area, and an anchor installation area. The suspended platform or work platform area is mainly used to identify the safety status of personnel and the number of people in the suspended platform. The plastering work area is mainly used to identify the plastering coverage status of the insulation board and the status of local plastering defects. The anchor installation area is mainly used to identify the morphology status of the anchors and the assembly sequence status of the anchors. The edge device can receive video image information from multiple camera devices and determine the identification task type corresponding to different image data by combining the construction progress information, thereby realizing safety identification, quality identification, anomaly classification, evidence uploading, and cloud closed-loop processing under the same control framework.

[0030] Includes the following steps: S1. Obtain video image information and construction progress information of the construction site of the external wall insulation suspended platform, preprocess the video image information to obtain the image data to be identified, and establish the correspondence between the image data to be identified and the construction progress information. In this embodiment, video image information and construction progress information of the external wall insulation suspended platform construction site are first acquired. The video image information reflects the real-time image status of workers, the suspended platform or work platform, the insulation board working surface, the plastering area, and the anchoring construction area at the external wall insulation suspended platform construction site. The construction progress information reflects the current construction stage, work area, and process status of the external wall insulation suspended platform construction. It should be noted that the video image information can be acquired by camera devices installed at relative positions on the suspended platform, work platform, floor edge, ground observation point, or external wall working surface. Each camera device is configured with a corresponding device number, building number, facade number, suspended platform number, construction section number, and target area type when connected to the edge device, so as to identify the source and purpose of different video image information in the future.

[0031] It should be noted that, in this embodiment, the anchors refer to the anchors, bolts and their assembly components used to fix the insulation board during the construction of external wall insulation. The sleeve target and the nail core target are identification objects used to characterize the assembly status of the same anchor hole position during the construction of the anchors.

[0032] Construction progress information can be formed by combining construction task configuration data, construction plan data, process flow data, suspended platform operation area configuration data, and edge process status records. Specifically, construction task configuration data identifies the current work surface of the construction team or the current suspended platform; construction plan data identifies the process arrangement of external wall insulation construction; process flow data identifies the status of tasks such as platform preparation, plastering, anchor installation, anomaly verification, and rectification; suspended platform operation area configuration data identifies the current building facade and construction section location of the suspended platform; and edge process status records record the changes in process status formed by the edge equipment during continuous processing of video images. It should also be noted that construction progress information is not limited to a single source. When the construction management platform has already issued a construction stage or operation area, the edge equipment directly reads the corresponding construction progress information; when the construction management platform has not issued a complete construction stage, the edge equipment can combine the work objects and continuous frame status in the video images to complete the status of the current construction stage or maintain the stage marker.

[0033] After acquiring video image information, the edge device performs unified access and decoding processing on each video stream, converting the continuous video stream into a sequence of image frames arranged according to the acquisition time, and writing the acquisition time, video channel number, camera device number, and frame sequence number into each image frame. Furthermore, to reduce time discrepancies between different camera devices, the edge device performs time synchronization processing on each video channel, enabling multiple image frames within the same construction period to be sorted and accessed according to a unified time reference, thereby avoiding incorrect combination of image frames from different construction stages or different work areas during subsequent task generation.

[0034] When preprocessing video image information, the edge device performs frame extraction, deduplication, image quality detection, target region cropping, and image size unification on image frames. Specifically, frame extraction is used to extract image frames that meet the recognition frequency requirements from the continuous video stream, avoiding repeated processing of continuous images with minimal content changes; deduplication is used to filter out duplicate frames that are highly similar to previous image frames and do not affect the judgment of construction status; image quality detection is used to determine whether image frames have severe blurring, occlusion, excessive darkness, overexposure, or jitter, and sets a low-quality mark for image frames that do not meet the minimum recognition requirements; target region cropping is used to extract the suspended platform or work platform area, plastering area, and anchor construction area from the image frames according to the target area type of the camera device or the preset work surface range; and image size unification is used to convert image frames from different sources into a unified data specification suitable for edge recognition tasks.

[0035] It should be noted that preprocessing does not change the construction status expressed by the construction site images, but rather structures and organizes the video image information so that it can be stably used in subsequent task generation and construction status recognition processes. For image frames that still do not meet the recognition requirements after preprocessing, the edge device can retain their source and time information and mark them as unacceptable image data or low-quality candidate image data; for image frames that meet the recognition requirements, the edge device encapsulates them into image data to be recognized, which includes the image frame itself, video source information, acquisition time information, target area type, image quality status, and corresponding work area identifier.

[0036] After obtaining the image data to be identified, the edge device establishes a correspondence between the image data and construction progress information. Specifically, the edge device uses the acquisition time, construction section number, suspended platform number, building facade number, video channel number, and target area type as the association criteria to match the image data to be identified within the same construction area, time range, and process status with the construction progress information, and writes the corresponding construction stage identifier and process status identifier for the successfully matched image data. Furthermore, when multiple video images exist in the same construction area, the edge device divides the image data to be identified into personnel operation area image data, plastering operation area image data, and anchor construction area image data according to the target area type of the camera equipment, so as to generate different construction identification tasks later based on the construction stage and target area type.

[0037] Furthermore, to avoid mismatches between image data and construction progress information during construction phase transitions, edge devices can be configured with phase hold time windows and phase transition confirmation conditions. When construction progress information changes in adjacent time periods, the edge device first marks the image data to be identified near the transition boundary as phase confirmation data, and then confirms it by combining changes in work objects, task configurations, or process status records in subsequent consecutive image frames. After confirmation, the corresponding construction phase identifier is written into the image data to be identified. This improves the accuracy of data association when multiple video streams, multiple construction areas, and multiple process statuses coexist.

[0038] Through the above processing, image data to be identified can be generated corresponding to the construction progress information. This image data not only contains image content that can be processed by the model or recognition rules, but also carries the construction stage, target area type, source device, acquisition time, and image quality status for subsequent task generation. This data structure provides the foundation for generating a set of construction identification tasks based on the current construction stage, enabling edge devices to distinguish which image data is used for personnel safety identification, which image data is used for plastering quality identification, and which image data is used for anchor morphology identification or anchor assembly sequence identification.

[0039] S2. Based on the construction progress information and the image data to be identified, determine the identification task type corresponding to the current construction stage and generate a set of construction identification tasks; After generating the image data to be identified corresponding to the construction progress information, such as Figure 3 As shown, the edge device further generates a construction recognition task set based on construction progress information and the image data to be recognized. The construction recognition task set is used to uniformly organize the recognition tasks that need to be performed at the current construction site. It does not simply send all the image data to be recognized into the model for processing, but first determines whether the current image data to be recognized matches the current construction stage and target area type, and then generates recognition tasks that are appropriate for the construction stage, thereby avoiding invalid recognition on irrelevant images.

[0040] It should be noted that the construction progress information may include the current construction stage identifier, work surface location, suspended platform number, construction section number, process start time, process switching status, and construction task configuration status. After reading the construction progress information, the edge device first determines the construction stage of the current suspended platform or work surface. The construction stage includes at least the platform preparation stage, the grouting stage, and the anchor installation stage. If the current construction stage has been issued by the construction management platform, the edge device directly uses the construction stage; if the current construction stage has not been explicitly issued, the edge device combines the target area type corresponding to the image data to be identified, the changes in the work objects in the continuous images, and the process status recorded in the previous time period to determine the current construction stage or maintain the previous valid construction stage.

[0041] Furthermore, the edge device reads the target area type from the image data to be identified. The target area type is used to characterize the main on-site area attributes corresponding to the image data to be identified, including the suspended platform or work platform area, the plastering operation area, and the anchor construction area. The suspended platform or work platform area is mainly used to observe whether the workers have entered the suspended platform, whether they are wearing safety protective equipment, and the number of people in the work platform; the plastering operation area is mainly used to observe the insulation board surface, the plastered and unplastered areas; the anchor construction area is mainly used to observe the anchor holes, sleeves, nail cores, and the local condition of the anchors after installation.

[0042] After determining the current construction stage and target area type, the edge device matches the two and determines the identification task type accordingly. When the current construction stage is the basket preparation stage and the target area type is a suspended basket or work platform area, the edge device generates a set of construction identification tasks that includes personnel safety status identification tasks and suspended basket personnel verification tasks. The personnel safety status identification task identifies the wearing status of workers' safety helmets, reflective safety clothing, and the status of personnel entering the work area. The suspended basket personnel verification task determines whether the number of personnel in the suspended basket or work platform meets the preset safety management requirements, thus ensuring that the safety status before the start of high-altitude work is prioritized for inclusion in the identification task set.

[0043] If the current construction stage is the plastering stage, and the target area type is the plastering operation area, the edge device generates a construction identification task set that includes the plastering quality identification task. The plastering quality identification task is used to identify and subsequently determine the insulation board area, plastered area, and unplastered area to ascertain the overall plastering coverage status and whether there are local plastering defects in key areas of the insulation board. It should be noted that during this stage, if video images of the suspended platform or work platform area continue to be received, the edge device can retain the basic monitoring status of the personnel safety status identification task, but the plastering quality identification task corresponding to the plastering operation area is written into the construction identification task set as a quality identification task directly related to the current process.

[0044] When the current construction phase is the anchor installation phase, and the target area type is the anchor installation area, the edge device configures the image data to be identified as an object related to the anchor installation quality, and generates an anchor morphology recognition task and an anchor assembly sequence recognition task. The anchor morphology recognition task is used to identify the local morphological state of the anchor after installation, and the anchor assembly sequence recognition task is used to identify the order of appearance, correlation, and installation completion status of the sleeve target and the nail core target in the same anchoring hole, thereby determining whether the anchor installation process meets the preset procedure requirements.

[0045] It should also be noted that when the target area type of the image data to be identified does not match the current construction stage, the edge device does not directly generate the corresponding high-precision identification task. Instead, it sets a mismatch mark, a low-priority monitoring mark, or a pending confirmation mark for the image data to be identified. For example, when an image of the anchor construction area is received during the plastering construction stage but the anchor work object has not yet appeared, the edge device may temporarily not generate an anchor assembly sequence identification task. When an image of the plastering work area is received during the anchor construction stage but the current image does not contain a valid insulation board work surface, the edge device can mark the image data as an object to be reviewed or delayed, thereby reducing invalid tasks from entering the subsequent edge task scheduling process.

[0046] Furthermore, when generating each construction identification task, the edge device writes task attribute information for that task. This task attribute information includes task type, task generation time, corresponding construction stage, target area type, video source information, image data index to be identified, and task status. The task type distinguishes between personnel safety status identification tasks, suspended platform personnel verification tasks, plastering quality identification tasks, anchor morphology identification tasks, and anchor assembly sequence identification tasks. The task generation time records the time when the task enters the construction identification task set. The corresponding construction stage represents the association between the task and the construction process. The target area type represents the image area corresponding to the task. The video source information traces the camera equipment and construction area corresponding to the task. The image data index to be identified is used to retrieve the corresponding image data. The task status indicates whether the task is in a pending, scheduled, completed, delayed, or unexecutable state.

[0047] Within the same scheduling cycle, edge devices can generate multiple construction identification tasks for multiple video channels and multiple target areas. These tasks are then grouped according to construction stage, target area type, and task type to form a construction identification task set. Once the construction identification task set is formed, the edge devices use it as input for subsequent edge task scheduling, enabling subsequent processing to sort, allocate, and execute safety identification tasks, quality identification tasks, and assembly sequence identification tasks within a unified task set.

[0048] S3. Perform edge task scheduling on the construction identification task set, and perform construction status identification according to the scheduling results to obtain collaborative perception output data including personnel safety identification results and construction quality identification results; After the construction identification task set is formed, such as Figure 4As shown, the edge device performs edge task scheduling on the set of construction recognition tasks and executes construction status recognition according to the scheduling results. It should be noted that the purpose of edge task scheduling is to enable the same edge device to prioritize construction recognition tasks that are highly relevant to the current construction stage, have a high risk level, and have high timeliness requirements when simultaneously accessing multiple video image information and generating multiple types of recognition tasks, while avoiding low-value image frames or repetitive tasks from continuously occupying edge computing resources.

[0049] When scheduling edge tasks, the edge device first reads each construction identification task from the set of construction identification tasks one by one, and extracts the attribute information of each construction identification task. The attribute information includes task type, generation time, corresponding construction stage, and task status. Furthermore, the task type is used to distinguish between personnel safety status identification tasks, suspended platform personnel verification tasks, plastering quality identification tasks, anchor morphology identification tasks, and anchor assembly sequence identification tasks; the generation time is used to determine the order in which construction identification tasks enter the scheduling process; the corresponding construction stage is used to determine the degree of correlation between the construction identification task and the current process; and the task status is used to indicate whether the construction identification task is in a pending scheduling, executing, completed, delayed processing, or unexecutable state.

[0050] The edge device also acquires edge device resource status information, which characterizes the computing and transmission capabilities that the edge device can provide to the construction status identification process within the current scheduling cycle. It should be noted that the edge device resource status information may include processor occupancy status, image processing unit occupancy status, memory occupancy status, device temperature status, inference task queue length, and network availability status. The edge device obtains this information by reading its own operating status and task cache status, and uses it as the basis for determining task priority, task queue allocation, computing resource quota determination, and model call order.

[0051] Computing resource quotas can be represented by the number of inference time slots, the number of concurrent tasks, the scope of cache resource usage, or the model call opportunities that each task queue can occupy within the current scheduling cycle.

[0052] After obtaining the attribute information of the construction identification tasks and the resource status information of the edge devices, the edge devices determine the scheduling priority of each construction identification task. Specifically, the edge devices can comprehensively sort the construction identification tasks according to the safety risk attributes corresponding to the task type, the degree of matching between the task and the current construction stage, the task waiting time, the availability of image data, and the current load status of the edge devices. In the basket preparation stage, the personnel safety status identification task and the basket personnel verification task have higher scheduling priority; in the plastering construction stage, the plastering quality identification task has higher scheduling priority; in the anchor construction stage, the anchor morphology identification task and the anchor assembly sequence identification task have higher scheduling priority. It should also be noted that for construction areas where suspected anomalies have appeared, the edge devices can increase the scheduling priority of subsequent identification tasks in the same area or for the same target object to ensure that the continuity of anomalies can be confirmed in a timely manner.

[0053] Edge devices allocate construction identification tasks to different levels of task queues according to scheduling priorities. Specifically, edge devices can set up real-time guarantee queues, regular identification queues, and delayed processing queues. The real-time guarantee queue is used to store tasks such as personnel safety status identification, suspended platform personnel verification, and suspected anomaly review. The regular identification queue is used to store tasks such as plastering quality identification, anchor morphology identification, and anchor assembly sequence identification that match the current construction stage. The delayed processing queue is used to store construction identification tasks that do not completely match the target area, have low image quality, or are weakly related to the current construction stage. Through the above task queue division, edge devices can prioritize the execution of critical identification tasks when computing resources are limited.

[0054] Furthermore, the edge devices determine the computing power resource quota corresponding to each task queue based on the task queue level and the edge device's resource status information. Specifically, when the edge device load is low and the network status is stable, the edge device can simultaneously execute construction identification tasks in both the real-time assurance queue and the regular identification queue. When the edge device load increases, it prioritizes construction identification tasks in the real-time assurance queue and reduces, merges, or delays the execution of tasks in the regular identification queue. When the edge device is under high load or the network is constrained, it suspends and delays processing tasks in the queue, retaining only identification tasks directly related to personnel safety anomalies, construction quality anomalies, and anomaly verification. This ensures that the edge device maintains its ability to identify critical construction risks under different operating conditions.

[0055] The edge device also determines the model invocation order based on the task queue level and the task type of the construction identification task. It should be noted that the model invocation order can adopt a layered processing approach of lightweight screening, dedicated identification, and high-precision verification. Lightweight screening is used to quickly determine whether there are workers, effective working surfaces, insulation board areas, or localized anchorage areas in the image frame. Dedicated identification is used to perform personnel safety status identification, suspended platform personnel verification, plastering quality identification, anchorage morphology identification, and anchorage assembly sequence identification on the screened image data. High-precision verification is used to confirm suspected safety anomalies, suspected plastering quality anomalies, and suspected anchorage construction anomalies. Through layered model invocation, the edge device can reduce the number of invalid high-precision identifications and improve the efficiency of edge computing power utilization.

[0056] When performing personnel safety status identification, the edge device retrieves the image data to be identified corresponding to the area of ​​the suspended platform or work platform, and identifies the personnel targets, safety helmet targets, reflective safety clothing targets, and the boundary area of ​​the work platform in the image. Further, the edge device performs spatial correspondence analysis between the personnel targets and the safety helmet and reflective safety clothing targets to determine whether the personnel are wearing safety helmets and reflective safety clothing, and counts the number of personnel within the boundary area of ​​the work platform to determine whether the number of personnel in the suspended platform or work platform exceeds the preset number requirement. After identification, the edge device generates a personnel safety identification result that includes the helmet wearing status, reflective safety clothing wearing status, and suspended platform number verification status.

[0057] When performing the plastering quality identification task, the edge device acquires the image data to be identified corresponding to the plastering quality identification task and performs region segmentation processing on the image data to be identified. Specifically, the edge device first identifies the area where the insulation board is located to obtain the insulation board area information, then identifies the plastering coverage area on the surface of the insulation board to obtain the plastering coverage area information, and determines the part of the insulation board area information that is not covered by plastering as the unplastered area information. It should be noted that the insulation board area information is used to limit the effective board surface range for plastering quality judgment, the plastering coverage area information is used to characterize the board surface part that has been plastered, and the unplastered area information is used to characterize the board surface part that may have leakage, insufficient plaster, or local construction deficiencies.

[0058] After obtaining information about the insulation board area and the grout coverage area, the edge-end equipment first determines the overall grout coverage status. Specifically, the edge-end equipment judges whether the overall grout coverage meets the construction requirements based on the extent of the grout coverage area within the insulation board area. If the overall grout coverage does not meet the preset requirements, the edge-end equipment directly generates a grout quality identification result indicating abnormal grout quality. If the overall grout coverage meets the preset requirements, the edge-end equipment continues to analyze the distribution of ungrouted areas within the key areas of the insulation board. The key areas of the insulation board can include the edge area, corner area, preset stress-concern area, and areas related to subsequent anchoring construction. The edge-end equipment determines whether there are local grout defects by judging the area, continuity, and concentration of ungrouted areas within these key areas, and generates a grout quality identification result based on the overall grout coverage status and local grout defect status.

[0059] When performing anchor morphology recognition and anchor assembly sequence recognition tasks, the edge device acquires image data to be recognized corresponding to the anchor construction area and identifies the sleeve targets and nail core targets in the image data. Further, using the area where the sleeve target is located as a reference, the edge device determines the matching area of ​​the corresponding anchor hole position and judges the relative position, overlap relationship, and center proximity between the nail core target and the sleeve target, thereby establishing an instance-level association relationship between the sleeve target and the nail core target within the same anchor hole position. It should be noted that the instance-level association relationship is used to distinguish between sleeves and nail cores in different anchor hole positions, avoiding the incorrect binding of nail core targets in adjacent holes to other sleeve targets.

[0060] After establishing instance-level associations, the edge device generates corresponding sleeve and nail events for anchoring holes based on the presence of sleeve and nail targets in consecutive image frames. Furthermore, the edge device caches these sleeve and nail events in consecutive image frames and determines whether the events are stable through multi-frame continuous confirmation. For single-frame anomalies caused by short-term occlusion, image jitter, or fluctuations in target detection, the edge device does not immediately change the construction status of the anchoring holes. Instead, it performs stabilization confirmation by combining the continuity of events in adjacent image frames to obtain a stable event sequence.

[0061] The edge device determines state transitions based on stable event sequences and instance-level associations. Specifically, if a sleeve event occurs first in a stable event sequence, followed by a nail core event within the same anchoring hole location, the edge device determines that the anchoring hole location has entered the normal assembly process. If a nail core event occurs without a stable sleeve event, the edge device determines that there is an abnormal assembly sequence. If the sleeve event is consistently established but the nail core event is not consistently established for a long period, the edge device determines that there is a missing component or incomplete installation. If both the sleeve event and the nail core event are consistently established and they maintain an instance-level association within the same anchoring hole location, the edge device determines that the anchoring hole location has reached the installation completion state. Thus, the edge device determines the anchor status of the corresponding anchoring hole location, including installation completion, missing component, and abnormal installation states, and generates anchor morphology recognition results and anchor assembly timing recognition results.

[0062] It should also be noted that the construction status identification process can simultaneously output personnel safety identification results and construction quality identification results, or it can output one type of identification result separately at different construction stages. Personnel safety identification results reflect the protective status of workers and the number of people in the suspended platform, while construction quality identification results reflect the grouting quality status, anchor morphology status, and anchor assembly sequence status. The edge device structures and organizes these identification results according to construction area, task type, collection time, and event source, forming collaborative sensing output data. This collaborative sensing output data provides a data foundation for subsequent identification, tiered processing, evidence extraction, and uploading / scheduling of construction anomalies.

[0063] S4. Based on the collaborative sensing output data, identify construction anomalies, classify and process them to obtain anomaly handling results. After generating collaborative sensing output data, the edge devices perform anomaly analysis on the data to determine if any construction anomalies exist. It should be noted that the collaborative sensing output data includes personnel safety identification results and construction quality identification results. Personnel safety identification results characterize the wearing status of safety helmets and reflective safety clothing, and the verification status of the number of workers in the suspended platform. Construction quality identification results characterize the quality of insulation board plastering, the morphology of anchors, and the assembly sequence of anchors. When performing anomaly analysis, the edge devices do not simply judge individual identification values ​​in isolation. Instead, they combine the corresponding construction area, task type, data collection time, target area type, and continuous identification status to form construction anomalies that can be subsequently processed and uploaded for scheduling and invocation.

[0064] Specifically, the edge device reads various identification results from the collaborative sensing output data and compares these results with preset anomaly judgment conditions. For personnel safety identification results, when it detects that workers are not wearing safety helmets or reflective safety clothing, or that the number of personnel in the suspended platform or work platform exceeds the preset number requirement, the edge device will determine the corresponding identification result as a personnel safety anomaly. For construction quality identification results, when it detects that the overall plastering coverage does not meet preset requirements, there are local plastering defects in key areas of the insulation board, there are missing parts in the anchors, or there are abnormal installation conditions or assembly sequence conditions, the edge device will determine the corresponding identification result as a construction quality anomaly. It should also be noted that for situations where insufficient image quality, severe target occlusion, or insufficient recognition continuity prevent direct confirmation of an anomaly, the edge device can generate a suspected anomaly marker or a review pending marker, which will serve as the triggering basis for subsequent anomaly review tasks or evidence extraction tasks.

[0065] After identifying an anomaly, the edge device extracts the anomaly type, location, and trigger time from the collaborative sensing output data. The anomaly type distinguishes between personnel safety anomalies and construction quality anomalies, and can further differentiate between anomalies related to safety helmets, reflective safety clothing, the number of personnel in the suspended platform, plastering quality, anchor morphology, and anchor assembly sequence. The anomaly location records the corresponding building number, facade number, construction section number, suspended platform number, video channel number, and target area type. The trigger time records the time when the anomaly first meets the judgment criteria or the confirmation time after consecutive anomalies are identified. Using this information, the edge device can convert abstract identification results into construction anomaly events with clear sources, locations, and times.

[0066] Furthermore, the edge device determines the anomaly level of the construction anomaly event based on the anomaly type. When the anomaly type is personnel safety anomaly, because it directly relates to the safety of high-altitude operations on the external wall insulation suspended platform, the edge device classifies the construction anomaly event as a stoppage-level anomaly event and generates a stoppage-level anomaly handling result. The stoppage-level anomaly handling result can include a stoppage notification status, on-site alarm status, management terminal push status, anomaly review pending status, and subsequent rectification tracking status, indicating that the anomaly event needs to be prioritized in the early warning and uploading process. When the anomaly type is construction quality anomaly, because it mainly relates to the quality of plastering, anchor installation quality, and process connection quality, the edge device classifies the construction anomaly event as a rectification-level anomaly event and generates a rectification-level anomaly handling result. The rectification-level anomaly handling result can include a rectification notification status, quality anomaly record status, evidence retention status, review pending status, and rectification tracking status, indicating that the anomaly event needs to enter the quality rectification and traceability process.

[0067] It should be noted that work stoppage-level and rectification-level anomalies are not merely notification messages, but serve as the data foundation for subsequent evidence extraction, uploading scheduling, and cloud-based management. When generating anomaly handling results, edge devices simultaneously generate event identifiers for construction anomalies. These event identifiers are used to establish a unified association between the identification results, handling results, evidence information, uploaded tasks, cloud-based closed-loop records, review results, and rectification status of the same anomaly. Event identifiers can be formed by combining construction area information, video channel information, trigger time information, and anomaly type information, or they can be uniquely numbered by the edge device according to the event generation sequence and bound to the construction area information, trigger time information, and anomaly type information.

[0068] Furthermore, the edge device associates the event identifier, anomaly type, anomaly level, anomaly location, trigger time, and anomaly handling result to form hierarchical anomaly event information. Hierarchical anomaly event information can at least characterize what kind of anomaly it is, where it occurs, when it is triggered, its level, and what handling is required. Further, hierarchical anomaly event information can also carry the anomaly source task, anomaly-related image data index, continuous confirmation status, and handling status. The anomaly source task is used to trace the anomaly event from personnel safety status identification tasks, suspended platform personnel verification tasks, plastering quality identification tasks, anchor morphology identification tasks, or anchor assembly sequence identification tasks. The anomaly-related image data index is used to locate the anomaly evidence information that needs to be extracted subsequently. The continuous confirmation status is used to indicate whether the anomaly event is a single trigger, a continuous trigger, or a verification confirmation trigger. The handling status is used to indicate whether the anomaly event is in a pending upload, uploaded, pending review, reviewed, pending rectification, or completed state.

[0069] It should also be noted that when multiple abnormal events exist within the same time period, edge devices can sort the tiered abnormal event information according to abnormality level, abnormality type, abnormality location, and trigger time. Stoppage-level abnormal events take precedence over rectification-level abnormal events in the subsequent evidence extraction and upload scheduling process; under the same abnormality level, consecutively triggered abnormal events take precedence over single-triggered abnormal events in subsequent processing; abnormal events that occur repeatedly within the same construction area and belong to the same abnormality type can be associated and merged to avoid generating multiple independent events for the same abnormal object. Through the above methods, edge devices can convert the collaborative sensing output data obtained from multi-task identification into structured, traceable, and schedulable abnormality handling results, providing a clear data foundation for the subsequent abnormal evidence information extraction and upload tasks.

[0070] S5. Extract the corresponding abnormal evidence information based on the abnormal handling results, construct an upload task based on the abnormal handling results and abnormal evidence information, schedule the upload task, and obtain the upload processing results. After generating hierarchical abnormal event information, such as Figure 5 As shown, the edge device extracts corresponding abnormal evidence information based on the abnormal handling results, and constructs an upload task based on the abnormal handling results and abnormal evidence information. It should be noted that the abnormal evidence information is used to prove the occurrence process, location, and abnormal state of the construction anomaly. It is not a standalone alarm text, but a combination of video image evidence data, recognition result data, and handling status data corresponding to the construction anomaly.

[0071] When extracting abnormal evidence information, the edge device first reads the event identifier, abnormality type, abnormality level, abnormality location, and trigger time from the hierarchical abnormal event information, and then locates the corresponding video channel and the image data to be identified based on the trigger time. For personnel safety abnormalities, the edge device extracts image frames from the video image information corresponding to the suspended platform or work platform area that can display the status of workers, safety helmets, reflective safety clothing, or the number of people in the suspended platform; for construction quality abnormalities, the edge device extracts image frames from the video image information corresponding to the plastering area or anchor construction area that can display the plastering status of the insulation board, the local status of the anchors, or the assembly sequence status of the anchors.

[0072] Furthermore, the edge device, centered on the trigger time, extracts alarm screenshots corresponding to the abnormal trigger moment from the video cache, and extracts video clips or keyframe sequences within a preset time range before and after the trigger time as supplementary evidence data. Alarm screenshots are used to visually display the on-site state at the moment of abnormal triggering, video clips are used to demonstrate the continuous changes before and after the anomaly occurs, and keyframe sequences are used to replace complete video clips for uploading when network limitations or insufficient upload resources exist. It should also be noted that the edge device retains the video channel number, camera device number, construction section number, target area type, and acquisition time when extracting evidence, ensuring that the abnormal evidence information remains consistent with the event identifier corresponding to the construction anomaly event.

[0073] After obtaining anomaly evidence, the edge device constructs an upload task based on the anomaly handling results and the anomaly evidence. The upload task converts anomaly event data requiring cloud-based management into schedulable, fragmentable, and recoverable transmission objects. The edge device configures each upload task with an event identifier, anomaly level, data type, data volume, upload time limit, and task status corresponding to the construction anomaly event. Specifically, the event identifier maintains a unique association between the upload task and the construction anomaly event; the anomaly level determines the priority of the upload task; the data type distinguishes between anomaly handling results, alarm screenshots, video clips, keyframe sequences, and identification result data; the data volume represents the size of the data to be uploaded; the upload time limit represents the time required to complete the upload; and the task status indicates whether the upload task is in a pending, sending, suspended, resumed, or completed state.

[0074] It should be noted that different anomaly levels correspond to different upload requirements. For shutdown-level anomalies, priority should typically be given to uploading the anomaly handling results, alarm screenshots, and necessary video clips to ensure timely warnings and verification by the cloud. For rectification-level anomalies, priority can be given to uploading the anomaly handling results and keyframe sequences, with complete video clips uploaded later when network conditions permit. By configuring data types and upload time limits for upload tasks, edge devices can prioritize the upload of high-level anomalies and high-value evidence data even in weak network environments.

[0075] Before scheduling uploads, edge devices acquire the current available network bandwidth and bandwidth fluctuation status. The current available network bandwidth characterizes the network capacity available for uploading abnormal data at the current moment or within the current scheduling window. The bandwidth fluctuation status characterizes whether the network bandwidth is stable, whether there is a significant decrease, or frequent jitter. Edge devices can determine the current available network bandwidth and bandwidth fluctuation status by statistically analyzing the recent data transmission rate, upload success status, retransmission status, and waiting time, and use this information as the basis for determining fragment size and adjusting the upload order.

[0076] When determining the fragment size, the edge device considers the current available network bandwidth, bandwidth fluctuation status, and the data volume of the upload task. When the available network bandwidth is high and bandwidth fluctuation is small, the edge device can configure a larger fragment size for the upload task to reduce the number of fragments and improve overall upload efficiency. When the available network bandwidth is low or bandwidth fluctuation is large, the edge device configures a smaller fragment size for the upload task to reduce the impact of a single fragment failure on the overall upload process. For video segments with large data volumes, the edge device divides them into multiple consecutive fragments for upload; for anomaly handling results or alarm screenshots with small data volumes, the edge device can send them as a single fragment for upload.

[0077] Furthermore, the edge devices determine the upload order of each upload task based on the anomaly level and upload time limit. Upload tasks corresponding to shutdown-level anomalies take priority over upload tasks corresponding to rectification-level anomalies; under the same anomaly level, upload tasks with more urgent upload time limits take priority over upload tasks with more lenient upload time limits; under the same event identifier, anomaly handling results and alarm screenshots take priority over sending complete video clips. Thus, in shared network or weak network conditions, edge devices can prioritize uploading core anomaly information required for cloud-based management, rather than having large video clips occupy the upload channel for extended periods.

[0078] When sending fragmented upload data based on the upload order, the edge device continuously monitors the actual service status of upload tasks of different anomaly levels within the current scheduling window. The actual service status can be characterized by the amount of data already sent, the number of completed tasks, the sending wait time, and the sending time taken for each anomaly level upload task. This is used to determine whether tasks of different anomaly levels have received upload service commensurate with their importance. The edge device compares the actual service status with preset service targets to obtain a warning level service bias status. The warning level service bias status is used to characterize whether there is insufficient or excessive service for upload tasks of a certain anomaly level within the current scheduling window.

[0079] In situations where there is insufficient service for high-level anomaly tasks, edge devices adjust the upload order of each upload task based on the service bias status of the warning level, giving higher priority to upload tasks with higher anomaly levels and insufficient service. For example, when an upload task corresponding to a shutdown-level anomaly event does not receive enough transmission opportunities within the current scheduling window due to network congestion, the edge device increases the transmission priority of this type of upload task during subsequent segment transmission. When a large video clip corresponding to a rectification-level anomaly event has already occupied a significant amount of upload resources, the edge device can reduce the transmission order of its subsequent segments to free up network resources for high-level anomaly tasks.

[0080] It should also be noted that, to avoid large-volume upload tasks blocking high-priority abnormal tasks for extended periods, the edge device checks for any pending upload tasks with higher priority before the current fragment is sent. If a higher-priority pending upload task exists, and the current upload task has already reached the fragment boundary, the edge device determines that the fragment boundary preemption condition is met. At the fragment boundary after the current fragment is sent, the device switches to the pending upload task and marks the switched upload task as suspended. This fragment boundary preemption method avoids directly interrupting the currently transmitting fragment data, reducing the risk of transmission failure and data corruption, while ensuring that high-priority abnormal tasks can obtain upload opportunities in a timely manner.

[0081] Upload tasks marked as suspended are not discarded; the edge device retains their event identifier, sent fragment position, remaining data volume, data type, and task status. When a switched upload task meets the sending conditions again, the edge device resumes sending based on its remaining data volume and continues uploading from the fragment position that was not yet completed. Meeting the sending conditions again can include network recovery, completion of current high-priority tasks, the pending task's upload timeout being approaching, or the completion of uploading core evidence corresponding to the same abnormal event. Through this suspension-to-resume process, the edge device can ensure that high-priority abnormal events are prioritized for upload while preventing low-priority or large-volume upload tasks from being permanently discarded.

[0082] After the upload scheduling is completed, the edge device generates the upload processing result. The upload processing result includes the event identifier, the data type uploaded, the range of data uploaded, the upload completion status, the suspended status, the resumed sending status, the upload failure status, and the corresponding time record. The upload processing result is used to indicate whether the abnormal handling results and abnormal evidence information have entered the cloud management process, and provides the transmission basis for the cloud to conduct early warning, retention, review, rectification, and tracing of abnormal events.

[0083] S6. Transmit the uploaded processing results to the cloud and perform cloud-based management and control, forming cloud-based management and control results that include early warning, retention, review, rectification and traceability, and feed back the review and rectification results to the edge task scheduling process.

[0084] After the upload processing result is generated, such as Figure 6 As shown, the edge device sends the uploaded processing results to the cloud, which receives and manages them. It should be noted that the uploaded processing results include not only the completion status of the anomaly handling results and anomaly evidence information, but also the event identifier corresponding to the construction anomaly, the uploaded data type, upload time, upload source, upload status, and data integrity status. The cloud aggregates various types of data for the same construction anomaly event through the event identifier, ensuring a consistent data link throughout the processes of edge identification, anomaly handling, evidence upload, cloud verification, rectification tracking, and subsequent tracing of the same anomaly event.

[0085] After receiving the uploaded processing results, the cloud first performs integrity verification and event attribution determination. Specifically, the cloud reads the event identifier, anomaly level, anomaly type, anomaly location, trigger time, data type, and upload source from the uploaded processing results, and determines whether a corresponding anomaly event record already exists for that event identifier. If the cloud does not find a corresponding anomaly event record, it creates a new anomaly event record based on the event identifier and writes the anomaly handling result and anomaly evidence information into that anomaly event record. If a corresponding anomaly event record already exists in the cloud, it associates the newly received data as supplementary data for that anomaly event record. Thus, even if alarm screenshots, keyframe sequences, and video clips of the same anomaly event are uploaded in batches, the cloud can still group them into the same closed-loop anomaly event record according to the same event identifier.

[0086] Furthermore, the cloud platform generates warning content based on the anomaly level and type. For work stoppage-level anomalies, the cloud marks them as high-priority warning objects and generates warning information corresponding to personnel safety anomalies. This warning information is used to indicate that the anomaly involves personnel safety risks during the construction of the external wall insulation suspended platform. For rectification-level anomalies, the cloud marks them as quality rectification warning objects and generates warning information corresponding to construction quality anomalies. This warning information is used to indicate that the anomaly involves anomalies related to plastering quality, anchor form, or anchor assembly sequence. It should also be noted that the warning content is stored in association with the event identifier, anomaly location, trigger time, and anomaly evidence information, enabling rapid identification of the source and evidence of the anomaly during subsequent review and rectification.

[0087] When storing abnormal evidence information in the cloud, alarm screenshots, keyframe sequences, video clips, identification result data, and abnormal handling results are uniformly linked and stored. Specifically, the cloud establishes an evidence index according to event identifiers and writes different data types into their corresponding storage locations, while recording the data source, upload time, data integrity status, and associated construction area. It should be noted that evidence retention is not simply saving images or videos separately, but rather binding evidence data with the abnormality type, abnormality level, abnormal location, trigger time, and handling status of the construction anomaly event, thereby forming a searchable, verifiable, and traceable chain of abnormal evidence.

[0088] During the review and management process, the cloud platform reads the corresponding anomaly handling results and anomaly evidence information based on the closed-loop record of anomaly events, and generates review objects. Review objects can include event identifiers, anomaly types, anomaly levels, anomaly locations, alarm screenshots, keyframe sequences, video clips, and edge recognition results. After review, the cloud platform records the review results, including anomaly established and anomaly not established states. When the review result is anomaly established, the cloud platform maintains the early warning status of the construction anomaly event and transfers it to the pending rectification or tracking and handling status. When the review result is anomaly not established, the cloud platform marks the construction anomaly event as excluded from review and retains the corresponding review records and evidence data for subsequent tracing of the edge recognition process and task scheduling process.

[0089] During the rectification management process, the cloud platform generates rectification tracking records based on the review results and the anomaly level. For work stoppage-level anomalies, the rectification tracking record records the handling status of personnel safety anomalies, including whether safety protection corrections have been completed, whether the work stoppage warning has been lifted, and whether subsequent construction tasks are allowed to continue. For rectification-level anomalies, the rectification tracking record records the handling status of construction quality anomalies, including whether plastering quality corrections have been completed, whether anchor installation status adjustments have been completed, whether process sequence anomalies have been handled, and whether subsequent reviews have been passed. Furthermore, when the rectification status meets the completion conditions, the cloud platform updates the construction anomaly event to a rectification completed status and writes the rectification completion time, rectification results, and related evidence into the anomaly event closed-loop record.

[0090] It should be noted that the closed-loop record for abnormal events is used throughout the entire process of early warning, retention, review, rectification, and traceability. The closed-loop record for abnormal events includes at least the event identifier, abnormality type, abnormality level, abnormality location, trigger time, abnormality handling result, abnormality evidence information, early warning status, review result, rectification status, and traceability record. Through this closed-loop record, the cloud can reconstruct the entire process of the construction abnormality event from edge detection and triggering to cloud-based handling completion during subsequent queries, preventing abnormal events from merely remaining at the level of a one-time alarm.

[0091] Furthermore, the cloud updates the establishment status of construction anomalies based on the review results, and updates the handling status based on the rectification status. The establishment status indicates whether the anomaly is still considered a valid anomaly after review; the handling status indicates whether the anomaly is currently in the pending review, pending rectification, rectification in progress, rectification completed, or review ruled out state. After updating the status, the cloud feeds back the updated establishment and handling statuses to the edge task scheduling process, enabling edge devices to adjust the generation and scheduling of subsequent construction identification tasks based on the cloud feedback.

[0092] After receiving the establishment and processing status feedback from the cloud, the edge device updates the task priority, task admission status, or identification strategy for subsequent scheduling cycles. The task admission status is used to characterize whether the construction identification task is allowed to enter the edge task scheduling process, and it includes the admitted status, the delayed admission status, and the inadmissible status.

[0093] Specifically, when the cloud reports that an anomaly has been identified in a construction area and is still in an incomplete handling state, the edge device increases the scheduling priority of the relevant identification tasks for that construction area and adds anomaly review tasks or continuous monitoring tasks for that area in subsequent scheduling cycles. When the cloud reports that an anomaly has been rectified, the edge device can reduce the anomaly review priority of the area corresponding to that event and restore that area to the normal identification strategy. When the cloud reports that an anomaly is in a review exclusion state, the edge device can adjust the task admission conditions or identification strategies for similar anomalies to reduce repeated false alarms in the same scenario.

[0094] It should also be noted that cloud feedback does not directly replace the real-time identification process of edge devices, but rather serves as a status input for the edge task scheduling process, assisting edge devices in allocating identification resources more rationally in subsequent scheduling cycles. Through the feedback of cloud management results to the edge task scheduling process, edge devices can differentiate and process confirmed abnormal areas, areas where rectification is incomplete, areas where rectification is complete, and areas excluded after review, thereby forming a closed-loop collaboration between edge identification, abnormal uploading, cloud management, and edge rescheduling.

[0095] Through the aforementioned cloud-based management and control process, the uploaded processing results are converted into cloud-based management and control results containing early warning, retention, review, rectification, and traceability information. The review and rectification results are then fed back to the edge task scheduling process. Therefore, this embodiment enables not only the identification and uploading of safety and quality anomalies at the external wall insulation suspended platform construction site, but also the formation of a structured closed-loop record in the cloud. This further influences the priority, access status, and identification strategy of subsequent edge identification tasks, thereby improving the continuity of collaborative perception and intelligent management and control at the external wall insulation suspended platform construction site.

[0096] Example 2 This embodiment provides a collaborative sensing and intelligent control system for the construction of external wall insulation suspended scaffolding, such as... Figure 2 As shown, the system includes edge devices and a cloud-based management and control unit; The edge device includes a video process association module, a task generation module, an edge scheduling and identification module, an anomaly classification module, and an upload scheduling module; The video process association module is used to acquire video image information and construction process information of the external wall insulation suspended platform construction site, preprocess the video image information to obtain the image data to be identified, and establish the correspondence between the image data to be identified and the construction process information. The task generation module is used to determine the recognition task type corresponding to the current construction stage based on the construction progress information and the image data to be recognized, and to generate a set of construction recognition tasks. The edge scheduling and identification module is used to schedule edge tasks for the construction identification task set and perform construction status identification according to the scheduling results, so as to obtain collaborative perception output data including personnel safety identification results and construction quality identification results. The anomaly classification module is used to identify construction anomalies based on collaborative sensing output data, classify and process these anomalies, and obtain anomaly handling results. The upload scheduling module is used to extract the corresponding abnormal evidence information based on the abnormal handling results, construct upload tasks based on the abnormal handling results and abnormal evidence information, schedule the upload tasks, obtain the upload processing results, and send the upload processing results to the cloud management unit. The cloud-based management and control unit is used to receive the uploaded processing results and perform cloud-based management and control, forming cloud-based management and control results that include early warning, retention, review, rectification and traceability, and feeding back the review and rectification results to the edge scheduling and identification module.

[0097] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A collaborative sensing and intelligent control method for the construction of external wall insulation suspended scaffolding, characterized in that, Includes the following steps: S1. Acquire video image information and construction progress information of the construction site of the external wall insulation suspended platform, preprocess the video image information to obtain image data to be identified, and establish the correspondence between the image data to be identified and the construction progress information; S2. Based on the construction progress information and the image data to be identified, determine the identification task type corresponding to the current construction stage and generate a construction identification task set; S3. Perform edge task scheduling on the construction identification task set, and perform construction status identification according to the scheduling result to obtain collaborative perception output data including personnel safety identification results and construction quality identification results; S4. Based on the collaborative sensing output data, determine the construction anomaly events, classify the construction anomaly events for processing, and obtain the anomaly handling results. S5. Extract the corresponding abnormal evidence information based on the abnormal handling result, construct an upload task based on the abnormal handling result and the abnormal evidence information, schedule the upload task for upload, and obtain the upload processing result. S6. Transmit the uploaded processing results to the cloud and perform cloud management to form a cloud management result that includes early warning, retention, review, rectification and traceability, and feed back the review and rectification results to the edge task scheduling process.

2. The collaborative perception and intelligent control method for the construction of external wall insulation suspended scaffolding according to claim 1, characterized in that, In step S2, generating the construction identification task set includes: The current construction stage is determined based on the construction progress information, and the recognition task type is determined based on the target area type corresponding to the current construction stage and the image data to be recognized. When the current construction stage is the basket preparation stage and the target area type is a suspended basket or work platform area, a construction identification task set is generated, which includes personnel safety status identification task and suspended basket personnel verification task. When the current construction stage is the plastering construction stage and the target area type is the plastering operation area, a construction identification task set containing plastering quality identification tasks is generated. When the current construction stage is the anchor installation stage and the target area type is the anchor installation area, a construction identification task set is generated, which includes anchor shape identification task and anchor assembly sequence identification task.

3. The collaborative perception and intelligent control method for the construction of external wall insulation suspended scaffolding according to claim 1, characterized in that, In step S3, edge task scheduling for the construction identification task set includes: Extract the attribute information of each construction identification task in the construction identification task set. The attribute information includes task type, generation time, corresponding construction stage and task status, and obtain the resource status information of edge devices. Based on the attribute information and the edge device resource status information, the scheduling priority of each construction identification task is determined, and each construction identification task is assigned to a task queue of different levels according to the scheduling priority. Based on the task queue level and edge device resource status information, determine the computing power resource quota and model call order corresponding to each task queue, and perform construction status identification according to the computing power resource quota and model call order.

4. The collaborative perception and intelligent control method for the construction of external wall insulation suspended scaffolding according to claim 1, characterized in that, In step S3, the construction status identification includes: When the construction identification task set includes a plastering quality identification task, the image data to be identified corresponding to the plastering quality identification task is obtained, the image data to be identified is segmented into regions to obtain insulation board area information and plastering coverage area information, and the information of unplastered areas is determined based on the insulation board area information and plastering coverage area information. The overall plastering coverage status is determined based on the insulation board area information and the plastering coverage area information. When the overall grout coverage does not meet the preset requirements, a grout quality identification result representing an abnormal grout quality is generated. When the overall grouting coverage meets the preset requirements, the local grouting defect status is determined based on the distribution of the ungrouted area information in the key area of ​​the insulation board, and a grouting quality identification result is generated based on the overall grouting coverage and the local grouting defect status.

5. The collaborative perception and intelligent control method for the construction of external wall insulation suspended scaffolding according to claim 1, characterized in that, In step S3, the construction status identification includes: When the construction identification task set includes anchor morphology identification task and anchor assembly sequence identification task, the image data to be identified corresponding to the anchor construction area is obtained, the sleeve target and nail core target in the image data to be identified are identified, and the instance-level association relationship within the same anchor hole position is established according to the spatial matching relationship between the sleeve target and the nail core target. Based on the instance-level association, sleeve events and nail core events corresponding to the anchoring hole positions are generated, and the sleeve events and nail core events in consecutive image frames are stabilized and confirmed to obtain a stable event sequence. Based on the stable event sequence and the instance-level association, state transition judgment is performed to determine the anchor status of the corresponding anchor hole position. The anchor status includes the installation completed state, the component missing state, and the installation abnormal state, and anchor morphology recognition result and anchor assembly timing recognition result are generated.

6. The collaborative perception and intelligent control method for construction of external wall insulation suspended scaffolding according to claim 1, characterized in that, In step S4, the graded processing of the construction anomaly includes: The abnormality type, abnormal location, and trigger time of the construction abnormal event are determined based on the collaborative sensing output data, and the abnormality level of the construction abnormal event is determined based on the abnormality type. When the anomaly type is a personnel safety anomaly, the construction anomaly event is identified as a work stoppage-level anomaly event, and a work stoppage-level anomaly handling result is generated. When the anomaly type is a construction quality anomaly, the construction anomaly event is identified as a rectification-level anomaly event, and a rectification-level anomaly handling result is generated. An event identifier is generated for the construction anomaly, and the event identifier, anomaly type, anomaly level, anomaly location, trigger time, and anomaly handling result are associated to form hierarchical anomaly event information.

7. The collaborative perception and intelligent control method for construction of external wall insulation suspended scaffolding according to claim 6, characterized in that, In step S5, the uploading and scheduling of the anomaly handling results and anomaly evidence information includes: An upload task is constructed based on the abnormal handling results and abnormal evidence information, and the upload task is configured with the event identifier, abnormality level, data type, data volume, upload time limit and task status corresponding to the construction abnormal event; Obtain the current available network bandwidth and bandwidth fluctuation status, determine the corresponding fragment size based on the current available network bandwidth, bandwidth fluctuation status and the data volume of the upload task, and divide the upload task into fragments for data upload according to the fragment size; The upload order of each upload task is determined based on the anomaly level and upload time limit, and the segmented upload data is sent based on the upload order to obtain the upload processing result.

8. The collaborative perception and intelligent control method for construction of external wall insulation suspended scaffolding according to claim 7, characterized in that, In step S5, sending the fragmented upload data based on the upload order includes: Based on the anomaly level and the amount of data sent for each upload task, the actual service status of upload tasks with different anomaly levels within the current scheduling window is statistically analyzed, and the actual service status is compared with the preset service target to obtain the warning level service bias status. The warning level corresponds to the anomaly level and is used to distinguish the service guarantee level of different abnormal events during the upload scheduling process; Pre-configured service targets can be based on the target upload ratio, target sending opportunities, or target completion time limit for work stoppage-level and rectification-level abnormal events. The upload order of each upload task is adjusted according to the service bias status of the warning level, so that upload tasks with higher abnormal levels and insufficient service receive higher sending priority. Before the current segment is sent, check if there is an uploading task with a higher sending priority than the current uploading task. When the detection result meets the segment boundary preemption condition, switch to the uploading task to be sent at the segment boundary after the current segment is sent and mark the switched uploading task as suspended. When the switched upload task meets the sending conditions again, it resumes sending based on the amount of data remaining to be sent.

9. The collaborative perception and intelligent control method for construction of external wall insulation suspended scaffolding according to claim 6, characterized in that, In step S6, cloud-based management includes: After receiving the uploaded processing results, the cloud will associate and store the abnormal handling results, abnormal evidence information, review results and rectification status according to the event identifier corresponding to the construction abnormal event, forming a closed-loop record of the abnormal event; The establishment status of the construction anomaly is updated based on the review results, the handling status of the construction anomaly is updated based on the rectification status, and the updated establishment and handling status are fed back to the edge task scheduling process. The edge task scheduling process updates the task priority, task admission status, or identification strategy in subsequent scheduling cycles based on the established status and the processing status.

10. A collaborative sensing and intelligent control system for the construction of external wall insulation suspended scaffolding, characterized in that, The collaborative sensing and intelligent control method for performing any one of claims 1 to 9 for construction of external wall insulation suspended scaffolding, the system comprising edge devices and a cloud control unit; The edge device includes a video process association module, a task generation module, an edge scheduling and identification module, an anomaly classification module, and an upload scheduling module; The video process association module is used to acquire video image information and construction process information of the external wall insulation suspended platform construction site, preprocess the video image information to obtain image data to be identified, and establish the correspondence between the image data to be identified and the construction process information. The task generation module is used to determine the recognition task type corresponding to the current construction stage based on the construction progress information and the image data to be recognized, and to generate a set of construction recognition tasks. The edge scheduling and identification module is used to schedule edge tasks for the construction identification task set and perform construction status identification according to the scheduling results to obtain collaborative perception output data that includes personnel safety identification results and construction quality identification results. The anomaly classification module is used to determine construction anomalies based on the collaborative sensing output data, classify the construction anomalies, and obtain anomaly handling results. The upload scheduling module is used to extract corresponding abnormal evidence information based on the abnormal handling result, construct an upload task based on the abnormal handling result and the abnormal evidence information, schedule the upload task, obtain the upload processing result, and send the upload processing result to the cloud management unit. The cloud management unit is used to receive the uploaded processing results and perform cloud management, forming a cloud management result that includes early warning, retention, review, rectification and traceability, and feeding back the review and rectification results to the edge scheduling and identification module.

Images