Construction worker safety behavior real-time analysis method and system based on edge computing

Abstract

The present application relates to the technical field of safety behavior analysis, in particular to a construction worker safety behavior real-time analysis method and system based on edge computing, which generates a person-rope association mapping table according to a person identifier and a rope identifier; when two or more safety rope objects are determined to have a cross-shading event according to object data of a monitoring video stream, a high-risk event signal is generated, and a control instruction is sent to a target camera unit to obtain a supplementary perspective video sequence of the safety rope objects that have the cross-shading event; a determination result is obtained by analyzing the supplementary perspective video sequence, and when the determination result is a physical cross, the safety rope object is marked with a physical cross risk identifier in the person-rope association mapping table; when any safety rope object associated with a construction worker object in the person-rope association mapping table is marked with a physical cross risk identifier, a safety state abnormal signal is output; the nature of the safety rope cross-shading event can be accurately distinguished, and potential safety risks can be discovered in a timely manner.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of safety behavior analysis, in particular to a construction worker safety behavior real-time analysis method and system based on edge computing. BACKGROUND

[0002] In the construction process, the safety rope is an important safety protection equipment for construction workers in dangerous environments such as high-altitude work, and its correct use and state are directly related to the safety of the construction workers. The real-time analysis technology of edge computing has the characteristics of low delay and high efficiency, which can quickly process the massive data generated on the construction site and timely discover potential safety hazards.

[0003] However, in the actual construction site environment, the use of safety ropes is complex and variable, and the cross-shading events between safety ropes occur frequently. When the safety ropes cross-shade, it is difficult to accurately determine whether the cross is a real physical entanglement (physical cross) or a visual overlap caused by a visual angle problem (non-physical cross). If a non-physical cross is misjudged as a physical cross, unnecessary alarms and construction interruptions may be triggered, affecting construction progress and efficiency; if a physical cross is missed, potential safety risks cannot be discovered in time, which may lead to serious safety accidents and threaten the safety of construction workers.

[0004] Therefore, there is an urgent need for a construction worker safety behavior real-time analysis method based on edge computing, which can accurately distinguish the nature of safety rope cross-shading events, timely discover potential safety risks, and effectively improve construction safety. SUMMARY

[0005] (1) Technical problem to be solved

[0006] The purpose of the present application is to provide a construction worker safety behavior real-time analysis method and system based on edge computing to solve the problem that in dangerous environments such as high-altitude work, the nature of safety rope cross-shading events cannot be accurately distinguished, resulting in the inability to timely discover potential safety risks.

[0007] (2) Technical solution

[0008] To achieve the above purpose, on the one hand, the present application provides a construction worker safety behavior real-time analysis method based on edge computing, which comprises:

[0009] S1, continuously acquiring monitoring video streams through image acquisition devices arranged in the construction area and inputting edge computing nodes, assigning a unique personnel identifier to each construction worker object in the monitoring video stream and a unique rope identifier to each safety rope object, and generating a person-rope association mapping table according to the correspondence between the personnel identifier and the rope identifier.

[0010] S2, the edge computing node processes object data corresponding to the monitoring video stream in real time, and when it is judged according to the object data that two or more safety rope objects have a cross-shielding event, a high-risk event signal is generated.

[0011] S3, when the high-risk event signal is monitored, the edge computing node generates a control instruction and sends the control instruction to a target camera unit in the image acquisition device; the control instruction is used to drive the target camera unit to adjust the shooting parameter to obtain a supplementary perspective video sequence of the safety rope object having the cross-shielding event.

[0012] S4, the edge computing node receives the supplementary perspective video sequence and performs data analysis to obtain a judgment result, and when the judgment result is that the cross-shielding event is a physical cross, the rope identifier corresponding to the safety rope object participating in the physical cross is determined, and the physical cross risk identifier is marked for the corresponding safety rope object in the person-rope association mapping table.

[0013] S5, when it is detected that any construction personnel object associated with the safety rope object in the person-rope association mapping table is marked with the physical cross risk identifier, a safety state abnormal signal is generated and output.

[0014] Further, when the high-risk event signal is monitored, the edge computing node generates a control instruction and sends the control instruction to a target camera unit in the image acquisition device, the method comprising:

[0015] extracting first spatial position data of the safety rope object having the cross-shielding event in the construction space from the cross-shielding event data of the high-risk event signal; obtaining deployment information of all image acquisition devices managed by the edge computing node, the deployment information including spatial position data and current perspective coverage range of the camera unit in the image acquisition device.

[0016] According to the first spatial position data, the spatial position data and the current perspective coverage range of the camera unit, a candidate camera unit is selected from all camera units, which is not completely covered by the current perspective coverage range or has a perspective shielding.

[0017] According to the spatial relationship between the spatial position data of the candidate camera unit and the first spatial position data, a target camera unit is selected, and a camera unit perspective adjustment parameter is calculated for the target camera unit.

[0018] The edge computing node generates a control instruction containing the camera unit perspective adjustment parameter, and sends the control instruction to the target camera unit.

[0019] Further, the method of screening out candidate camera units from all camera units, whose current visual angle coverage range does not completely cover the first spatial position data or there is visual angle obstruction, comprises:

[0020] According to the spatial position data of each camera unit, the current visual angle coverage range parameter and the first spatial position data, it is calculated whether the first spatial position data is located in the stereoscopic spatial region formed by the current visual angle coverage range of the camera unit.

[0021] When the first spatial position data is located in the stereoscopic spatial region, it is judged according to the pre-constructed construction scene three-dimensional model whether there is an obstruction on the line-of-sight path from the spatial position of the camera unit to the spatial point indicated by the first spatial position data, and if there is an obstruction, it is determined that the camera unit has visual angle obstruction to the first spatial position data.

[0022] When the first spatial position data is not located in the stereoscopic spatial region, or is located in the stereoscopic spatial region but there is visual angle obstruction, the camera unit is recorded as a candidate camera unit.

[0023] Further, the method of judging according to the pre-constructed construction scene three-dimensional model whether there is an obstruction on the line-of-sight path from the spatial position of the camera unit to the spatial point indicated by the first spatial position data, comprises:

[0024] The pre-constructed construction scene three-dimensional model is obtained, and the construction scene three-dimensional model contains geometric data and spatial position information of fixed structures and equipment in the construction scene.

[0025] According to the spatial position data of the camera unit and the first spatial position data, a line-of-sight vector connecting the two points is calculated; line-of-sight projection analysis is performed along the line-of-sight vector in the construction scene three-dimensional model, it is detected whether the path passed by the line-of-sight vector intersects with the three-dimensional entity represented by the geometric data of any fixed structure or equipment in the construction scene three-dimensional model; if intersection is detected, it is determined that there is an obstruction on the line-of-sight path; if no intersection is detected, it is determined that there is no obstruction on the line-of-sight path.

[0026] Further, the method of selecting a target camera unit according to the spatial relationship between the spatial position data of the candidate camera unit and the first spatial position data, comprises:

[0027] According to the mechanical motion parameters of the candidate camera unit, the estimated adjustment time required for the visual angle center axis to adjust from the current direction to point to the spatial point corresponding to the first spatial position data is calculated, and the mechanical motion parameters include horizontal rotation angular velocity and pitch rotation angular velocity.

[0028] According to a spatial geometric relationship between the spatial position data of the candidate camera unit and the first spatial position data, an expected image quality of imaging the target region at the adjusted view angle is evaluated, the spatial geometric relationship including a relative distance and a relative angle.

[0029] A comprehensive selection index of the candidate camera unit is obtained through data analysis of the estimated adjustment time and the expected image quality, the comprehensive selection index being set to preferentially select the candidate camera unit with shorter estimated adjustment time on the premise of meeting a preset minimum expected image quality requirement; all candidate camera units are sorted according to the comprehensive selection index, and the candidate camera unit with the optimal comprehensive selection index in the sorting result is selected as the target camera unit.

[0030] Further, the method of evaluating the expected image quality of imaging the target region at the adjusted view angle according to the spatial geometric relationship between the spatial position data of the candidate camera unit and the first spatial position data includes:

[0031] A straight-line distance from an optical center point of the candidate camera unit to a center point of the target region is calculated as the relative distance, and an included angle between a current optical axis direction of the candidate camera unit and a vector pointing from the candidate camera unit to the center point of the target region is calculated as the relative angle.

[0032] According to the relative distance and a focal length parameter of the candidate camera unit, a pixel coverage number of the target region on an imaging plane is calculated through an imaging model to evaluate an imaging resolution; a perspective distortion coefficient is obtained according to a tilt angle component and a horizontal deflection angle component in the relative angle, the perspective distortion coefficient representing a degree of tilting of the target region relative to the imaging plane; and the expected image quality is obtained through a weighted fusion algorithm according to the imaging resolution and the perspective distortion coefficient.

[0033] Further, the method of calculating the camera unit view angle adjustment parameter for the target camera unit includes:

[0034] A three-dimensional coordinate of spatial position data of the target camera unit and a current optical axis direction vector are obtained; a three-dimensional coordinate of a target space point corresponding to the first spatial position data is obtained; and a target direction vector pointing from the spatial position of the target camera unit to the target space point is calculated.

[0035] According to the current optical axis direction vector of the target camera unit and the calculated target direction vector, a horizontal deflection angle of rotation around a vertical axis and a tilt angle of rotation around a horizontal axis required for the current optical axis direction vector to coincide with the target direction vector are calculated.

[0036] According to a three-dimensional space distance between the spatial position of the target camera unit and the target space point, in combination with a lens optical parameter of the target camera unit, a focal length adjustment amount required for clear imaging of the target space point is calculated; and the horizontal deflection angle, the pitch angle and the focal length adjustment amount jointly constitute the camera unit view angle adjustment parameter.

[0037] Further, the method for receiving the supplementary view video sequence and performing data analysis to obtain a determination result by the edge computing node comprises:

[0038] The edge computing node parses the supplementary view video sequence to perform object contour segmentation on an image region where the crossing-shielded safety rope object is located to obtain a contour topological structure of the safety rope object near the intersection point.

[0039] The connection relationship of the contour topological structure at the intersection point is analyzed, if the contour topological structure shows that there are independent and mutually disconnected safety rope object contours at the intersection point, it is determined that the crossing-shielded event is non-physical crossing; if the contour topological structure shows that there are shared contours or contour connection points at the intersection point, so that the contours of different safety rope objects cannot be clearly segmented into independent objects, it is determined that the crossing-shielded event is physical crossing.

[0040] Further, the method for performing object contour segmentation on an image region where the crossing-shielded safety rope object is located to obtain a contour topological structure of the safety rope object near the intersection point comprises:

[0041] An image frame containing the intersection point region is extracted from the supplementary view video sequence and image preprocessing is performed to enhance the contour features, and an edge detection algorithm is used to identify all potential contour lines in the image frame; according to the morphological prior knowledge of the safety rope object, candidate contour lines belonging to the safety rope object are selected from all potential contour lines, and the morphological prior knowledge includes a length threshold of the contour, a height-width ratio range and linearity.

[0042] The candidate contour lines are subjected to connectivity analysis to construct a topological connection relationship between the contours, if there is a distance between the end points of the candidate contour lines less than a preset connectivity threshold and the direction continuity meets a preset condition, the candidate contour lines are connected as continuous contours of the same safety rope object; and the continuous contours and the topological connection relationship between the contours jointly represent the contour topological structure.

[0043] On the other hand, based on the same inventive concept, the present application also provides an edge computing-based real-time analysis system for safety behaviors of construction personnel, which comprises an identifier allocation and person-rope mapping module, a high-risk event judgment module, a camera unit control module, a determination result analysis and marking module and a safety state abnormal signal generation module, and the modules are sequentially and communicatively connected.

[0044] An identifier allocation and person-rope mapping module is configured to continuously acquire a monitoring video stream by an image acquisition device laid in a construction area and input the monitoring video stream into an edge computing node, allocate a unique person identifier to each construction worker object in the monitoring video stream and allocate a unique rope identifier to each safety rope object, and generate a person-rope association mapping table according to the correspondence between the person identifier and the rope identifier.

[0045] A high-risk event judgment module is configured to process object data corresponding to the monitoring video stream in real time by the edge computing node, and generate a high-risk event signal when it is judged according to the object data that two or more safety rope objects have a crossing shielding event.

[0046] A camera unit control module is configured to generate a control instruction by the edge computing node when a high-risk event signal is monitored, and send the control instruction to a target camera unit in the image acquisition device; the control instruction is used to drive the target camera unit to adjust a shooting parameter to acquire a supplementary perspective video sequence of the safety rope object having the crossing shielding event.

[0047] A judgment result analysis and identification module is configured to receive the supplementary perspective video sequence by the edge computing node and perform data analysis to obtain a judgment result, and determine a rope identifier corresponding to the safety rope object participating in the physical crossing and mark a physical crossing risk identifier for the corresponding safety rope object in the person-rope association mapping table when the judgment result is that the crossing shielding event is a physical crossing.

[0048] A safety state abnormal signal generation module is configured to generate and output a safety state abnormal signal when any construction worker object associated with a safety rope object in the person-rope association mapping table is marked with the physical crossing risk identifier.

[0049] (3) Advantageous effects

[0050] Compared with the prior art, the advantageous effects of the present application are:

[0051] 1. By allocating unique identifiers to construction workers and safety ropes and generating a person-rope association mapping table, in combination with real-time processing of the monitoring video stream by the edge computing node, the crossing shielding event of the safety rope can be quickly judged and a high-risk event signal can be generated, the camera unit can be adjusted in time to acquire a supplementary perspective video sequence, and the nature of the crossing shielding event can be accurately determined, thereby effectively improving the monitoring accuracy and timeliness of the safety state of the construction workers and providing strong support for construction safety protection.

[0052] 2. When a high-risk event occurs, based on detailed spatial location data, construction scene three-dimensional model, and mechanical motion parameters of the camera unit, candidate camera units are screened from multiple dimensions, and the target camera unit is selected by comprehensively considering factors such as estimated adjustment time and expected image quality, and the camera unit view angle adjustment parameters are accurately calculated to ensure that clear and effective supplementary view angle videos can be obtained in the optimal way, and the reliability and effectiveness of safety monitoring are further improved.

[0053] 3. After receiving the supplementary view angle video sequence, the safety rope object near the intersection is finely segmented and the contour topological structure is analyzed by parsing the video, the contour features are accurately identified by using the morphological prior knowledge of the safety rope object, and whether the intersection occlusion event is a physical intersection is accurately determined according to the connection relationship of the contour topological structure at the intersection, thereby providing a reliable basis for subsequent safety state abnormal signal generation. BRIEF DESCRIPTION OF DRAWINGS

[0054] Figure 1 The flowchart of the construction personnel safety behavior real-time analysis method based on edge computing of the embodiment 1 of the present application.

[0055] Figure 2 The module composition schematic diagram of the construction personnel safety behavior real-time analysis system based on edge computing of the embodiment 2 of the present application. DETAILED DESCRIPTION

[0056] The technical solutions in the embodiments of the present application will be described clearly and completely in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the present application.

[0057] Before examples are given, the application scenario of the inventive concept needs to be described. The present application is applied to actual construction. When multiple workers simultaneously perform high-altitude operations, their safety ropes often appear to cross in the air. At this time, it needs to be determined whether the crossing is real safety rope entanglement or just looks overlapped but actually one precedes the other without contact. If the physical crossing of the safety rope is not determined in time, safety risks will be caused.

[0058] Embodiment 1: As shown in the figure, the present embodiment provides a construction personnel safety behavior real-time analysis method based on edge computing, which comprises: Figure 1

[0059] ​S1, the image acquisition equipment laid through the construction area continuously acquires the monitoring video stream and inputs the edge computing node, each construction personnel object in the monitoring video stream is allocated a unique personnel identifier and each safety rope object is allocated a unique rope identifier; a person-rope association mapping table is generated according to the correspondence of the personnel identifier and the rope identifier; a plurality of high-definition cameras are deployed at key positions in the construction site, and the cameras are connected to the edge computing node through wired or wireless network to form a complete monitoring network. After the edge computing node receives the continuous monitoring video stream, computer vision technology is used to identify and classify the target objects in the picture. Each construction worker in the picture is automatically identified and assigned a unique digital number as a personnel identifier, such as "Worker001", "Worker002", etc. At the same time, each safety rope is identified and assigned a corresponding rope identifier, such as "Rope001", "Rope002", etc. The allocation process of such identifiers uses a deep learning object detection algorithm to accurately distinguish different construction personnel and safety rope objects in a complex construction environment. After completing the object identification, a person-rope association mapping table is established to record the correspondence between each construction worker and the safety rope he uses. For example, when "Worker001" is detected using "Rope001", the association record "Worker001→Rope001" is established in the person-rope mapping table.

[0060] S2, the edge computing node processes the object data corresponding to the monitoring video stream in real time, and when it is judged according to the object data that two or more safety rope objects have a cross-shading event, a high-risk event signal is generated; the edge computing node continuously analyzes the object data in the monitoring video stream, and focuses on monitoring the spatial relationship changes between safety rope objects. When two or more safety ropes overlap and cross in the picture, the cross-shading event detection mechanism will identify the various cross modes that may occur between safety rope objects. Once a cross-shading event is detected, a high-risk event signal is immediately generated, containing the specific position coordinates of the cross point, the safety rope identifiers involved, and other key information.

[0061] S3, when the high-risk event signal is monitored, the edge computing node generates a control instruction and sends the control instruction to the target camera unit in the image acquisition equipment; the control instruction is used to drive the target camera unit to adjust the shooting parameters to obtain a supplementary angle video sequence of the safety rope object that has a cross-shading event.

[0062] S4, the edge computing node receives the supplementary view video sequence and performs data analysis to obtain a determination result, when the determination result is that the intersection occlusion event is a physical intersection, then the rope identifier corresponding to the safety rope object participating in the physical intersection is determined, and the physical intersection risk identifier is marked for the corresponding safety rope object in the person-rope association mapping table.

[0063] S5, when any construction personnel object associated with the safety rope object in the person-rope association mapping table is marked with the physical intersection risk identifier, a safety state abnormal signal is generated and output. The state of the person-rope mapping table is continuously monitored, and as soon as it is found that any construction personnel associated safety rope is marked with the physical intersection risk identifier, a safety state abnormal signal is immediately generated and output. The abnormal signal is presented in various forms such as audible and visual alarm, mobile terminal push, monitoring center display, etc., to ensure that relevant personnel can receive warning information in time and take appropriate safety measures.

[0064] When the high-risk event signal is monitored, the edge computing node generates a control instruction and sends the control instruction to the target camera unit in the image acquisition device.

[0065] The first spatial position data of the safety rope object intersecting and occluding in the construction space is extracted from the intersection occlusion event data of the high-risk event signal; the deployment information of all image acquisition devices managed by the edge computing node is obtained, and the deployment information includes spatial position data and current view coverage range of the camera unit in the image acquisition device; when the edge computing node receives the high-risk event signal, the precise position information of the safety rope object intersecting in the three-dimensional space is parsed from the intersection occlusion event data included in the signal, which is the first spatial position data. The intersection point coordinates in the two-dimensional image are converted into specific positions in the three-dimensional coordinate system through multi-view geometric calculation and depth estimation algorithm, including X, Y, Z three direction coordinate values. For example, when two safety ropes intersect at a spatial position 15 meters away from the ground, 8 meters away from the east wall of the building, and 12 meters away from the south wall, the system will record this precise three-dimensional coordinate as the first spatial position data. The detailed deployment information of all image acquisition devices within the management range of the edge computing node is obtained, which includes spatial position data and current view coverage range of each camera unit. The spatial position data includes installation position three-dimensional coordinates, and the current view coverage range includes current shooting direction, field of view angle range, effective shooting distance and other parameters. Through these information, the three-dimensional space area covered by the current view of each camera unit can be calculated.

[0066] According to the first spatial position data and the spatial position data and the current visual angle coverage range of the camera unit, candidate camera units are screened out from all camera units, which do not completely cover the first spatial position data or have visual angle obstruction in the current visual angle coverage range.

[0067] According to the spatial relationship between the spatial position data of the candidate camera unit and the first spatial position data, a target camera unit is selected, and a camera unit visual angle adjustment parameter is calculated for the target camera unit.

[0068] The edge computing node generates control instructions containing the camera unit visual angle adjustment parameter, and sends the control instructions to the target camera unit. The edge computing node encapsulates the precisely calculated adjustment parameter into standardized control instructions, and sends the control instructions to the target camera unit through a communication network, so that the camera unit can accurately perform the adjustment action and obtain a high-quality supplementary visual angle video sequence.

[0069] The method of screening out candidate camera units from all camera units, which do not completely cover the first spatial position data or have visual angle obstruction in the current visual angle coverage range, comprises:

[0070] According to the spatial position data of each camera unit, the current visual angle coverage range parameter and the first spatial position data, it is calculated whether the first spatial position data is located in a three-dimensional space region formed by the current visual angle coverage range of the camera unit; the installation position three-dimensional coordinates, the current optical axis direction vector, the field of view angle parameter and other basic data of the camera unit are obtained and a three-dimensional visual angle coverage region model is constructed, which is usually represented as a conical or truncated conical space region with the camera unit position as the vertex. The boundary of this region is determined by the horizontal field of view angle, the vertical field of view angle, the nearest focusing distance and the farthest effective shooting distance of the camera unit. For example, a camera unit installed in a corner of a building with a height of 20 meters, a horizontal field of view angle of 60 degrees and a vertical field of view angle of 45 degrees, its visual angle coverage region is a conical space in a specific direction and angle.

[0071] When the first spatial position data is located in the three-dimensional space region, it is judged according to the pre-constructed three-dimensional model of the construction scene whether there is an obstruction on the line-of-sight path from the spatial position of the camera unit to the spatial point indicated by the first spatial position data. If there is an obstruction, it is determined that the camera unit has visual angle obstruction to the first spatial position data.

[0072] When the first spatial position data is not located in the stereoscopic space region, or is located in the stereoscopic space region but there is a visual angle obstruction, the camera unit is marked as a candidate camera unit. When it is necessary to determine whether the first spatial position data is in the visual angle coverage range of a certain camera unit, accurate point and conical region position relationship calculation is performed, and the position relationship calculation process adopts three-dimensional geometric operation, including steps of vector dot product calculation, angle comparison, distance verification, etc. If the intersection point represented by the first spatial position data is located outside the boundary of the conical region, it indicates that the current visual angle cannot completely cover the target region, and the camera unit is marked as a candidate unit. A more complex case is that the intersection point is located in the visual angle coverage range, but there is a judgment of visual line obstruction. The system calls the pre-constructed three-dimensional model of the construction scene to perform obstruction analysis, and through the accurate spatial geometric analysis and obstruction analysis, all camera units that need to adjust the visual angle can be accurately identified, thereby providing reliable basic data for subsequent target camera unit selection and visual angle adjustment.

[0073] The method for judging whether there is an obstruction on the visual line path from the spatial position of the camera unit to the spatial point indicated by the first spatial position data according to the pre-constructed three-dimensional model of the construction scene comprises the following steps:

[0074] The pre-constructed three-dimensional model of the construction scene is acquired, and the three-dimensional model of the construction scene comprises geometric data and spatial position information of fixed structures and equipment in the construction scene; the three-dimensional model comprises detailed geometric information of all fixed structures in the construction site, such as building main structure, scaffold system, tower crane equipment, temporary building, large construction machinery, etc. Each structure exists in the form of an accurate three-dimensional geometric body in the model, and comprises accurate spatial position, size parameter and shape feature. For example, when a camera unit located on the north side of a building needs to observe a safety rope intersection point on the south side, the system detects whether the visual line path from the camera unit to the intersection point is blocked by the intermediate scaffold, beam column or other equipment. If it is detected that the visual line path intersects with the steel pipe structure of the scaffold, it is determined that there is a visual angle obstruction, and the camera unit is marked as a candidate camera unit. Technical personnel use laser scanning equipment, unmanned aerial vehicle aerial photography, photogrammetry and other technical means to perform all-around three-dimensional data acquisition on the construction site, and the three-dimensional data is processed to form an accurate three-dimensional model comprising all fixed structures and main equipment. Each object in the three-dimensional model comprises detailed geometric data, such as wall thickness of a building, cross-sectional size of a beam column, steel pipe diameter and connection mode of a scaffold, arm length and turning radius of a tower crane, etc. Meanwhile, the three-dimensional model also records accurate spatial position information of each structure, establishes a unified coordinate system, and ensures that the relative position relationship of all objects in space is accurate and correct.

[0075] According to the spatial position data of the camera unit and the first spatial position data, a line-of-sight vector connecting the two points is calculated; a line-of-sight projection analysis is performed along the line-of-sight vector in the construction scene three-dimensional model, and it is detected whether the path passed by the line-of-sight vector intersects with a three-dimensional entity represented by the geometric data of any fixed structure or equipment in the construction scene three-dimensional model; if intersection is detected, it is determined that there is an occlusion on the line-of-sight path; if no intersection is detected, it is determined that there is no occlusion on the line-of-sight path. The calculation of the line-of-sight vector is based on the basic geometric principle of determining a straight line between two points in a three-dimensional space. The three-dimensional coordinate point A of the camera unit and the target point B corresponding to the first spatial position data are obtained, and then the direction vector from A to B is calculated. For example, if the camera unit is located at coordinates (x1, y1, z1) and the intersection point is located at coordinates (x2, y2, z2), then the line-of-sight vector can be represented as the direction vector of (x2-x1, y2-y1, z2-z1). Along the calculated line-of-sight vector, step-by-step sampling detection is performed at a certain step size, and at each sampling point, the system determines whether the point is located inside or on the boundary of any structure in the three-dimensional model. Specific intersection detection algorithms use different mathematical methods for different types of geometric bodies. For cylindrical structures (such as scaffold steel pipes), the intersection is determined by comparing the point-to-axis distance with the radius; for complex irregular geometric bodies, the intersection point calculation method of the ray and the triangular patch is used.

[0076] The method for selecting a target camera unit according to the spatial relationship between the spatial position data of the candidate camera unit and the first spatial position data comprises:

[0077] According to the mechanical motion parameters of the candidate camera unit, the estimated adjustment time required for adjusting the center axis of the view angle from the current direction to point to the spatial point corresponding to the first spatial position data is calculated, and the mechanical motion parameters include the horizontal rotation angular velocity and the pitch rotation angular velocity. The calculation of the estimated adjustment time is based on the motion characteristics of the mechanical structure of the camera unit. Each camera unit is equipped with a precise motor drive system and can realize accurate rotation in the horizontal direction (azimuth angle) and the vertical direction (pitch angle). The system calculates the angular difference between the current optical axis direction of the camera unit and the target direction, and then estimates the adjustment time required according to the mechanical motion parameters of the camera unit. For example, if the horizontal rotation angular velocity of a certain camera unit is 5 degrees per second and the pitch rotation angular velocity is 3 degrees per second, and the current horizontal rotation is 30 degrees and the pitch rotation is 15 degrees to point to the intersection point, then the adjustment time needs to consider the time of motion in both directions, and the larger value is usually taken, i.e. max(30÷5, 15÷3) = 6 seconds.

[0078] According to the spatial geometric relationship between the spatial position data of the candidate camera unit and the first spatial position data, the expected image quality of the target region imaged under the adjusted viewing angle is evaluated, and the spatial geometric relationship includes relative distance and relative angle. The relative distance is an important factor affecting the image quality. Too close distance will result in insufficient field of view range, and too far distance will affect the detail resolution. The relative angle is also critical. If the observation angle between the camera unit and the intersection point is too inclined, serious perspective distortion will occur, affecting the accurate judgment of the safety rope intersection state. For example, consider the case of two candidate camera units: camera unit A is 12 meters away from the intersection point, and the observation angle is 30 degrees inclined; camera unit B is 8 meters away from the intersection point, but the observation angle is 60 degrees inclined. Although camera unit B is closer, it can theoretically obtain higher resolution, but due to the observation angle being too inclined, it may cause the safety rope to be severely perspective compressed in the picture, which is not conducive to the judgment of the intersection state. While camera unit A is slightly farther away, but the observation angle is relatively appropriate, which can provide clearer and more accurate intersection point images.

[0079] The estimated adjustment time and the expected image quality are analyzed by data to obtain a comprehensive selection index of the candidate camera unit, and the comprehensive selection index is set to preferentially select the candidate camera unit with shorter estimated adjustment time under the premise of meeting the preset minimum expected image quality requirement; all candidate camera units are sorted according to the comprehensive selection index, and the candidate camera unit with the optimal comprehensive selection index in the sorting result is selected as the target camera unit. The system will set a minimum expected image quality threshold, and only the candidate camera unit meeting this threshold requirement will enter the final selection process. For the candidate camera unit meeting the expected image quality requirement, the camera unit with shorter adjustment time is preferentially selected to ensure that the supplementary viewing angle can be quickly obtained and the safety state judgment can be completed in time. This reflects the strict requirement for response speed in the actual construction environment, because the handling of safety incidents often has timeliness, and delayed judgment may lead to the expansion of safety risks. Both the acquisition quality of the supplementary viewing angle and the response time are optimized.

[0080] The method of evaluating the expected image quality of the target region imaged under the adjusted viewing angle according to the spatial geometric relationship between the spatial position data of the candidate camera unit and the first spatial position data includes:

[0081] The straight-line distance from the optical center point of the candidate camera unit to the center point of the target region is calculated as the relative distance according to the spatial position data of the candidate camera unit and the first spatial position data, and the included angle between the current optical axis direction of the candidate camera unit and the vector pointing from the candidate camera unit to the center point of the target region is calculated as the relative angle; the optical center point coordinate of the candidate camera unit is obtained, which is usually located at the principal point position of the camera lens. The straight-line distance from the optical center point to the center point of the target region (i.e. the intersection position) is the relative distance. The included angle between the current optical axis direction vector and the direction vector from the camera unit to the target region is the relative angle, which can be decomposed into a horizontal deflection angle and a pitch angle.

[0082] According to the relative distance and the focal length parameter of the candidate camera unit, the number of pixel coverage of the target region on the imaging plane is calculated through an imaging model to evaluate the imaging resolution; the degree of tilt of the target region relative to the imaging plane is calculated according to the pitch angle component and the horizontal deflection angle component in the relative angle to obtain a perspective distortion coefficient; and the expected image quality is obtained through a weighted fusion algorithm according to the imaging resolution and the perspective distortion coefficient. According to the focal length parameter of the camera unit and the relative distance, the projection size of the target region on the imaging plane is calculated using the principle of similar triangles. If the size of the target region in the actual space is 2 meters x 2 meters, the focal length of the camera unit is 50 millimeters, and the relative distance is 10 meters, then according to the imaging formula, the projection size of the target region on the imaging plane is about 10 millimeters x 10 millimeters. Combined with the pixel density parameter of the camera unit, it can be calculated how many pixels the target region will occupy, so as to evaluate the resolution level of the imaging. The calculation of the perspective distortion coefficient takes into account the influence of the observation angle on the image geometry. When the camera unit observes the target region at an inclined angle, it will produce a perspective compression or stretching effect, affecting the accurate judgment of the safety rope intersection state. The system analyzes the pitch angle component and the horizontal deflection angle component in the relative angle, and calculates the degree of distortion caused by these angle deviations to the geometry of the target region. The pitch angle deviation mainly affects the perspective distortion in the vertical direction, and the horizontal deflection angle deviation mainly affects the perspective distortion in the horizontal direction. The specific distortion coefficient calculation uses the perspective transformation matrix in projective geometry. The weighted fusion algorithm of the expected image quality needs to reasonably balance the two indicators of resolution and distortion degree. Usually, the resolution indicator reflects the degree of detail and clarity of the image, and the perspective distortion coefficient reflects the geometric accuracy of the image. For the judgment of the safety rope intersection state, geometric accuracy is often more important than extremely high resolution, because slight distortion may lead to misjudgment of the intersection property. Using weighted averaging or more complex fusion algorithms to calculate the final image quality score can accurately predict the imaging effect of different camera units after adjustment, providing a reliable technical basis for the selection of the target camera unit.

[0083] The method for calculating the camera unit perspective adjustment parameter for the target camera unit comprises:

[0084] The three-dimensional coordinates of the spatial position data of the target camera unit and the current optical axis direction vector are obtained; the three-dimensional coordinates of the target spatial point corresponding to the first spatial position data are obtained; the target direction vector from the spatial position of the target camera unit to the target spatial point is calculated; the current optical axis direction vector reflects the shooting direction of the camera unit at this time, which can be obtained in real time through the encoder or sensor built in the camera unit, or can be calculated according to the current horizontal angle and the pitch angle. For example, if the current horizontal angle of the camera unit is 120 degrees (0 degrees in the positive east direction), and the pitch angle is -15 degrees (a negative value indicates downward inclination), then the optical axis direction vector can be represented as a unit vector in three-dimensional space. The calculation of the target direction vector is a standard spatial geometry problem, which will calculate the connecting vector between the position coordinates of the camera unit and the coordinates of the target spatial point (intersection point), and after normalization processing, the ideal target direction vector is obtained.

[0085] According to the current optical axis direction vector of the target camera unit and the calculated target direction vector, the horizontal deflection angle required for rotating the current optical axis direction vector around the vertical axis to coincide with the target direction vector and the pitch angle required for rotating around the horizontal axis are calculated; the rotation transformation required for converting the current optical axis direction vector to the target direction vector is usually decomposed by Euler angle or quaternion method. For the horizontal deflection angle, the two direction vectors are projected onto the horizontal plane, and the included angle between the projection vectors is calculated; for the pitch angle, the angle change in the vertical direction is considered. By decomposing the current direction vector and the target direction vector into horizontal and vertical components, the horizontal component reflects the projection on the XY plane, and the vertical component is the value of the Z-axis direction. The horizontal deflection angle can be calculated by the inverse tangent function, and the calculation of the pitch angle needs to consider the elevation angle of the vector.

[0086] According to the three-dimensional spatial distance between the spatial position of the target camera unit and the target spatial point, combined with the lens optical parameters of the target camera unit, the focal length adjustment amount required for clear imaging of the target spatial point is calculated; the horizontal deflection angle, the pitch angle and the focal length adjustment amount jointly constitute the camera unit view angle adjustment parameters. According to the distance between the camera unit and the target spatial point, combined with the optical parameters of the lens, the focal length setting required to obtain the best imaging effect is calculated, which needs to consider multiple factors such as the focal length range of the lens, the closest focusing distance, the depth of field requirement, etc. For the fixed focal length lens, mainly calculate the adjustment of the focusing distance; for the zoom lens, the adjustment of the focal length also needs to be calculated to optimize the field of view range. For example, when the target distance is 12 meters, and the current focusing distance is set to infinity, it is calculated that the focusing distance needs to be adjusted to about 12 meters. The calculated horizontal deflection angle, pitch angle, focal length adjustment amount and other parameters are integrated into a standardized control instruction format to ensure that the camera unit can accurately perform the adjustment action.

[0087] The method for the edge computing node to receive the supplementary view video sequence and obtain a decision result through data analysis comprises:

[0088] The edge computing node analyzes the supplementary view video sequence to perform object contour segmentation on the image area where the safety rope object occurs cross occlusion to obtain the contour topological structure of the safety rope object near the cross point.

[0089] The connection relationships of the contour topology at the intersection points are analyzed. If the contour topology shows that there are independent, unconnected safety rope object contours at the intersection points, the intersection occlusion event is determined to be a non-physical intersection. If the contour topology shows that there are shared contours or contour connection points at the intersection points, making it impossible to clearly separate the contours of different safety rope objects into independent objects, the intersection occlusion event is determined to be a physical intersection. After the edge computing node receives the supplementary view video sequence transmitted by the target camera unit, it selects several frames with the best image quality from the supplementary view video sequence as the analysis objects. Usually, the frames with the clearest intersection points and the most complete safety rope contours are selected for focused analysis. If it is a non-physical intersection (i.e., visual overlap but no actual contact), then under the supplementary view, the contours of the two safety ropes should be independent of each other. Even if they appear to overlap visually, they should be clearly separated into two independent continuous curves in the contour analysis. Conversely, if it is a physical intersection (i.e., actual rope entanglement), then contour connections or fusion phenomena will occur at the intersection points. For example, two originally independent contour lines merge into a connected region at the intersection point, or characteristic connection structures such as Y-shapes or X-shapes appear. The system determines the nature of the connection relationship by analyzing indicators such as the number of connected components, branch point features, and local geometry of the contours. If it is merely visual overlap, then under a suitable supplementary viewpoint, the system should be able to identify two complete and independent contour lines, each with a clear start and end point, and no physical connection to the other contour line in between. However, if actual entanglement occurs, then near the intersection point, the two contour lines will exhibit actual geometric connection, forming a more complex connected structure that the system cannot clearly separate into two independent contours.

[0090] The method for obtaining the contour topology of the safety rope object near the intersection point by performing object contour segmentation on the image region where the safety rope object is located during the intersection point includes:

[0091] The image frames containing intersection regions are extracted from the supplementary view video sequence and image pre-processing is performed to enhance the contour features, and an edge detection algorithm is used to identify all potential contour lines in the image frames; candidate contour lines belonging to the safety rope object are selected from all potential contour lines according to the morphological prior knowledge of the safety rope object, and the morphological prior knowledge includes a length threshold of the contour, an aspect ratio range, and linearity; the clearest key frames are selected from the supplementary view video sequence, and usually 3-5 frames are selected for comprehensive analysis to improve the reliability of the determination. A series of pre-processing operations are performed on each selected image frame to enhance the contour features, including histogram equalization to improve the contrast of the image, Gaussian filtering to reduce noise interference, sharpening filtering to enhance edge features, and the like. Color space conversion is performed according to the color features of the safety rope, such as conversion from RGB space to HSV space, so as to better separate the safety rope object from the background environment. A combination strategy of multiple edge detection algorithms is adopted to overcome the limitations of a single algorithm. The Canny edge detection algorithm can provide delicate edge information and good edge continuity, but may produce too many false edges in a noisy environment; the Sobel operator has good noise suppression ability, but the edge positioning accuracy is relatively low; the Laplace operator is sensitive to small changes in details and can detect small contour changes. The system adaptively adjusts the parameters of these algorithms according to the specific features of the image, such as the high and low threshold values of the Canny algorithm, the convolution kernel size of the Sobel operator, and the like. The application of morphological prior knowledge is a key technology to improve the accuracy of contour recognition. The safety rope, as a special industrial product, has obvious morphological characteristics: slender linear structure, relatively fixed diameter range, continuous geometric shape, and the like. The system uses these prior knowledge to screen and verify the detected contour lines. For example, the length threshold is used to exclude short noise contours, and is usually set to the minimum expected value of the actual safety rope length; the aspect ratio range is used to identify the slender linear structure; the linearity index is used to evaluate the straightness of the contour, although the safety rope may have bends, but the local should maintain a relatively smooth curve feature. The linearity of the contour line is calculated, and the smoothness of the contour is evaluated by fitting a straight line or a spline curve. If the contour is too curved or has jagged changes, it may be a noise or edge of other objects, and will be excluded from the candidate contours.

[0092] The candidate contour lines are subjected to connectivity analysis to construct topological connection relationship between the contours, and if the distance between the end points of the candidate contour lines is less than a preset connectivity threshold and the direction continuity satisfies a preset condition, the candidate contour lines are connected as continuous contours of the same safety rope object; the continuous contours and the topological connection relationship between the contours are collectively represented as a contour topological structure. The spatial adjacency relationship between the candidate contour lines is analyzed to identify contour segments that can belong to the same safety rope object. The setting of the connectivity threshold needs to consider multiple factors such as image resolution, actual diameter of the safety rope, observation distance, etc. Generally, the connectivity threshold is set to 1.5-2 times the number of pixels corresponding to the diameter of the safety rope in the image, which can connect the contour breaks caused by occlusion or light changes, and can also avoid incorrect connection of the contours of different safety ropes. The tangent vector analysis method is used to judge the direction continuity. For the end points of two contour lines that can be connected, the system calculates the tangent directions at the end points, and judges whether the two directions are continuous. If the included angle between the two tangent vectors is less than a preset threshold (usually 30-45 degrees) and the end point distance satisfies the connectivity condition, it is considered that the two contour segments belong to the same continuous contour. This direction continuity check can effectively avoid incorrect connection of the intersecting contours of different safety ropes into one object. The final representation of the contour topological structure uses a data structure of graph theory, and each continuous contour is represented as a node, and the topological connection relationship between the contours is represented as an edge.

[0093] Embodiment 2: based on the same inventive concept, as Figure 2 shown, the embodiment also provides an edge computing-based real-time analysis system for safety behaviors of construction personnel, which comprises an identifier allocation and man-rope mapping module, a high-risk event judgment module, a camera unit control module, a judgment result analysis and marking module, and a safety state abnormal signal generation module, and the modules are sequentially and communicatively connected;

[0094] The identifier allocation and man-rope mapping module is used to continuously acquire a monitoring video stream through an image acquisition device arranged in a construction area and input the monitoring video stream into an edge computing node, allocate a unique personnel identifier to each construction personnel object in the monitoring video stream and allocate a unique rope identifier to each safety rope object, and generate a man-rope association mapping table according to the correspondence between the personnel identifier and the rope identifier.

[0095] The high-risk event judgment module is used to process object data corresponding to the monitoring video stream in real time through the edge computing node, and when it is judged according to the object data that two or more safety rope objects have a crossing and occlusion event, a high-risk event signal is generated.

[0096] The camera unit control module is configured to, when a high-risk event signal is monitored, generate a control instruction by the edge computing node and send the control instruction to a target camera unit in the image acquisition device; the control instruction is used to drive the target camera unit to adjust a shooting parameter to obtain a supplementary perspective video sequence of the safety rope object that causes the cross-shielding.

[0097] The determination result analysis and identification module is configured to receive the supplementary perspective video sequence by the edge computing node and perform data analysis to obtain a determination result; when the determination result is that the cross-shielding event is a physical cross-shielding, a rope identifier corresponding to the safety rope object participating in the physical cross-shielding is determined, and a physical cross-shielding risk identifier is marked for the corresponding safety rope object in the person-rope association mapping table.

[0098] The safety state abnormal signal generation module is configured to, when any construction personnel object associated with a safety rope object in the person-rope association mapping table is marked with the physical cross-shielding risk identifier, generate and output a safety state abnormal signal.

[0099] It should be noted that, as for the system in the above-mentioned embodiments, the specific manner in which each module performs operations has been described in detail in the embodiments related to the method, and will not be described in detail here.

[0100] Finally, it should be noted that: although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacements for part of the technical features, any modification, equivalent replacement, improvement, etc. within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method for real-time analysis of safety behavior of construction workers based on edge computing, characterized in that, The method comprises: The image acquisition device arranged through the construction area continuously acquires a monitoring video stream and inputs an edge computing node, each construction personnel object in the monitoring video stream is assigned a unique personnel identifier, and each safety rope object is assigned a unique rope identifier; a person-rope association mapping table is generated according to the correspondence of the personnel identifier and the rope identifier; The edge computing node processes object data corresponding to the monitoring video stream in real time, and when it is judged according to the object data that two or more safety rope objects have a cross-shading event, a high-risk event signal is generated; When the high-risk event signal is monitored, the edge computing node generates a control instruction and sends the control instruction to a target camera unit in the image acquisition device; the control instruction is used to drive the target camera unit to adjust the shooting parameters to obtain a supplementary perspective video sequence of the safety rope object that has the cross-shading event; The edge computing node receives the supplementary perspective video sequence and performs data analysis to obtain a judgment result; when the judgment result is that the cross-shading event is a physical cross, the rope identifier corresponding to the safety rope object participating in the physical cross is determined, and a physical cross risk identifier is marked for the corresponding safety rope object in the person-rope association mapping table; When it is detected that any safety rope object associated with a construction personnel object in the person-rope association mapping table is marked with the physical cross risk identifier, a safety state abnormal signal is generated and output. 2.The edge computing based construction worker safety behavior real-time analysis method according to claim 1, characterized in that, The method when the high-risk event signal is monitored, the edge computing node generates a control instruction and sends the control instruction to a target camera unit in the image acquisition device comprises: First spatial position data of the safety rope object that has the cross-shading event in the construction space is extracted from the cross-shading event data of the high-risk event signal; deployment information of all image acquisition devices managed by the edge computing node is acquired, and the deployment information includes spatial position data and current perspective coverage range of the camera unit in the image acquisition device; According to the first spatial position data, the spatial position data of the camera unit, and the current perspective coverage range, a candidate camera unit is selected from all camera units whose current perspective coverage range does not completely cover the first spatial position data or has a perspective shading; According to the spatial relationship between the spatial position data of the candidate camera unit and the first spatial position data, a target camera unit is selected, and camera unit perspective adjustment parameters are calculated for the target camera unit; The edge computing node generates a control instruction containing the camera unit perspective adjustment parameters and sends the control instruction to the target camera unit. 3.The edge computing based construction worker safety behavior real-time analysis method of claim 2, wherein, The method of selecting a candidate camera unit from all camera units whose current perspective coverage range does not completely cover the first spatial position data or has a perspective shading comprises: According to the spatial position data, the current perspective coverage range parameter of each camera unit, and the first spatial position data, it is calculated whether the first spatial position data is located in the three-dimensional space region formed by the current perspective coverage range of the camera unit; When the first spatial position data is located in the stereoscopic space region, it is determined according to the pre-constructed construction scene three-dimensional model whether there is an occlusion on the line-of-sight path from the spatial position of the camera unit to the spatial point indicated by the first spatial position data, and if there is an occlusion, it is determined that the camera unit has a visual angle occlusion to the first spatial position data. When the first spatial position data is not located in the stereoscopic space region or is located in the stereoscopic space region but has a visual angle occlusion, the camera unit is recorded as a candidate camera unit. 4.The edge computing based construction worker safety behavior real-time analysis method of claim 3, wherein, The method for determining whether there is an occlusion on the line-of-sight path from the spatial position of the camera unit to the spatial point indicated by the first spatial position data according to the pre-constructed construction scene three-dimensional model comprises: obtaining a pre-constructed construction scene three-dimensional model, wherein the construction scene three-dimensional model comprises geometric data and spatial position information of fixed structures and equipment in the construction scene; calculating a line-of-sight vector connecting the two points according to the spatial position data of the camera unit and the first spatial position data; performing line-of-sight projection analysis along the line-of-sight vector in the construction scene three-dimensional model, detecting whether the path passed by the line-of-sight vector intersects with a three-dimensional entity represented by the geometric data of any fixed structure or equipment in the construction scene three-dimensional model; if intersection is detected, it is determined that there is an occlusion on the line-of-sight path; if no intersection is detected, it is determined that there is no occlusion on the line-of-sight path. 5.The edge computing based construction worker safety behavior real-time analysis method of claim 2, wherein, The method for selecting a target camera unit according to the spatial relationship between the spatial position data of the candidate camera unit and the first spatial position data comprises: calculating the estimated adjustment time required for the visual angle center axis to adjust from the current direction to point to the spatial point corresponding to the first spatial position data according to the mechanical motion parameters of the candidate camera unit, wherein the mechanical motion parameters include horizontal rotation angular velocity and pitch rotation angular velocity; evaluating the expected image quality of imaging the target region under the adjusted visual angle according to the spatial geometric relationship between the spatial position data of the candidate camera unit and the first spatial position data, wherein the spatial geometric relationship includes relative distance and relative angle; obtaining a comprehensive selection index of the candidate camera unit by data analysis of the estimated adjustment time and the expected image quality, wherein the comprehensive selection index is set to preferentially select the candidate camera unit with shorter estimated adjustment time under the premise of meeting the preset minimum expected image quality requirement; and sorting all candidate camera units according to the comprehensive selection index, and selecting the candidate camera unit with the optimal comprehensive selection index in the sorting result as the target camera unit. 6.The edge computing based construction worker safety behavior real-time analysis method of claim 5, wherein, The method for evaluating the expected image quality of imaging the target region under the adjusted visual angle according to the spatial geometric relationship between the spatial position data of the candidate camera unit and the first spatial position data comprises: calculating the straight-line distance from the optical center point of the candidate camera unit to the center point of the target region as the relative distance, and calculating the included angle between the current optical axis direction of the candidate camera unit and the vector pointing from the candidate camera unit to the center point of the target region as the relative angle; and According to the relative distance and the focal length parameter of the candidate camera unit, a pixel coverage number of the target region on an imaging plane is calculated through an imaging model to evaluate an imaging resolution; a perspective distortion coefficient is calculated according to a pitch angle component and a horizontal deflection angle component in the relative angle; and an expected image quality is obtained through a weighted fusion algorithm according to the imaging resolution and the perspective distortion coefficient. 7.The edge computing based construction worker safety behavior real-time analysis method of claim 2, wherein, The method for calculating the camera view angle adjustment parameter for the target camera unit comprises: acquiring a three-dimensional coordinate of spatial position data of the target camera unit and a current optical axis direction vector; acquiring a three-dimensional coordinate of a target space point corresponding to the first spatial position data; and calculating a target direction vector from the spatial position of the target camera unit to the target space point; calculating a horizontal deflection angle of rotation around a vertical axis and a pitch angle of rotation around a horizontal axis required for the current optical axis direction vector to coincide with the target direction vector according to the current optical axis direction vector of the target camera unit and the calculated target direction vector; calculating a focal length adjustment amount required for the target space point to be clearly imaged according to a three-dimensional spatial distance between the spatial position of the target camera unit and the target space point in combination with lens optical parameters of the target camera unit; and combining the horizontal deflection angle, the pitch angle and the focal length adjustment amount to form the camera view angle adjustment parameter. 8.The edge computing based construction worker safety behavior real-time analysis method of claim 1, wherein, The method for receiving the supplementary view video sequence and performing data analysis to obtain a determination result by the edge computing node comprises: The edge computing node analyzes the supplementary view video sequence to perform object contour segmentation on an image region where the safety rope objects that have crossed and occluded each other are located to obtain a contour topological structure of the safety rope objects near the intersection point; if the contour topological structure shows that there are independent and mutually disconnected safety rope object contours at the intersection point, it is determined that the crossed and occluded event is non-physical; and if the contour topological structure shows that there are shared contours or contour connection points at the intersection point, so that the contours of different safety rope objects cannot be clearly segmented into independent objects, it is determined that the crossed and occluded event is physical. 9.The edge computing based construction worker safety behavior real-time analysis method of claim 8, wherein, The method for performing object contour segmentation on an image region where the safety rope objects that have crossed and occluded each other are located to obtain a contour topological structure of the safety rope objects near the intersection point comprises: image frames containing the intersection point region are extracted from the supplementary view video sequence and image preprocessing is performed to enhance contour features, and an edge detection algorithm is used to identify all potential contour lines in the image frames; candidate contour lines belonging to the safety rope objects are selected from all potential contour lines according to morphological prior knowledge of the safety rope objects, and the morphological prior knowledge includes a length threshold of the contour, a height-width ratio range and linearity; The candidate contour lines are subjected to connectivity analysis to construct topological connection relationship between the contours, and if the distance between the end points of the candidate contour lines is less than a preset connectivity threshold and the direction continuity satisfies a preset condition, the candidate contour lines are connected as continuous contours of the same safety rope object; the continuous contours and the topological connection relationship between the contours are collectively represented as a contour topological structure.

10. A real-time analysis system for construction worker safety behavior based on edge computing, characterized in that, The system comprises an identifier allocation and human rope mapping module, a high-risk event judgment module, a camera unit control module, a judgment result analysis and identification module, and a safety state abnormal signal generation module, and the modules are sequentially and communicatively connected; The identifier allocation and human rope mapping module is configured to continuously acquire a monitoring video stream by an image acquisition device arranged in a construction area and input the monitoring video stream into an edge computing node, allocate a unique personnel identifier to each construction personnel object in the monitoring video stream and allocate a unique rope identifier to each safety rope object, and generate a human rope association mapping table according to the correspondence between the personnel identifier and the rope identifier; The high-risk event judgment module is configured to process object data corresponding to the monitoring video stream in real time by the edge computing node, and generate a high-risk event signal when it is determined according to the object data that two or more safety rope objects have a cross-shading event; The camera unit control module is configured to generate a control instruction by the edge computing node and send the control instruction to a target camera unit in the image acquisition device when a high-risk event signal is detected, and the control instruction is used to drive the target camera unit to adjust a shooting parameter to acquire a supplementary perspective video sequence of the safety rope objects having the cross-shading event; The judgment result analysis and identification module is configured to receive the supplementary perspective video sequence by the edge computing node and perform data analysis to obtain a judgment result, determine a rope identifier corresponding to the safety rope objects participating in the physical cross when the judgment result is that the cross-shading event is a physical cross, and mark the corresponding safety rope objects with a physical cross risk identifier in the human rope association mapping table; The safety state abnormal signal generation module is configured to generate and output a safety state abnormal signal when it is detected that any safety rope object associated with a construction personnel object in the human rope association mapping table is marked with the physical cross risk identifier.

Images