Construction site violation visual identification and dynamic closed-loop management method based on edge computing

Abstract

The application discloses a construction site violation visual identification and dynamic closed-loop management and control method based on edge computing, and comprises the following steps: constructing a swinging physical model of a hoisted object based on collected real-time operation parameters, environmental wind speed data and positioning data of surrounding workers, predicting the swinging angle and swinging envelope radius of the hoisted object in a future predetermined time period, and generating a dynamic risk envelope area; calculating the collision probability of each worker and the dynamic risk envelope area; generating hierarchical early warning information according to the comparison result of the collision probability and a preset threshold value, and sending the hierarchical early warning information to a management platform; and in response to confirmation, on-site disposal and rectification feedback operations performed by a management personnel through an APP, receiving disposal feedback information from the management platform. The application realizes real-time collision probability evaluation and hierarchical early warning of a dynamic risk area of a hoisted object, and solves the problem that an existing static electronic fence cannot adapt to a dynamic irregular dangerous area caused by swinging of a hoisted object, leading to invalid alarm or missed report.

Description

Technical Field

[0001] This invention belongs to the field of construction site management technology, and in particular relates to a method for visual recognition and dynamic closed-loop management of construction site violations based on edge computing. Background Technology

[0002] Tower cranes are indispensable large-scale lifting equipment on construction sites. During operation, the loads (such as steel bars, concrete buckets, and formwork) sway due to inertia, wind, and other factors during lifting, rotation, luffing, and braking. When the load sways, the area directly below it and covered by the swaying trajectory constitutes a dynamic high-risk zone. Personnel entering this zone are highly susceptible to collisions or being struck by falling objects. Therefore, real-time identification of the load's swaying state and delineation of dynamic danger zones are critical issues in construction site safety management.

[0003] In existing construction site safety management, the following technical means are commonly used to protect tower crane operating areas. First, static electronic fence technology: Based on personnel positioning systems (such as UWB high-precision positioning, Bluetooth AOA positioning), fixed danger zone polygons (e.g., the fan-shaped area covered by the tower crane boom) are pre-drawn on the management platform map. When a worker wearing a positioning terminal enters this fixed area, the system triggers an alarm. This technology has been deployed in many large infrastructure construction sites to restrict personnel from entering the tower crane's slewing radius. Second, tower crane operation status monitoring systems: Real-time operating parameters are collected through angle sensors, amplitude sensors, and load sensors installed on the tower crane for overload protection, collision prevention, etc., and are commonly found in tower crane black box products. Third, video surveillance and manual inspection: Cameras are installed at the highest points of the construction site, and safety management personnel take turns viewing the footage from the monitoring center. When personnel are found entering under the boom, they are stopped by verbal warnings via walkie-talkie.

[0004] However, the aforementioned existing technologies still suffer from a major problem in practical applications: they cannot provide real-time warnings for dynamic, irregularly shaped hazardous areas caused by the swinging of suspended loads. Static electronic fences cover the projected area of ​​the tower crane boom with a fixed fan-shaped or circular region. However, when the load swings, the actual hazardous area is not the entire static area below the boom, but a dynamic envelope that constantly changes with the swing angle and direction. Static fences often generate numerous invalid alarms due to their large area (personnel are below the boom but far from the swinging load), leading to safety personnel becoming desensitized to alarms; or they may miss alarms due to their small area (the swinging load exceeds the static fence's range), resulting in the failure to identify the real risks. Tower crane status monitoring systems only focus on the safety of the equipment itself (such as torque limitations) and do not involve warnings for personnel below. Video surveillance relies on human attention and cannot calculate the spatiotemporal relationship between the swinging trajectory of the load and the position of personnel in real time, resulting in a significant delay in response. Therefore, the following solutions are proposed to address these problems. Summary of the Invention

[0005] The purpose of this invention is to provide a method for visual recognition and dynamic closed-loop management of construction site violations based on edge computing. By integrating the prediction of the swing trajectory of tower crane loads with the spatiotemporal collision probability calculation method of personnel movement trends, it can generate dynamic danger envelope areas that are adapted to the swing characteristics of loads in real time and implement graded early warning. This solves the problem that existing static electronic fences cannot adapt to the dynamic irregular danger areas caused by the swing of loads, resulting in invalid alarms or missed alarms.

[0006] To solve the above-mentioned technical problems, the present invention is achieved through the following technical solution:

[0007] This invention relates to a method for visual recognition and dynamic closed-loop management of construction site violations based on edge computing, comprising the following steps: Step S1: In response to the identification of the tower crane hoisting operation scene, collect the tower crane's real-time operating parameters, ambient wind speed data, and the location data of surrounding workers; Step S2: Based on the collected data, construct a physical model of the suspended object's swing, predict the swing angle and swing envelope radius of the suspended object within a predetermined time period in the future, and then generate a dynamic risk envelope region; Step S3: Based on the worker's current position and velocity vector, predict the movement trajectory of each worker within a predetermined time period, and calculate the collision probability of each worker with the dynamic risk envelope area. Step S4: Based on the comparison result between the collision probability and the preset threshold, generate graded early warning information and send the graded early warning information to the management platform so that the management platform can push alarms to the corresponding worker terminals and management personnel terminals; Step S5: Respond to the confirmation, on-site handling and rectification feedback operations performed by the management personnel through the APP, and receive handling feedback information from the management platform to achieve a closed management loop.

[0008] Furthermore, the generation of the dynamic risk envelope region in step S2 specifically includes: solving the following swing differential equation to predict the angle between the suspension rope and the vertical direction. and its azimuth :

[0009] In the formula, The angle between the suspension rope and the vertical direction; It is a time variable; This refers to the air damping coefficient; For the weight of the suspended object; This refers to the length of the suspension rope. It is the acceleration due to gravity; The tangential acceleration at the point of suspension of the object; The component of the wind load in the swing direction is represented; the coordinates of the projected point of the suspended object on the ground are expressed as the coordinates of the tower crane center, the amplitude of the luffing trolley, the length of the hoisting rope, and the predicted swing angle. and azimuth angle The operation relationship is calculated, and the projection point and swing envelope radius of all times within the future predetermined time period are calculated. The union of the circular regions centered on the projection point and with the sum of the swing envelope radius and the safety margin as the radius is taken as the dynamic risk envelope region.

[0010] Furthermore, the method for calculating the collision probability in step S3 is as follows: For each worker, a uniform linear motion model is used to predict their position within a predetermined future time period, and the collision probability is defined as:

[0011] In the formula, For the first The probability of collision for each worker; For indicator functions; Predict the location of workers; This is a dynamic risk envelope region; This is a weighting function that decays over time; The current moment; The predetermined time period; the weighting function The value increases as it gets closer to the current moment.

[0012] Furthermore, the graded warning in step S4 includes: triggering a level 1 alarm when the collision probability is greater than or equal to a first preset threshold; triggering a level 2 alarm when the collision probability is between a second preset threshold and a first preset threshold; and only logging when the collision probability is less than a second preset threshold but greater than zero; wherein the first preset threshold is greater than the second preset threshold.

[0013] Furthermore, for workers who trigger a Level 1 or Level 2 alarm, the management platform also performs the following linked operations: highlighting the dynamic risk envelope area and the worker's real-time location and predicted trajectory on the GIS map; sending a voice prompt containing the evacuation direction to the positioning terminal worn by the worker; and pushing an alarm card containing the worker's identity, collision probability, and predicted location to the APP of the team leader of the worker's work group and the safety officer of the tower crane operation area.

[0014] Further, the handling feedback information received in step S5 is generated by the following process: In response to the manager's touch operation of the confirmation button on the APP, the APP records the confirmation time and starts timing the handling duration; In response to the on-site intervention photos, rectification instructions and handling result codes uploaded by the manager, the APP sends the intervention information to the management platform; The management platform calls the relationship between the worker's real-time location and the dynamic risk envelope area to verify the rectification. When the worker has left the dynamic risk envelope area and the direction of movement is away from the area, it is determined to be a valid closed loop and a closed loop record is generated; If the handling duration exceeds the preset handling time limit and no closed loop record is received, the management platform automatically reports the alarm information level by level to the terminal of the next higher-level manager and sends the final handling result as the handling feedback information to the edge computing node.

[0015] Furthermore, it also includes a closed-loop rate statistics and parameter adaptive adjustment step: the edge computing node receives a parameter adjustment instruction from the management platform, and updates the predetermined time period, preset threshold, and safety margin parameters of the dynamic risk envelope region stored locally according to the parameter adjustment instruction; wherein, the parameter adjustment instruction is generated by the management platform according to the following process: statistically analyze the effective closed-loop rate of each edge computing node in multiple consecutive statistical periods, and when the effective closed-loop rate of an edge computing node is lower than the preset standard, the management platform issues the parameter adjustment instruction to the node, the parameter adjustment instruction including at least one of the following: extending the predetermined time period, reducing the first preset threshold, and increasing the safety margin of the dynamic risk envelope region.

[0016] Furthermore, the method for edge computing nodes to identify tower crane hoisting operation scenarios in step S1 is as follows: by acquiring video streams through high-altitude cameras, and using a deep learning object detection model to detect in multiple consecutive frames that there are suspended objects below the hook and the height of the suspended objects above the ground is greater than a preset height threshold, it is determined that the tower crane hoisting operation is in progress.

[0017] Furthermore, the real-time operating parameters of the tower crane collected in step S1 include the boom slewing angle, luffing trolley amplitude, hook height, lifting weight, and rope length; the environmental wind speed data includes the average wind speed and the angle between the wind direction and the boom axis; and the worker's positioning data includes the worker's three-dimensional coordinates and movement speed vector.

[0018] An edge computing-based visual recognition and dynamic closed-loop management device for construction site violations, used to execute the method described in any of the above-mentioned embodiments, including: The front-end sensing unit is used to collect video streams, tower crane operating parameters, ambient wind speed and worker positioning data at the construction site, and transmit the collected data to the edge computing node; The edge computing node is communicatively connected to the front-end sensing unit. In response to the identification of tower crane hoisting operation scenario, it constructs a physical model of the hoisting object's swing based on the collected data and solves the swing differential equation. It predicts the swing trajectory of the hoisting object within a predetermined time period to generate a dynamic risk envelope area. At the same time, it predicts the worker's motion trajectory based on the worker's current position and velocity vector, calculates the collision probability of each worker with the dynamic risk envelope area, and uploads the collision probability and related data to the management platform. The management platform is connected to the edge computing nodes and is used to generate graded early warning information based on the comparison result of the collision probability and the preset threshold, push alarms to the corresponding worker terminals and management personnel terminals, and count the closed loop rate of each edge computing node. Based on the result that the closed loop rate is lower than the preset standard, the predetermined time period, the preset threshold and the safety margin parameters of the dynamic risk envelope area are automatically adjusted. The APP terminal communicates with the management platform and is used by managers to confirm alarms, handle on-site and provide feedback on rectification, record handling time and upload rectification verification results to form a management closed loop.

[0019] The present invention has the following beneficial effects: This invention achieves graded early warning based on collision probability by integrating the prediction of the swing trajectory of tower crane loads with the movement trend of personnel, which helps reduce invalid alarms that do not match dynamic risks. Physical parameters such as load, wind speed, and rope length collected by edge computing nodes are used to construct dynamic hazard envelope areas, improving the adaptability of risk areas. Spatiotemporal collision analysis combined with personnel positioning data makes the early warning results more targeted. At the same time, the prediction parameters of edge nodes are adjusted in reverse through closed-loop data processing via the APP, forming a feedback mechanism that can gradually improve the risk response matching degree of the system during use.

[0020] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0021] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a flowchart illustrating the edge computing-based visual recognition and dynamic closed-loop management method for construction site violations according to the present invention. Figure 2 This is a schematic diagram of the construction site violation visual recognition and dynamic closed-loop management device based on edge computing according to the present invention. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] Please see Figure 1 As shown, this invention is a method for visual recognition and dynamic closed-loop management of construction site violations based on edge computing, comprising the following steps: Step S1: In response to recognizing the tower crane hoisting operation scenario, the edge computing node collects the tower crane's real-time operating parameters, ambient wind speed data, and the location data of surrounding workers; Step S2: Based on the collected data, construct a physical model of the suspended object's swing, predict the swing angle and swing envelope radius of the suspended object within a predetermined time period in the future, and then generate a dynamic risk envelope region; Step S3: Based on the worker's current position and velocity vector, predict the movement trajectory of each worker within a predetermined time period, and calculate the collision probability of each worker with the dynamic risk envelope area. Step S4: Based on the comparison result between the collision probability and the preset threshold, generate graded early warning information and send the graded early warning information to the management platform so that the management platform can push alarms to the corresponding worker terminals and management personnel terminals; Step S5: Respond to the confirmation, on-site handling and rectification feedback operations performed by the management personnel through the APP, and receive handling feedback information from the management platform to achieve a closed management loop.

[0025] Step S2, generating the dynamic risk envelope region, specifically includes: solving the following swing differential equation to predict the angle between the suspension rope and the vertical direction. and its azimuth :

[0026] In the formula, The angle between the suspension rope and the vertical direction; It is a time variable; This refers to the air damping coefficient; For the weight of the suspended object; This refers to the length of the suspension rope; It is the acceleration due to gravity; The tangential acceleration at the suspension point of the object; The component of the wind load in the swing direction is represented; the coordinates of the projected point of the suspended object on the ground are expressed as the coordinates of the tower crane center, the amplitude of the luffing trolley, the length of the hoisting rope, and the predicted swing angle. and azimuth angle The operation relationship is calculated, and the projection point and swing envelope radius of all times within the future predetermined time period are calculated. The union of the circular regions centered on the projection point and with the sum of the swing envelope radius and the safety margin as the radius is taken as the dynamic risk envelope region.

[0027] The method for calculating the collision probability in step S3 is as follows: For each worker, a uniform linear motion model is used to predict their position within a predetermined future time period, and the collision probability is defined as:

[0028] In the formula, For the first The probability of collision for each worker; For indicator functions; Predict the location of workers; This is a dynamic risk envelope region; This is a weighting function that decays over time; The current moment; The predetermined time period; the weighting function The value increases as it gets closer to the current moment.

[0029] The graded warning in step S4 includes: triggering a level 1 alarm when the collision probability is greater than or equal to the first preset threshold; triggering a level 2 alarm when the collision probability is between the second preset threshold and the first preset threshold; and only logging when the collision probability is less than the second preset threshold but greater than zero; wherein the first preset threshold is greater than the second preset threshold.

[0030] For workers who trigger a Level 1 or Level 2 alarm, the management platform also performs the following linked operations: highlights the dynamic risk envelope area and the worker's real-time location and predicted trajectory on the GIS map; sends a voice prompt containing the evacuation direction to the positioning terminal worn by the worker; and pushes an alarm card containing the worker's identity, collision probability, and predicted location to the APP of the team leader of the worker's work group and the safety officer of the tower crane operation area.

[0031] The handling feedback information received in step S5 is generated by the following process: In response to the manager's touch operation of the confirmation button on the APP, the APP records the confirmation time and starts timing the handling duration; In response to the on-site intervention photos, rectification instructions and handling result codes uploaded by the manager, the APP sends the intervention information to the management platform; The management platform calls the relationship between the worker's real-time location and the dynamic risk envelope area to verify the rectification. When the worker has left the dynamic risk envelope area and the direction of movement is away from the area, it is determined to be a valid closed loop and a closed loop record is generated; If the handling duration exceeds the preset handling time limit and no closed loop record is received, the management platform automatically reports the alarm information to the terminal of the next higher level manager and sends the final handling result as handling feedback information to the edge computing node.

[0032] It also includes a closed-loop rate statistics and parameter adaptive adjustment step: the edge computing node receives a parameter adjustment instruction from the management platform and updates the predetermined time period, preset threshold, and safety margin parameters of the dynamic risk envelope region stored locally according to the parameter adjustment instruction; wherein, the parameter adjustment instruction is generated by the management platform according to the following process: the effective closed-loop rate of each edge computing node is counted in multiple consecutive statistical periods, and when the effective closed-loop rate of an edge computing node is lower than the preset standard, the management platform issues a parameter adjustment instruction to the node, which includes at least one of the following: extending the predetermined time period, reducing the first preset threshold, and increasing the safety margin of the dynamic risk envelope region.

[0033] The method for edge computing nodes to identify tower crane hoisting operation scenarios in step S1 is as follows: by collecting video streams through high-altitude cameras, and using a deep learning object detection model to detect in multiple consecutive frames that there are suspended objects below the hook and the height of the suspended objects above the ground is greater than a preset height threshold, it is determined that the tower crane hoisting operation is in progress.

[0034] The real-time operating parameters of the tower crane collected in step S1 include the boom slewing angle, luffing trolley amplitude, hook height, lifting weight, and rope length; the environmental wind speed data includes the average wind speed and the angle between the wind direction and the boom axis; and the worker's positioning data includes the worker's three-dimensional coordinates and movement speed vector.

[0035] Please see Figure 2 As shown, the present invention is a construction site violation visual recognition and dynamic closed-loop management device based on edge computing, used to perform the method described above, comprising: The front-end sensing unit is used to collect video streams, tower crane operating parameters, ambient wind speed and worker positioning data at the construction site, and transmit the collected data to the edge computing node; Edge computing nodes communicate with front-end sensing units to respond to the identification of tower crane hoisting operation scenarios. Based on the collected data, they construct a physical model of the hoisted object's swing and solve the swing differential equation. They predict the swing trajectory of the hoisted object within a predetermined time period to generate a dynamic risk envelope area. At the same time, they predict the worker's movement trajectory based on the worker's current position and velocity vector, calculate the collision probability of each worker with the dynamic risk envelope area, and upload the collision probability and related data to the management platform. The management platform communicates with edge computing nodes and generates graded early warning information based on the comparison between collision probability and preset threshold. It pushes alarms to the corresponding worker terminals and management personnel terminals, and calculates the closed-loop rate of each edge computing node. Based on the result that the closed-loop rate is lower than the preset standard, it automatically adjusts the safety margin parameters of the preset time period, preset threshold and dynamic risk envelope area. The APP terminal communicates with the management platform and is used by managers to confirm alarms, handle on-site and provide feedback on rectification, record handling time and upload rectification verification results to form a management closed loop.

[0036] The specific application of this embodiment is as follows: Step S1: Identify the tower crane operation scene and collect multi-dimensional status data at the edge. The AI ​​video analysis box, deployed in the tower crane cab or on the tower crane structure, analyzes the video stream in real time through a connected high-altitude camera. When the deep learning object detection model built into the AI ​​video analysis box detects a suspended object below the hook for three consecutive frames and the suspended object is more than 1.5 meters above the ground, it determines that the tower crane hoisting operation is in progress. After this state is triggered, the edge computing node immediately performs the following data acquisition actions: Real-time operating parameters, including the current slewing angle of the boom, are read from the tower crane PLC controller via the Modbus TCP protocol. The distance between the luffing trolley and the center of the tower body Hook height Current lifting weight Length of the hoisting rope .

[0037] The average wind speed at the current work site over the past 10 seconds is collected using miniature wind speed sensors connected to the edge node. and the angle between the instantaneous maximum wind direction and the boom axis .

[0038] Synchronize the latest real-time coordinates of all worker positioning terminals (UWB or Bluetooth AOA) within a 200-meter radius of the tower crane from the management platform. and the corresponding movement speed vector ,in, Indicates the first One worker; For the first A worker on the horizontal plane Direction coordinates; For the first A worker on the horizontal plane Direction coordinates; For the first Elevation coordinates of each worker; For the velocity vector in Components in direction; For the velocity vector in Components in direction.

[0039] All the collected data were tagged with a uniform timestamp. And cache it locally on the edge nodes.

[0040] Step S2: Construct a prediction model for the swing trajectory of the suspended load and calculate the dynamic hazard envelope region. The edge computing nodes establish a physical model of the suspended load's swing based on the tower crane status data collected in step S1. The suspended load is considered as a simple pendulum system subjected to gravity, rope tension, air resistance, and inertial forces. After leaving the ground, the load will swing under the influence of tower boom rotation, braking, or wind disturbances. Swing angle. The variation over time can be approximately described by the following oscillating differential equation:

[0041] In the formula, The angle between the suspension rope and the vertical direction; It is a time variable; Angular velocity; Angular acceleration; This is the air damping coefficient, and an empirical value of 0.05 is used for the construction site environment. For the weight of the suspended object, It is the acceleration due to gravity; The tangential acceleration at the suspension point of the object; The component of the wind load in the oscillation direction is determined based on the real-time wind speed. Angle with wind direction calculate: , air density, This is the resistance coefficient of the suspended object. The windward area of ​​the suspended object. This is a constant factor in the wind pressure calculation formula.

[0042] Edge computing nodes utilize the fourth-order Runge-Kutta method with a step size of 0.1 seconds, starting from the current time. Predicting the future Swing angle within seconds and azimuth angle The coordinates of the projected point of the suspended object on the ground. for:

[0043]

[0044] In the formula, Fix the coordinates of the tower crane's center point; This refers to the distance between the luffing trolley and the center of the tower.

[0045] Swing envelope radius Define dynamic hazard areas. With the ground projection point of the suspended object as the center and a radius of... The circular region, in the formula, The meter is a safety margin, taking into account the size of the suspended object and the reaction distance of personnel.

[0046] Calculate all moments within the next 5 seconds The dangerous areas are identified, and the union of these areas is taken as the final dynamic risk envelope area. .

[0047] Step S3: Trajectory prediction and collision probability calculation based on personnel movement trends For each worker obtained in step S1 Its current position is known. and velocity vector Assuming the worker's motion is uniform linear motion in the short term, then in the future... The predicted location is:

[0048]

[0049] In the formula, For the first A worker in the future Prediction coordinate; For the first A worker in the future Prediction coordinate; Use the current timestamp. Define the worker. The condition for a collision with a dynamic hazard zone is that there exists an arbitrary predicted moment. This allows workers to predict locations. Entering the dynamic risk envelope region Calculate the collision probability. as follows:

[0050] wherein, is an indicator function, which takes the value 1 when the condition is satisfied, and 0 otherwise; is a weight function that decays over time, , is the base of the natural logarithm; is the attenuation coefficient, with a value of 0.5 .

[0051] If for a worker has not entered within the entire prediction interval , then . If the worker is already located in the envelope area, the system automatically sets . This probability value is reported to the management platform together with the worker's ID, current position, and predicted trajectory data.

[0052] Step S4: Hierarchical warning generation and dynamic fence visualization After the management platform receives the collision probability of each worker , it executes the hierarchical warning logic: If , it is determined as high risk, and a first-level alarm is triggered immediately; If , it is determined as caution, and a second-level alarm is triggered; If and greater than 0, it is determined as observation, only the log is recorded and no alarm is actively pushed; If , no alarm operation is performed.

[0053] For workers who trigger a first-level or second-level alarm, the management platform performs the following linkage operations: On the GIS map at the platform end, the dynamic risk envelope area is highlighted with translucent red , and the real-time positions and predicted trajectory lines of high-risk and caution workers are marked with points of different colors.

[0054] Send vibration and voice prompts to the positioning terminal worn by the worker: "Swing area of hoisted objects, please move towards evacuate immediately", wherein the direction is calculated according to the relative orientation between the worker's current position and the nearest boundary of the envelope area.

[0055] Push an alarm card to the APP of the team leader of the team the worker belongs to and the full-time safety officer of the tower crane operation area, the content of the card includes: worker name / ID, current collision probability value, predicted position 3 seconds later, screenshot of the dynamic envelope area, and a "Confirm Disposal" button.

[0056] Meanwhile, to prevent alarm storms, the system is configured to only send duplicate alarms for the same worker on the same hoisting operation if the probability of a previous alarm is not confirmed and is still ≥0.7 after 15 seconds.

[0057] Step S5: APP closed-loop processing and processing data feedback After receiving the alarm via the APP, the team leader or safety officer completes the closed loop according to the following procedure: Confirmation and Attendance: Click the "Confirm Received" button, and the app will automatically record the confirmation time. And start timing the processing time. , This is a real-time variable that is continuously updated during the handling process. The APP calls the phone's GPS, and if the GPS shows that the safety officer's location has entered a 20-meter radius centered on the worker, it is automatically marked as having arrived on site.

[0058] Intervention Records: Safety officers can take photos of on-site interventions, such as when workers have been evacuated to a safe area; enter rectification instructions, such as when workers have been educated and removed from the danger zone; and select the handling result code: 0 - evacuated, 1 - warned, 2 - fine issued.

[0059] Rectification Verification: After the safety officer completes the intervention, they must click "Complete Handling" again. At this point, the system will recalculate the real-time relationship between the worker's current position and the dynamic envelope area by calling the edge nodes. If the shortest distance between the worker's position and the envelope area is greater than 2 meters, and the worker's speed direction is moving away from the envelope area, then it is considered a valid closed loop; otherwise, the safety officer is required to intervene again.

[0060] Reporting timeouts through hierarchical channels: If from The processing time limit has been exceeded If the system fails to complete the process within a specified number of minutes, it will automatically send an alarm message and the current handler's information to the next higher-level administrator's app and send an SMS notification until the process is closed.

[0061] Each closed-loop record includes alarm ID, worker ID, collision probability, and handling time. Whether the loop is effectively closed and the handling result code are all stored in the violation database of the management platform.

[0062] Step S6: Adaptive parameter adjustment of edge nodes based on loop closure rate The management platform regularly compiles daily statistics on the alarm closed-loop data associated with each edge node, i.e., the AI ​​video analytics box corresponding to each tower crane. Node definition. The effective closed-loop rate on the same day for:

[0063] In the formula For nodes The total number of Level 1 and Level 2 alarms triggered on that day; The time for processing those marked as effective closed loops is... Number of minutes (15-minute buffer allowed).

[0064] If for three consecutive days If the platform determines that workers in the area are slow to react to the risks of dynamic suspended objects or that existing early warning strategies are insufficient, it will automatically send alerts to the relevant nodes. Issue parameter adjustment instructions: Predict the duration of the swing. The warning time has been increased from 5 seconds to 8 seconds to expand the lead time. The high-risk alarm threshold was lowered from 0.7 to 0.6 to improve alarm sensitivity; Safety margin The physical fence area has been expanded from 2.0 meters to 3.5 meters.

[0065] Meanwhile, the platform includes tower crane operators and safety officers corresponding to nodes with low closed-loop rates in the special training list for the next day's safety morning meeting, achieving complete closed-loop optimization from algorithm to management.

[0066] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0067] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for visual recognition and dynamic closed-loop management of construction site violations based on edge computing, characterized in that, The described control method includes the following steps: Step S1: In response to recognizing a tower crane lifting operation scenario, collect real-time operation parameters of the tower crane, ambient wind speed data, and positioning data of surrounding workers; Step S2: Based on the collected data, construct a physical model of the suspended load swing, predict the swing angle and swing envelope radius of the suspended load within a predetermined future time period, and further generate a dynamic risk envelope area; Step S3: Predict the movement trajectory of each worker within the predetermined time period according to the current position and velocity vector of the worker, and calculate the collision probability of each worker with the dynamic risk envelope area; Step S4: Generate hierarchical warning information according to the comparison result between the collision probability and a preset threshold, and send the hierarchical warning information to a management platform, so that the management platform pushes an alarm to the corresponding worker terminal and manager terminal; Step S5: In response to confirmation, on-site disposal and rectification feedback operations performed by the manager through the APP, receive disposal feedback information from the management platform to realize a closed-loop management.

2. The method for visual recognition and dynamic closed-loop management of construction site violations based on edge computing as described in claim 1, characterized in that, The generation of the dynamic risk envelope region in step S2 specifically includes: solving the following swing differential equation to predict the angle between the suspension rope and the vertical direction. and its azimuth : In the formula, The angle between the suspension rope and the vertical direction; It is a time variable; This refers to the air damping coefficient; For the weight of the suspended object; This refers to the length of the suspension rope; It is the acceleration due to gravity; The tangential acceleration at the point of suspension of the object; This is the component of the wind load in the oscillating direction; The coordinates of the projected point of the suspended load on the ground are expressed as the coordinates of the tower crane center, the amplitude of the luffing trolley, the length of the hoisting rope, and the predicted swing angle. and azimuth angle The operational relationships are calculated, and the projection points and swing envelope radii for all times within a predetermined future time period are calculated; Take the union of circular areas centered on the projection point at each moment and with the sum of the swing envelope radius and the safety margin as the radius, as the dynamic risk envelope area.

3. The method for visual recognition and dynamic closed-loop management of construction site violations based on edge computing as described in claim 2, characterized in that, The method for calculating the collision probability in step S3 is: for each worker, use a uniform linear motion model to predict their position within the predetermined future time period, and define the collision probability as: In the formula, For the first The probability of collision for each worker; For indicator functions; Predict the location of workers; This is a dynamic risk envelope region; This is a weighting function that decays over time; The current moment; The predetermined time period; the weighting function The value increases as it gets closer to the current moment.

4. The method for visual recognition and dynamic closed-loop management of construction site violations based on edge computing as described in claim 3, characterized in that, The hierarchical warning in step S4 includes: Trigger a first-level alarm when the collision probability is greater than or equal to a first preset threshold; Trigger a second-level alarm when the collision probability is between a second preset threshold and the first preset threshold; Only record a log when the collision probability is less than the second preset threshold and greater than zero; wherein the first preset threshold is greater than the second preset threshold.

5. The method for visual recognition and dynamic closed-loop management of construction site violations based on edge computing as described in claim 4, characterized in that, For workers who trigger the first-level alarm or the second-level alarm, the management platform further executes the following linkage operations: Highlight the dynamic risk envelope area, the real-time position and the predicted trajectory of the worker on a GIS map; Send a voice prompt containing the evacuation direction to the positioning terminal worn by the worker; Push an alarm card containing the worker's identity, collision probability and predicted position to the APP of the team leader of the team to which the worker belongs and the safety officer of the tower crane operation area.

6. The method for visual recognition and dynamic closed-loop management of construction site violations based on edge computing as described in claim 5, characterized in that, The disposal feedback information received in step S5 is generated by the following process: In response to the touch operation of the manager on the confirmation button on the APP, the APP records the confirmation time and starts timing the disposal duration; In response to the on-site intervention photos, rectification descriptions and disposal result codes uploaded by the manager, the APP sends the intervention information to the management platform; The management platform invokes the relationship between the real-time position of the worker and the dynamic risk envelope area to perform rectification verification. When the worker has left the dynamic risk envelope area and the moving direction is away from the area, it is determined as an effective closed loop and a closed-loop record is generated; When the closed-loop record is not received after the disposal duration exceeds the preset disposal time limit, the management platform automatically reports the alarm information step by step to the terminal of the superior manager, and sends the final disposal result as the disposal feedback information to the edge computing node.

7. The method for visual recognition and dynamic closed-loop management of construction site violations based on edge computing as described in claim 6, characterized in that, The method further includes a closed-loop rate statistics and parameter adaptive adjustment step: Edge computing nodes receive parameter adjustment instructions from the management platform and update the predetermined time period, preset threshold, and safety margin parameters of the dynamic risk envelope region stored locally according to the parameter adjustment instructions. The parameter adjustment instructions are generated by the management platform according to the following process: calculating the effective loop closure rate of each edge computing node in multiple consecutive statistical periods; when the effective loop closure rate of an edge computing node is lower than a preset standard, the management platform issues the parameter adjustment instructions to that node. The parameter adjustment instructions include at least one of the following: extending the predetermined time period, reducing the first preset threshold, and increasing the safety margin of the dynamic risk envelope region.

8. The method for visual recognition and dynamic closed-loop management of construction site violations based on edge computing as described in claim 1, characterized in that, The method for edge computing nodes to identify tower crane hoisting operation scenarios in step S1 is as follows: by collecting video streams through high-altitude cameras, and using a deep learning object detection model to detect in multiple consecutive frames that there are suspended objects below the hook and the height of the suspended objects above the ground is greater than a preset height threshold, it is determined that the tower crane hoisting operation is in progress.

9. The method for visual recognition and dynamic closed-loop management of construction site violations based on edge computing as described in claim 1, characterized in that, The real-time operating parameters of the tower crane collected in step S1 include the boom slewing angle, luffing trolley amplitude, hook height, lifting weight, and rope length; the environmental wind speed data includes the average wind speed and the angle between the wind direction and the boom axis; and the worker's positioning data includes the worker's three-dimensional coordinates and movement speed vector.

10. A construction site violation visual recognition and dynamic closed-loop management device based on edge computing, used to execute the method of any one of claims 1 to 9, characterized in that, include: The front-end sensing unit is used to collect video streams, tower crane operating parameters, ambient wind speed and worker positioning data at the construction site, and transmit the collected data to the edge computing node; The edge computing node is communicatively connected to the front-end sensing unit. In response to the identification of tower crane hoisting operation scenario, it constructs a physical model of the hoisting object's swing based on the collected data and solves the swing differential equation. It predicts the swing trajectory of the hoisting object within a predetermined time period to generate a dynamic risk envelope area. At the same time, it predicts the worker's motion trajectory based on the worker's current position and velocity vector, calculates the collision probability of each worker with the dynamic risk envelope area, and uploads the collision probability and related data to the management platform. The management platform is connected to the edge computing nodes and is used to generate graded early warning information based on the comparison result of the collision probability and the preset threshold, push alarms to the corresponding worker terminals and management personnel terminals, and count the closed loop rate of each edge computing node. Based on the result that the closed loop rate is lower than the preset standard, the predetermined time period, the preset threshold and the safety margin parameters of the dynamic risk envelope area are automatically adjusted. The APP terminal communicates with the management platform and is used by managers to confirm alarms, handle on-site and provide feedback on rectification, record handling time and upload rectification verification results to form a management closed loop.

Images