Iron tower operation safety early warning system and method

Abstract

The invention discloses an iron tower operation safety early warning system and method, and relates to the technical field related to intelligent safety early warning, and the system comprises a data obtaining unit which is used for obtaining Beidou three-dimensional coordinate data and an operation site panoramic image in real time; the operation scene model establishing unit is used for establishing a dynamic three-dimensional operation scene model; the prediction model building unit is used for tracking the motion trail of each operation target based on the coordinate data and building a prediction model fusing kinematics rules and machine learning; and the risk level judgment unit is used for carrying out risk level judgment according to the spatial position and the attribute characteristics of the intersection area and generating operation early warning information. The technical problems of insufficient safety early warning accuracy and timeliness caused by insufficient high-altitude operation target dynamic track monitoring precision and risk early warning lagging in the prior art are solved, and the purposes of accurately identifying the movement track intersection risk in real time, dynamically evaluating the safety level and improving the safety early warning efficiency are achieved. And the early warning accuracy and timeliness of the operation safety of the iron tower are improved.

Description

Technical Field

[0001] This application relates to the field of intelligent safety early warning technology, specifically to a safety early warning system and method for tower operations. Background Technology

[0002] Safety management in tower operations faces severe challenges. The complex environment of high-altitude operations, involving multiple targets such as personnel, equipment, and machinery, makes them prone to accidents due to spatial overlap or conflicting movement trajectories. Traditional safety warning systems struggle to identify dynamic risks in real-time and accurately, especially in complex three-dimensional work scenarios, lacking intelligent prediction and collision warning capabilities for multi-target movement trajectories. While the BeiDou Navigation Satellite System provides centimeter-level three-dimensional coordinate data and drone aerial photography can quickly acquire panoramic images of the work site, the key lies in utilizing this data for accurate trajectory prediction and risk warning. Furthermore, existing trajectory prediction methods are ill-suited to the complex dynamic interactions of multiple targets in tower operations, failing to effectively integrate attributes such as equipment type, movement speed, and work stage for dynamic risk assessment, thus impacting the safety and efficiency of high-altitude operation management.

[0003] Therefore, current technologies suffer from insufficient accuracy in monitoring the dynamic trajectory of targets operating at heights and delayed risk warnings, resulting in inadequate accuracy and timeliness of safety warnings. Summary of the Invention

[0004] This application provides a safety early warning system and method for tower operations, which solves the technical problems of insufficient accuracy in monitoring the dynamic trajectory of high-altitude operation targets and delayed risk warning in the existing technology, resulting in insufficient accuracy and timeliness of safety early warning. It achieves the technical effect of real-time and accurate identification of the risk of intersection of movement trajectories, dynamic assessment of safety level, and improvement of the accuracy and timeliness of safety early warning for tower operations.

[0005] This application provides a tower operation safety early warning system, which includes: a data acquisition unit for acquiring BeiDou 3D coordinate data of each operation target in real time and simultaneously acquiring panoramic images of the operation site through UAV aerial photography; an operation scene model building unit for fusing the 3D coordinate data with the panoramic images of the operation site through spatiotemporal registration to build a dynamic 3D operation scene model containing all targets; a prediction model building unit for tracking the movement trajectory of each operation target based on time-series coordinate data, building a prediction model that integrates kinematic rules and machine learning, and predicting the intersection area between the movement paths of the targets; and a risk level determination unit for determining the risk level based on the spatial location of the intersection area and the attribute characteristics of the corresponding target, and generating operation early warning information.

[0006] In a possible implementation, the tower operation safety early warning system also performs the following processing: establishing a mapping relationship between the UAV camera coordinate system and the BeiDou three-dimensional coordinate system; spatially aligning the three-dimensional coordinate data with the panoramic image of the operation site based on the mapping relationship; performing point cloud registration of the tower structure feature points extracted from the UAV image with the pre-stored tower BIM model; using the coordinates after point cloud registration as a reference, and according to the coordinate mapping relationship, fusing the three-dimensional coordinate data into the point cloud registration coordinates to establish a dynamic three-dimensional operation scene model containing all targets.

[0007] In a possible implementation, the tower operation safety early warning system also performs the following processing: acquiring UAV POS data, including longitude, latitude, altitude, pitch angle, and yaw angle, and binding the UAV POS data and image data to the same timestamp to ensure time synchronization; based on the UAV POS data, constructing a transformation matrix and translation vector between image pixel coordinates and BeiDou world coordinates through perspective projection transformation to determine the mapping relationship.

[0008] In a possible implementation, the tower operation safety early warning system also performs the following processing: extracting tower structure feature points from the UAV imagery and converting the extracted feature points to the BeiDou world coordinate system using mapping relationships; extracting corresponding feature points from the BIM model; and aligning the image feature points with the BIM model feature points using a point cloud registration algorithm to complete the point cloud registration of the UAV imagery features with the pre-stored tower BIM model.

[0009] In a possible implementation, the tower operation safety early warning system also performs the following processing: initializing rotation and translation parameters using the symmetry features of the tower; calculating the spatial distance between image feature points and BIM model points; and iteratively optimizing the registration accuracy based on the rotation and translation parameters, spatial distance, and steel structure curvature similarity until the target matching error is reached, thereby determining the point cloud registration.

[0010] In a possible implementation, the tower operation safety early warning system also performs the following processing: constructing a kinematic sub-model based on the equipment's physical parameters and outputting the rigid body motion trajectory; constructing a machine learning sub-model through a long short-term memory network to learn personnel behavior patterns; and using the kinematic sub-model and the machine learning sub-model to construct a prediction model, which is used to output a probability cloud map of the motion path within a future time window.

[0011] In a possible implementation, the tower operation safety early warning system further performs the following processing: randomly sampling the probability distribution of random factors during equipment movement to generate position samples for each time step; combining a kinematic sub-model with each sample point to simulate the equipment's trajectory from the current moment to a future time window, statistically analyzing the trajectories of all sample points, calculating the probability of the equipment appearing at each position, and generating the probability distribution of the equipment's movement path; extracting features from historical trajectory data of personnel, including position, speed, acceleration, and direction of movement, and using an attention-based neural network model to model the personnel's trajectory, calculating the importance weight of each time step, and weighting the key path points in the personnel's trajectory based on the attention weight; using the weighted features as input, predicting the personnel's movement path within a future time window through a neural network model, and generating the probability distribution of the personnel's movement path; superimposing the probability distribution of the equipment's movement path with the probability distribution of the personnel's movement path, and calculating the intersection probability of the two at each position; determining the intersection probability using a preset probability threshold, marking the region probability according to the determined probability result, and generating a movement path probability cloud map.

[0012] In a possible implementation, the tower operation safety early warning system also performs the following processing: establishing a risk assessment index system based on the spatial location, size, and shape of the intersection area, as well as the speed, direction, and type attributes of the corresponding target; using a fuzzy logic algorithm to comprehensively evaluate the risk assessment indexes, classifying them into at least three risk levels: high, medium, and low; and generating corresponding operation early warning signals based on the risk level, including the early warning level, early warning time, early warning location, and response measures.

[0013] In a possible implementation, the tower operation safety early warning system also performs the following processing: collecting signal attenuation patterns of typical tower types and constructing a multi-path error compensation library for steel structures; based on the multi-path error compensation library for steel structures, matching the characteristics of the operation area in real time, correcting error points, and resetting the position coordinates of the target; and predicting the intersection area based on the reset target position coordinates.

[0014] This application also provides a method for safety early warning of tower operations, including: acquiring BeiDou three-dimensional coordinate data of each operation target in real time, and simultaneously acquiring panoramic images of the operation site through drone aerial photography; fusing the three-dimensional coordinate data with the panoramic images of the operation site through spatiotemporal registration to establish a dynamic three-dimensional operation scene model containing all targets; tracking the movement trajectory of each operation target based on time-series coordinate data, establishing a prediction model that integrates kinematic rules and machine learning, and predicting the intersection area between the movement paths of the targets; determining the risk level based on the spatial location of the intersection area and the attribute characteristics of the corresponding targets, and generating operation early warning information.

[0015] The proposed tower operation safety early warning system and method includes: a data acquisition unit for real-time acquisition of BeiDou 3D coordinate data and panoramic images of the operation site; an operation scene model building unit for establishing a dynamic 3D operation scene model; a prediction model building unit for tracking the motion trajectory of each operation target based on coordinate data and establishing a prediction model integrating kinematic rules and machine learning; and a risk level determination unit for determining the risk level based on the spatial location and attribute characteristics of the intersection area and generating operation early warning information. This system solves the technical problems of insufficient accuracy in monitoring the dynamic trajectory of high-altitude operation targets and delayed risk warnings in existing technologies, resulting in insufficient accuracy and timeliness of safety early warnings. It achieves the technical effect of real-time and accurate identification of motion trajectory intersection risks, dynamic assessment of safety levels, and improved accuracy and timeliness of tower operation safety early warnings. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of this disclosure, the accompanying drawings of the embodiments of this disclosure will be briefly described below. Flowcharts are used in this application to illustrate the operations performed by the system according to the embodiments of this application. It should be understood that the preceding or following operations are not necessarily performed precisely in sequence. Instead, various steps can be processed in reverse order or simultaneously as needed. Furthermore, other operations can be added to these processes, or one or more steps can be removed from these processes.

[0017] Figure 1 This is a schematic diagram of the structure of the tower operation safety early warning system provided in the embodiment of this application.

[0018] Figure 2 This is a schematic diagram of the safety early warning method for tower operations provided in an embodiment of this application.

[0019] Figure labeling: Data acquisition unit 10, operation scenario model building unit 20, prediction model building unit 30, risk level determination unit 40. Detailed Implementation

[0020] The above description is merely an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, specific embodiments of this application are given below.

[0021] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description of this application will be provided in conjunction with the accompanying drawings. The described embodiments should not be considered as limitations on this application. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0022] In the following description, references to "some embodiments" describe a subset of all possible embodiments. However, it is understood that "some embodiments" can be the same or different subsets of all possible embodiments and can be combined with each other without conflict. The terms "first" and "second" are used merely to distinguish similar objects and do not represent a specific ordering of objects. The terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, system, product, or server that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or modules not explicitly listed or inherent to these processes, products, or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only.

[0023] This application provides a tower operation safety early warning system, such as Figure 1 As shown, the system includes:

[0024] The data acquisition unit 10 is used to acquire the BeiDou three-dimensional coordinate data of each operation target in real time, and simultaneously acquire panoramic images of the operation site through drone aerial photography.

[0025] Preferably, the BeiDou 3D coordinate data of each work target is acquired in real time, including the 3D coordinate data of personnel, equipment, tools, etc., such as longitude, latitude, and altitude. Specifically, workers wear BeiDou positioning terminals, such as handheld devices or safety helmets with built-in modules, to acquire personnel 3D coordinate data. Cranes, elevators and other mechanical equipment are equipped with BeiDou high-precision positioning modules to collect equipment 3D coordinate data. Large tools such as tower hoisting equipment are equipped with lightweight positioning modules to acquire tool 3D coordinate data. This is used to track the precise position of all dynamic targets in real time. Simultaneously, panoramic images of the work site are acquired through drone aerial photography. The drone is equipped with a high-resolution camera and a POS system, as well as an RTK positioning module to ensure accurate shooting position. The POS system is used to record the drone's own position and attitude. The drone flies over the work area along a preset route, capturing multi-angle panoramic images of the work site in real time. It simultaneously records the timestamp, latitude and longitude, pitch angle, yaw angle, etc. of each frame of the image. The SLAM algorithm is used to stitch multiple frames of images into a panoramic image, which can extract and identify feature points such as tower structures, equipment, and personnel in the images, such as the edges of tower materials and the outline of cranes. This provides a visual global perspective of the work site and makes up for the lack of environmental information in Beidou data.

[0026] The operation scene model building unit 20 is used to fuse three-dimensional coordinate data with panoramic images of the operation site through spatiotemporal registration to build a dynamic three-dimensional operation scene model containing all targets.

[0027] Preferably, spatiotemporal registration refers to the time synchronization and spatial alignment of BeiDou coordinate data and UAV panoramic images from different sources, at different times, and in different coordinate systems. Specifically, this includes: establishing a three-dimensional spatial coordinate system with the base of the tower as the origin, where the X-axis points due east, the Y-axis points due north, and the Z-axis points vertically upward; using the Network Time Protocol (NTP) to ensure that the timestamps of BeiDou positioning data and each frame of images captured by the UAV are strictly matched, with a time synchronization accuracy of no less than 1 millisecond; establishing a mathematical mapping relationship between the BeiDou world coordinate system and the UAV image coordinate system through a coordinate transformation matrix T (T=[R|t], where R is a 3×3 rotation matrix and t is a 3×1 translation vector); fusing the registered BeiDou coordinate data with environmental information in the UAV imagery, including coordinate system mapping and feature point registration, and then constructing a three-dimensional scene with spatiotemporal attributes, i.e., a dynamic three-dimensional operation scene model containing all targets, to display the real-time position and movement status of personnel, equipment, and the tower structure.

[0028] Furthermore, the specific configuration of the operation scene model establishment unit 20 also includes: establishing a mapping relationship between the UAV camera coordinate system and the Beidou three-dimensional coordinate system; spatially aligning the three-dimensional coordinate data with the panoramic image of the operation site based on the mapping relationship; performing point cloud registration of the tower structure feature points extracted from the UAV image with the pre-stored tower BIM model; using the coordinates after point cloud registration as a reference, and according to the coordinate mapping relationship, fusing the three-dimensional coordinate data into the point cloud registration coordinates to establish a dynamic three-dimensional operation scene model containing all targets.

[0029] Preferably, the images captured by the UAV are in a 2D pixel coordinate system, while the BeiDou positioning data is in a 3D world coordinate system. A mapping relationship is established between the UAV camera coordinate system and the BeiDou 3D coordinate system. Specifically, the latitude, longitude, pitch, and yaw angle data are determined using the POS system onboard the UAV, and the real-world coordinates corresponding to each pixel in the image are calculated. Then, a transformation matrix between world coordinates and pixel coordinates is established, ensuring strict alignment of the timestamps of the image frames and BeiDou data. Then, based on the mapping relationship, the 3D coordinate data is spatially aligned with the panoramic image of the work site. The tower structure feature points extracted from the UAV images are then registered with the pre-stored tower BIM model as point clouds. Specifically, from the UAV images... The process involves extracting steel structure feature points from the tower, such as bolt holes, welds, and angle steel edges. Corresponding 3D feature points are then extracted from the pre-stored tower BIM model. An iterative nearest-point algorithm or feature matching algorithm is used to align the tower structure feature points in the image with the pre-stored tower BIM model, optimizing rotation and translation parameters. Simultaneously, the symmetry of the tower and the consistency of the steel structure curvature are utilized to iteratively reduce registration errors. Finally, using the coordinates after point cloud registration as a reference, the 3D coordinate data is mapped and fused into the point cloud registration coordinates through coordinate mapping relationships, generating a dynamic 3D operational scene model containing all targets, such as the pre-stored tower BIM model, real-time updated equipment and personnel coordinates and movement trajectories, and drone imagery.

[0030] Furthermore, the specific configuration of the operation scenario model establishment unit 20 also includes acquiring UAV POS data, including longitude, latitude, altitude, pitch angle, and yaw angle, and binding the UAV POS data and image data to the same timestamp to ensure time synchronization; based on the UAV POS data, constructing a transformation matrix and translation vector between image pixel coordinates and BeiDou world coordinates through perspective projection transformation to determine the mapping relationship.

[0031] Preferably, the UAV POS data is acquired through the POS system onboard the UAV. This data includes the spatial attitude and position information recorded by the UAV when capturing each frame of imagery, such as longitude, latitude, altitude, pitch angle, and yaw angle. The imagery data captured by the UAV and the POS data are time-aligned, i.e., bound to the same timestamp, to avoid coordinate mapping errors caused by transmission delays. Then, based on the UAV POS data, a transformation matrix and translation vector between image pixel coordinates and BeiDou world coordinates are constructed through perspective projection transformation. Specifically, using the UAV's longitude, latitude, altitude, pitch angle, and yaw angle POS data, the transformation matrix from 3D world coordinates to 2D pixel coordinates is calculated through collinearity equations, ensuring that the timestamps of the image frames and BeiDou data are strictly aligned. The offset of the UAV's current position relative to the BeiDou coordinate origin is then determined as the translation vector, ultimately establishing the mapping relationship.

[0032] Furthermore, the specific configuration of the operation scenario model establishment unit 20 also includes: using mapping relationships to transform the tower structure feature points extracted from the UAV image into the BeiDou world coordinate system; extracting corresponding feature points in the BIM model; and aligning the image feature points with the BIM model feature points through a point cloud registration algorithm to complete the point cloud registration of the UAV image features with the pre-stored tower BIM model.

[0033] Preferably, the tower feature points in the 2D images captured by the drone are precisely matched with the 3D feature points of the pre-stored tower BIM model to establish a spatial correspondence, ultimately achieving precise alignment of the dynamic 3D scene. Specifically, based on a CNN network, tower material feature points are extracted from the drone images, including geometric features such as angle steel edges, bolt holes, and welds, as well as texture features such as rust patches and signage text. Then, using the mapping relationship determined by the perspective projection matrix, the extracted feature points are transformed into the BeiDou world coordinate system. That is, the camera coordinate system coordinates of the extracted feature points are calculated through back projection, and then the camera coordinates are transformed... The coordinates are changed to BeiDou world coordinates; the 3D point cloud of the tower or the CAD model of key components of the tower is exported from the pre-stored tower BIM model, and feature points that match the image are extracted, such as the center coordinates of the tower foot bolt holes. The image feature points are aligned with the BIM model feature points through feature matching and iterative nearest point matching. This includes estimating the initial rotation matrix using the symmetry of the tower, calculating the Euclidean distance between the image feature points and the BIM model points, and then solving the optimal rotation and translation matrices respectively. This completes the point cloud registration of the UAV image features and the pre-stored tower BIM model, realizing the virtual-real fusion of UAV image and BIM model.

[0034] Furthermore, the specific configuration of the operation scenario model establishment unit 20 also includes: initializing the rotation and translation parameters using the symmetry features of the iron tower; calculating the spatial distance between the image feature points and the BIM model points; and iteratively optimizing the registration accuracy based on the rotation and translation parameters, spatial distance, and steel structure curvature similarity until the target matching error is reached, thus determining the point cloud registration.

[0035] Preferably, efficient and high-precision point cloud registration is achieved through tower symmetry initialization and curvature optimization iteration. Specifically, tower symmetry feature analysis is performed, i.e., typical tower structures have axisymmetric (such as four-corner towers) or central symmetry (such as steel pipe towers) characteristics. By detecting the symmetry plane / axis in the BIM model, the parameter search range is greatly reduced. Then, the main symmetry axis direction of the BIM model is extracted. According to the drone shooting angle, the image feature points are rotated to align with the symmetry direction of the BIM model. The initial rotation and translation parameters are calculated. For example, if the tower is four-corner symmetric, the image feature points are rotated by 90°, 180°, and 270° respectively during initialization. The angle with the smallest matching error is selected as the initial value.

[0036] Preferably, Euclidean distance is used to calculate the spatial distance between the nearest neighbor pairs of image feature points and BIM model points, and outlier point pairs with a distance greater than a threshold (e.g., 10cm) are removed to avoid noise interference. Then, based on rotation and translation parameters and spatial distance, the registration accuracy is iteratively optimized by combining the similarity of steel structure curvature. Specifically, the curvature features of the steel structure are extracted, that is, the local curvature of the BIM model and image feature points is calculated to reflect the degree of bending of the steel structure. Then, an objective function is constructed to minimize the distance error and the difference in steel structure curvature. The weight coefficient of steel structure curvature is usually 0.1~0.3, until the target matching error is reached, for example, when the average registration error is less than 1cm, fast and high-precision point cloud registration is achieved, providing a reliable dynamic three-dimensional benchmark for safety early warning.

[0037] The prediction model building unit 30 is used to track the motion trajectory of each task target based on time series coordinate data, build a prediction model that integrates kinematic rules and machine learning, and predict the intersection area between the target motion paths.

[0038] The specific configuration of the prediction model building unit 30 further includes: constructing a kinematic sub-model based on the physical parameters of the equipment and outputting the rigid body motion trajectory; constructing a machine learning sub-model through a long short-term memory network to learn the behavior patterns of personnel; and superimposing the kinematic sub-model and the machine learning sub-model through a weighted fusion method to construct a prediction model, with the fusion weight dynamically adjusted according to the prediction accuracy, for outputting a probability cloud map of the motion path within the future time window.

[0039] Preferably, a predictive model is established by integrating kinematic rules and machine learning behavior to predict the future movement paths of dynamic targets such as personnel and equipment in tower operation scenarios, and to identify potential intersection areas between different target trajectories, i.e., collision risk points. Specifically, the time-series coordinate data includes the BeiDou coordinate data of each operation target, thereby tracking and determining the movement trajectory of each operator and piece of equipment. Then, for mechanical equipment such as cranes and elevators, a rigid body motion model is constructed based on physical parameters such as the length of its robotic arm, rotation speed, and acceleration limits, as a kinematic sub-model, and the future trajectory is predicted using Newton's equations of motion. For personnel behavior, a Long Short-Term Memory (LSI) network is used. TM takes time-series data such as position, velocity, acceleration, and orientation angle as input, learns motion patterns from historical trajectory data, such as walking path preferences and pausing habits, and determines machine learning sub-models to output the probability distribution of personnel positions within future time windows. Then, it merges the kinematics sub-model and the machine learning sub-model to obtain a fused prediction model, generates a motion path probability cloud map, quantifies the probability of each spatial point being occupied in the future, and predicts the intersection area between target motion paths. For example, when the probability distribution overlap area between equipment and personnel exceeds a threshold (such as 30%), it is marked as a high-risk intersection area, thereby achieving accurate prediction of motion trajectories and dynamic risk identification in complex scenarios.

[0040] Furthermore, the specific configuration of the prediction model building unit 30 also includes: randomly sampling the probability distribution of random factors during the movement of the device, generating position samples at each time step; combining a kinematic sub-model with each sample point to simulate the movement trajectory of the device from the current moment to the future time window, performing statistical analysis on the movement trajectories of all sample points, calculating the probability of the device appearing at each position, and generating the probability distribution of the device's movement path; extracting features from the historical trajectory data of personnel, including position, speed, acceleration, and direction of movement, using a neural network model based on an attention mechanism to model the personnel trajectory, calculating the importance weight of each time step, and weighting the key path points in the personnel trajectory based on the attention weight; using the weighted features as input, predicting the movement path of personnel within the future time window through the neural network model, and generating the probability distribution of the personnel's movement path; superimposing the probability distribution of the device's movement path with the probability distribution of the personnel's movement path, calculating the intersection probability of the two at each position; using a preset probability threshold to determine the intersection probability, marking the region probability according to the determined probability result, and generating a movement path probability cloud map.

[0041] Preferably, by using Monte Carlo random sampling and neural network prediction, the probability distributions of future movement paths of equipment and personnel are generated respectively. The intersection probability of the two in three-dimensional space is calculated by superimposing the results, and finally a visualized movement path probability cloud map is formed, thereby realizing quantitative early warning of collision risk. Specifically, the movement of equipment such as cranes is affected by random factors such as mechanical control errors and wind disturbances. It is necessary to model the probability distribution. At each time step, the probability distribution of random factors in the movement process of the equipment is generated by Monte Carlo random sampling to determine possible position samples. For each sample point, the kinematic sub-model is combined to simulate the movement trajectory of the equipment from the current moment to the future time window. That is, the position sample is substituted into the rigid body kinematic equation to calculate the future movement trajectory, and the frequency of occurrence of the spatial positions passed by the movement trajectory corresponding to all samples in the future time window is statistically analyzed. The probability of the equipment appearing at each position is calculated, that is, the ratio of the number of times the trajectory passes through that point to the total number of samples, and the probability distribution of the movement path of the equipment is generated.

[0042] Preferably, features are extracted from historical trajectory data of personnel, including location, speed, acceleration, and direction of movement. An attention mechanism is used to calculate the importance weight of each time step, highlighting key behaviors. Then, a neural network model is used to model the personnel trajectory, weighting key path points in the trajectory based on attention weights. Using the weighted features as input, the neural network model predicts the personnel's movement path within a future time window, outputting a multivariate Gaussian distribution of the future path as the probability distribution of the personnel's movement path, i.e., a probability grid map, where each grid stores the probability value of personnel arrival. Next, the probability distributions of equipment and personnel movement paths are superimposed, i.e., the product of the probability distributions of equipment and personnel at each spatial grid point is calculated as the intersection probability. A preset probability threshold (0.3) is then used to determine the intersection probability. Based on the determined probability results, regional probability marking is performed, with areas where the intersection probability is greater than the preset probability threshold marked as high-risk areas and marked in red. Finally, a movement path probability cloud map is generated, thereby achieving dynamic risk visualization and enabling accurate safety early warning for high-altitude operations.

[0043] The risk level determination unit 40 is used to determine the risk level based on the spatial location of the intersection area and the attribute characteristics of the corresponding target, and generate operation early warning information.

[0044] The specific configuration of the risk level determination unit 40 also includes: establishing a risk assessment index system based on the spatial location, size, and shape of the intersection area, as well as the speed, direction, and type attributes of the corresponding target, including distance, speed, and time risk indicators; using a fuzzy logic algorithm to comprehensively evaluate the risk assessment indicators, classifying them into at least three risk levels: high, medium, and low; and generating corresponding operational early warning signals based on the risk level, including the early warning level, early warning time, early warning location, and response measures.

[0045] Preferably, the risk level is determined based on the spatial location, size, and shape of the intersection area, as well as the speed, direction, and type of the corresponding target. Spatial location refers to the position of the intersection area relative to key components of the tower, such as proximity to high-voltage lines or the tower top; spatial size refers to the volume of the intersection area; spatial shape may be elongated (e.g., the area swept by a crane boom) or dotted (e.g., a personnel standing point); speed is the movement speed of the equipment or personnel, such as a hook moving at 2 m / s or a person walking at 1 m / s; direction is the angle of the movement vector; type refers to the target type weight, such as crane 1.0 > personnel 0.7 > tools 0.3. Then, core risk assessment indicators based on spatial characteristics, dynamic characteristics, and attribute characteristics are established to create a risk assessment indicator system. For example, an intersection area volume greater than 1 m³ is considered high risk; a narrow, elongated area is more dangerous than a dispersed area; the risk doubles when the relative speed is greater than 2 m / s; opposing movements are three times riskier than same-direction movements; the collision weight for a crane-personnel collision is 0.9; and the collision weight for a drone-tool collision is 0.4.

[0046] Preferably, a fuzzy logic algorithm is used to comprehensively evaluate the risk assessment indicators. This involves converting the precisely input spatial location, size, and shape of the intersection area, as well as the speed, direction, and type attributes of the corresponding target, into a fuzzy set. Basic safety thresholds are set: 3m between personnel and equipment, and 1.5m between personnel and tools. Dynamic correction coefficients are introduced, such as wind speed coefficient = 0.05 × wind speed value, and visibility coefficient = 0.02 × insufficient visibility value. The risk assessment indicators are then comprehensively evaluated, classifying them into at least three risk levels: high, medium, and low. Corresponding operational warning signals are generated based on the risk level, including the warning level, warning time, warning location, and response measures. The warning time is the estimated collision time, such as an estimated danger occurring in 8.3 seconds. The warning location is the estimated three-dimensional coordinates of the collision, such as longitude 116.404°, latitude 39.915°, and altitude 152m. Response measures include automated commands, such as the crane immediately stopping lateral movement. When the predicted minimum distance is less than or equal to the safety threshold (the product of the basic safety threshold and the dynamic correction coefficient), if the predicted minimum distance is less than or equal to 0.4 times the safety threshold, it is considered high risk. Operations should be stopped immediately, and audible and visual alarms should be activated, safety ropes should be locked, and the drone rescue net should be launched. If the predicted minimum distance is less than or equal to 0.4 times the safety threshold and less than or equal to 0.6 times the safety threshold, it is considered medium risk. A voice prompt should be given to avoid the obstacle, and the equipment braking components should be activated to reduce the equipment speed. If the predicted minimum distance is greater than or equal to 0.6 times the safety threshold, it is considered low risk. An AR visual warning should be triggered, such as a yellow label on the monitoring interface.

[0047] Furthermore, the tower operation safety early warning system also includes: collecting signal attenuation patterns of typical tower types and constructing a multi-path error compensation library for steel structures; based on the multi-path error compensation library for steel structures, matching the characteristics of the operation area in real time, correcting error points, and resetting the target's position coordinates; and predicting the intersection area based on the reset target position coordinates.

[0048] Preferably, testing equipment is deployed on typical iron towers such as wine glass towers, cat head towers, and converter stations to record signal attenuation characteristics at different locations, including carrier phase jumps and signal-to-noise ratio fluctuations, forming a signal attenuation pattern library. Then, machine learning analysis is used to construct the mapping relationship between tower layout and errors, and a multi-path error compensation library for steel structures is built. The multi-path error compensation library for steel structures is continuously optimized by inverting multi-path errors through dual-frequency RTK terminals worn by operators. Next, the multi-path error compensation library for steel structures is matched with the characteristics of the work area in real time. That is, according to the BIM model of the current tower type, the corresponding error compensation library is loaded, the location of the target and its distance from the nearest steel beam are detected in real time, the compensation value is queried to obtain the correction amount, and then the error point is corrected to reset the position coordinates of the target. Finally, the reset target position coordinates are input into the motion trajectory prediction model to predict the intersection area, thereby improving the accuracy of intersection area prediction and thus improving the accuracy and timeliness of tower operation safety early warning.

[0049] In the above text, refer to Figure 1 A tower operation safety early warning system according to an embodiment of the present invention has been described in detail. Next, reference will be made to... Figure 2 A method for early warning of safety during tower operations according to an embodiment of the present invention is described. The method for early warning of safety during tower operations, such as... Figure 2 As shown, the method includes: acquiring BeiDou 3D coordinate data of each operation target in real time, and simultaneously acquiring panoramic images of the operation site through UAV aerial photography; fusing the 3D coordinate data with the panoramic images of the operation site through spatiotemporal registration to establish a dynamic 3D operation scene model containing all targets; tracking the motion trajectory of each operation target based on time-series coordinate data, establishing a prediction model that integrates kinematic rules and machine learning to predict the intersection area between the target motion paths; determining the risk level based on the spatial location of the intersection area and the attribute characteristics of the corresponding target, and generating operation early warning information.

[0050] In one possible implementation, the tower operation safety early warning method further includes: establishing a mapping relationship between the UAV camera coordinate system and the BeiDou three-dimensional coordinate system; spatially aligning the three-dimensional coordinate data with the panoramic image of the operation site based on the mapping relationship; performing point cloud registration of the tower structure feature points extracted from the UAV image with the pre-stored tower BIM model; using the coordinates after point cloud registration as a reference, and according to the coordinate mapping relationship, fusing the three-dimensional coordinate data into the point cloud registration coordinates to establish a dynamic three-dimensional operation scene model containing all targets.

[0051] In one possible implementation, the tower operation safety early warning method further includes: acquiring UAV POS data, including longitude, latitude, altitude, pitch angle, and yaw angle, and binding the UAV POS data and image data to the same timestamp to ensure time synchronization; based on the UAV POS data, constructing a transformation matrix and translation vector between image pixel coordinates and BeiDou world coordinates through perspective projection transformation to determine the mapping relationship.

[0052] In one possible implementation, the tower operation safety early warning method further includes: extracting tower structure feature points from the UAV imagery, converting the extracted feature points to the BeiDou world coordinate system using a mapping relationship; extracting corresponding feature points from the BIM model; and aligning the image feature points with the BIM model feature points using a point cloud registration algorithm to complete the point cloud registration of the UAV image features with the pre-stored tower BIM model.

[0053] In one possible implementation, the tower operation safety early warning method further includes: initializing rotation and translation parameters using the symmetry characteristics of the tower; calculating the spatial distance between image feature points and BIM model points; and iteratively optimizing the registration accuracy based on the rotation and translation parameters, spatial distance, and steel structure curvature similarity until the target matching error is reached, thereby determining the point cloud registration.

[0054] In one possible implementation, the tower operation safety early warning method further includes: constructing a kinematic sub-model based on the equipment's physical parameters and outputting the rigid body motion trajectory; constructing a machine learning sub-model through a long short-term memory network to learn personnel behavior patterns; and using the kinematic sub-model and the machine learning sub-model to construct a prediction model, which is used to output a probability cloud map of the motion path within a future time window.

[0055] In one possible implementation, the tower operation safety early warning method further includes: randomly sampling the probability distribution of random factors during equipment movement to generate position samples at each time step; combining a kinematic sub-model with each sample point to simulate the equipment's movement trajectory from the current moment to a future time window, statistically analyzing the movement trajectories of all sample points, calculating the probability of the equipment appearing at each position, and generating a probability distribution of the equipment's movement path; extracting features from historical trajectory data of personnel, including position, speed, acceleration, and direction of movement, and using an attention-based neural network model to model the personnel's trajectory, calculating the importance weight of each time step, and weighting the key path points in the personnel's trajectory based on the attention weight; using the weighted features as input, predicting the personnel's movement path within a future time window through a neural network model, and generating a probability distribution of the personnel's movement path; superimposing the probability distribution of the equipment's movement path with the probability distribution of the personnel's movement path, and calculating the intersection probability of the two at each position; determining the intersection probability using a preset probability threshold, marking the region probability according to the determined probability result, and generating a movement path probability cloud map.

[0056] In one possible implementation, the tower operation safety early warning method further includes: establishing a risk assessment index system based on the spatial location, size, and shape of the intersection area, as well as the speed, direction, and type attributes of the corresponding target; using a fuzzy logic algorithm to comprehensively evaluate the risk assessment indexes, classifying them into at least three risk levels: high, medium, and low; and generating corresponding operation early warning signals based on the risk levels, including the early warning level, early warning time, early warning location, and response measures.

[0057] In one possible implementation, the tower operation safety early warning method further includes: collecting signal attenuation patterns of typical tower types and constructing a steel structure multipath error compensation library; based on the steel structure multipath error compensation library, matching the characteristics of the operation area in real time, correcting error points, and resetting the target's position coordinates; and predicting the intersection area based on the reset target position coordinates.

[0058] The tower operation safety early warning system provided in this embodiment of the invention can execute the tower operation safety early warning method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0059] Although this application makes various references to certain modules in the system according to the embodiments of this application, any number of different modules can be used and run on user terminals and / or servers. The various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of each functional unit are only for easy distinction between each other and are not used to limit the scope of protection of this invention.

[0060] The specific embodiments described above do not constitute a limitation on the scope of protection of this application. Those skilled in the art should understand that various modifications, combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A safety warning system for tower work, characterized in that, The application relates to a method for constructing a dynamic three-dimensional work scene model and a method for predicting a work risk. The method comprises the following steps: a data acquisition unit is used for acquiring Beidou three-dimensional coordinate data of each work target in real time, and acquiring panoramic images of a work site through aerial photography of a UAV; a work scene model establishing unit is used for fusing the three-dimensional coordinate data and the panoramic images of the work site through space-time registration, and establishing a dynamic three-dimensional work scene model containing all targets; a prediction model establishing unit is used for tracking a motion trajectory of each work target based on time sequence coordinate data, establishing a prediction model fusing kinematics rules and machine learning, and predicting an intersection area between target motion paths; 2. The tower operation safety warning system according to claim 1, characterized in that, a risk level judging unit is used for judging a risk level according to a space position of the intersection area and attribute features of corresponding targets, and generating work warning information. The work scene model establishing unit executes the following steps: a mapping relationship between a UAV camera coordinate system and a Beidou three-dimensional coordinate system is established; based on the mapping relationship, the three-dimensional coordinate data and the panoramic images of the work site are spatially aligned, and tower material structure feature points extracted from the UAV images are point cloud registered with a pre-stored tower BIM model; 3. The tower operation safety warning system according to claim 2, characterized in that, based on the point cloud registered coordinates, the three-dimensional coordinate data is fused into the point cloud registered coordinates according to the coordinate mapping relationship, and a dynamic three-dimensional work scene model containing all targets is established. The work scene model establishing unit executes the following steps: UAV POS data including longitude, latitude, altitude, pitch angle and yaw angle are acquired, and the UAV POS data and image data are bound on the same timestamp to ensure time synchronization; 4. The iron tower operation safety early warning system according to claim 3, characterized in that, based on the UAV POS data, a conversion matrix and a translation vector of image pixel coordinates and Beidou world coordinates are constructed through perspective projection transformation, and the mapping relationship is determined. The work scene model establishing unit executes the following steps: tower material structure feature points extracted from the UAV images are converted into the Beidou world coordinate system by using the mapping relationship; corresponding feature points are extracted from the BIM model; 5. The iron tower operation safety early warning system according to claim 4, characterized in that, image feature points and BIM model feature points are aligned through a point cloud registration algorithm, and point cloud registration of the UAV image features and the pre-stored tower BIM model is completed. The work scene model establishing unit executes the following steps: rotation and translation parameters are initialized by using the symmetry features of the tower; the spatial distance between the image feature points and the BIM model points is calculated; 6. The tower work safety warning system of claim 1, wherein, based on the rotation and translation parameters and the spatial distance, the registration accuracy is iteratively optimized in combination with the curvature similarity of the steel structure until the target matching error is reached, and the point cloud registration is determined. The prediction model establishing unit executes the following steps: a kinematics sub-model is constructed according to device physical parameters, and a rigid body motion trajectory is output; a machine learning sub-model is constructed through a long short-term memory network, and personnel behavior patterns are learned; 7. The tower work safety warning system of claim 6, wherein, the kinematics sub-model and the machine learning sub-model are superposed to construct a prediction model, which is used for outputting a motion path probability cloud diagram in a future time window. The prediction model establishing unit executes the following steps: random factors in a device motion process are randomly sampled to generate position samples at each time step; The kinematic sub-model is combined with each sample point to simulate the motion trajectory of the device from the current time to the future time window, statistical analysis is performed on the motion trajectories of all sample points, the appearance probability of the device at each position is calculated, and a motion path probability distribution of the device is generated; Features are extracted from the historical trajectory data of the personnel, including position, speed, acceleration, and motion direction, a neural network model based on an attention mechanism is used to model the personnel trajectory, the importance weight of each time step is calculated, and the key path points in the personnel trajectory are weighted based on the attention weight; The weighted features are used as input to predict the motion path of the personnel in the future time window through a neural network model, and a motion path probability distribution of the personnel is generated; The motion path probability distribution of the device and the motion path probability distribution of the personnel are superimposed to calculate the intersection probability of the two at each position; The intersection probability is determined by using a preset probability threshold, the region probability is marked according to the determination probability result, and a motion path probability cloud map is generated.

8. The tower work safety warning system of claim 1, wherein, The steps performed by the risk level determination unit include: According to the spatial position, size, shape of the intersection area, and the speed, direction, type attribute characteristics of the corresponding target, a risk assessment index system is established; Fuzzy logic algorithm is used to comprehensively judge the risk assessment index, and at least three risk levels of high, medium and low are divided; According to the risk level, a corresponding operation warning signal is generated, including warning level, warning time, warning position and response measures.

9. The tower work safety warning system of claim 1, wherein, It also includes: Collecting the signal attenuation mode of typical tower type, constructing a steel structure multi-path error compensation library; Based on the steel structure multi-path error compensation library, real-time matching of the operation area characteristics is performed, error point correction is performed, and the position coordinates of the target are reset; Based on the reset target position coordinates, the intersection area is predicted.

10. A method for early warning of tower operation safety, characterized in that, The method is applied to the iron tower operation safety warning system of any one of claims 1-9, and the method comprises: Real-time acquisition of Beidou three-dimensional coordinate data of each operation target, synchronous acquisition of operation site panoramic image through unmanned aerial vehicle aerial photography; Through space-time registration, the three-dimensional coordinate data and the operation site panoramic image are fused to establish a dynamic three-dimensional operation scene model containing all targets; Based on the time sequence coordinate data, the motion trajectory of each operation target is tracked, a prediction model combining kinematic rules and machine learning is established, and the intersection area between the target motion paths is predicted; According to the spatial position of the intersection area and the attribute characteristics of the corresponding target, the risk level is determined, and operation warning information is generated.

Images