Construction equipment safety early warning and prediction method based on dynamic perception of unmanned aerial vehicle

Abstract

The invention provides a construction equipment safety early warning and prediction method based on unmanned aerial vehicle dynamic perception. The method comprises the following steps that real-time video data of a construction site and high-precision pose data of an unmanned aerial vehicle are collected through an inspection unmanned aerial vehicle; processing the video data, identifying construction equipment, key components and constructors, and obtaining geographic coordinates of the construction equipment, the key components and the constructors; generating a dynamic electronic fence changing along with the operation state based on the key component; predicting a future movement track of the construction equipment and analyzing track abnormity in combination with historical data and task types of the equipment; calculating a collision risk level between the construction equipment, and optimizing risk judgment in combination with an out-of-range overlapping degree; according to different construction equipment types, dynamically planning an unmanned aerial vehicle inspection path through a differential inspection strategy; and risk analysis is performed in combination with the dynamic electronic fence and the prediction result, construction equipment related safety early warning is generated, and the safety management level of the construction site can be improved through the method.

Description

Technical Field

[0001] This invention relates to the field of construction safety, specifically to a method for early warning and prediction of construction equipment safety based on dynamic perception by unmanned aerial vehicles (UAVs). Background Technology

[0002] In large-scale construction sites in fields such as construction, transportation, and energy, construction equipment often operates under high intensity and involves multiple devices working in coordination, accompanied by a large number of construction workers moving around, posing a severe challenge to safety management. Traditional construction safety monitoring mainly relies on two methods: manual inspection and fixed video surveillance. Manual inspection is limited by personnel physical strength, field of vision, and inspection frequency, making it difficult to achieve real-time full coverage of a vast construction site. Especially in scenarios with dense equipment and dispersed work areas, blind spots are prone to occur, resulting in the inability to promptly detect malfunctions in critical equipment components or personnel accidentally entering dangerous areas. Fixed video surveillance, due to its fixed installation location, cannot adjust its monitoring range to follow the dynamic operation trajectory of construction equipment, and it is difficult to accurately obtain the high-precision geographical coordinates of equipment, critical components, and personnel, failing to meet the monitoring requirements for spatial accuracy in complex construction scenarios.

[0003] Existing safety early warning systems for construction equipment also have significant shortcomings: Firstly, most systems use static electronic fences to delineate safety zones, failing to consider the dynamic impact of construction equipment's operating status on safety boundaries. This results in poor adaptability of the electronic fences, making them prone to false or missed warnings. Secondly, existing early warning methods often rely on simple risk assessment based on real-time distance, lacking integrated analysis of historical equipment operating data and current task types. This makes it difficult to accurately predict future equipment trajectories and identify risks such as trajectory deviations and potential collisions between equipment in advance. Furthermore, some systems using drones for inspection fail to optimize inspection paths based on the dynamic risk distribution of the construction site, leading to low drone inspection coverage efficiency and failing to maximize their dynamic perception advantages. These issues collectively result in insufficient timeliness, accuracy, and comprehensiveness of safety early warnings for construction sites, seriously threatening the lives of workers and construction progress.

[0004] Therefore, there is an urgent need to provide a construction equipment safety early warning and prediction method based on UAV dynamic perception to solve problems such as insufficient traditional monitoring coverage, low early warning accuracy, weak trajectory prediction capability and lack of inspection path optimization, and improve the level of construction site safety management. Summary of the Invention

[0005] The purpose of this invention is to solve the technical problems mentioned above, and to propose a method for safety early warning and prediction of construction equipment based on UAV dynamic perception, including the following steps: S1. Collect real-time video data of the construction site and high-precision pose data of the drone itself through inspection drones; S2. Process video data, identify construction equipment, key components and construction personnel, and obtain the geographical coordinates of construction equipment, key components and construction personnel; S3. Generate dynamic electronic fences that change with the operation status based on key components; S4. Based on historical equipment data and task type, predict the future movement trajectory of construction equipment and analyze trajectory anomalies; S5. Calculate the collision risk level between construction equipment and optimize the risk assessment based on the degree of overlap beyond the permitted range; S6. Based on different types of construction equipment, dynamically plan the inspection path of the drone through differentiated inspection strategies; S7. Combine dynamic electronic fences with prediction results to conduct risk analysis and generate safety warnings related to construction equipment.

[0006] In the preferred embodiment, step S3, the step of generating a dynamic electronic fence, includes: S31. Determine the geographical coordinates of the rotation center of the construction equipment through the key point detection model; S32. Identify the geographic coordinates of the tip of the boom of the construction equipment; S33. Calculate the real-time distance d between the center of rotation and the tip of the boom; S34. With the center of rotation as the center, and... A circular dynamic electronic fence is generated with radius [r, y], where [r, y] This is a preset safety margin.

[0007] In the preferred embodiment, in step S33, the real-time distance is calculated using the great circle distance formula: The formula for great circle distance is: ; in, The average radius of the Earth; , These are the longitude and latitude of the center of rotation of the construction equipment. , These are the longitude and latitude of the boom tip, respectively. The difference in longitude between the two key points.

[0008] In the preferred embodiment, step S4 further includes the following steps: S410. Based on the historical location and speed data of the construction equipment, construct the motion state vector of the construction equipment; identify the current task type of the construction equipment; when the task type is a special task, supplement the motion state vector with the estimated task duration, cooperating equipment ID and pre-positioning coordinate parameters of the overlapping area. S420. Using a time-series prediction model, the location of construction equipment at several points in time within a specific future period is predicted to form a predicted trajectory. S430. Calculate the distance between the predicted trajectories of different construction equipment at various future time points, determine the minimum predicted distance and expected collision time between the predicted trajectories, and calculate the collision risk level. When the collision risk level exceeds the set threshold, trigger a collision warning. S440. Compare the predicted trajectory of the construction equipment with the preset original work trajectory, and calculate the trajectory deviation. , ,when If the deviation exceeds the preset threshold, it is determined to be a trajectory deviation risk, and the risk type and level analysis process is initiated. S450. For specific tasks, calculate the activity frequency of the device in the overlapping region: Determine the start time of construction equipment entering the overlapping area and the end time of its departure from the overlapping area; Calculate the time it takes for the construction equipment to reach key points on the trajectory; Output the collision avoidance time window between construction equipment.

[0009] In the preferred embodiment, the formula for calculating the collision risk level in step S430 is as follows: ; in, Collision risk level, The minimum predicted distance is TTC, the predicted collision time is TTC, and the weighting coefficient is [missing information]. Overlap parameter ; The risk type and level process includes: The trajectory deviates from the risk level. The risk level is defined as ∈[5,10) meters. The risk level is defined as ∈[10,20) meters. ≥20 meters is classified as Level 3 risk; The risk level for exceeding the boundary is as follows: Level 1 risk is defined as the distance exceeding the boundary ∈ [1,3) meters, Level 2 risk is defined as the distance exceeding the boundary ∈ [3,5) meters, and Level 3 risk is defined as the distance exceeding the boundary ≥ 5 meters. Collision risk level is calculated using the collision risk level formula: when When the value is greater than 0.8, a level-two warning is triggered; when When the value is greater than 1.2, a Level 1 warning is triggered.

[0010] In the preferred embodiment, the types of construction equipment in step S6 include high-altitude rotating equipment, ground operation equipment, mobile operation equipment, and manned operation equipment. The dimensions of the differentiated inspection strategy include: inspection path shape, hovering point density, inspection height, key coverage area, and inspection speed. Each dimension is strongly linked to the equipment's operating characteristics.

[0011] In the preferred solution, the differentiated inspection strategy is as follows: For high-altitude rotating equipment: the inspection path is segmented and reciprocating along the length of the boom, with hovering points set according to the distribution of key risk points on the boom, and the hovering time is adapted to the identification requirements of key components; the inspection height avoids obstructing the boom luffing mechanism to ensure a clear monitoring view; and high-frequency fault risk points are given priority coverage. For ground-based equipment: the inspection path covers the core operating area of ​​the equipment in a grid pattern, and the density of hovering points is adapted to the bucket operating radius and track travel range; dedicated hovering points are added at the bucket lifting limit position to enhance the monitoring of key actions; the inspection perspective is close to the working surface to ensure clear capture of the bucket's movement and the contact state with the ground; Mobile operation equipment: The inspection path is dynamically adjusted to follow the equipment's operation line, prioritizing coverage of safety-sensitive areas; the inspection speed is adapted to the equipment's movement speed to avoid identification errors caused by excessive flight speed; fixed inspection points are set up at the unloading point to ensure that the entire unloading process is captured; For manned operation equipment: the inspection path is circular, surrounding the platform's lifting channel and the area where workers stand, and the inspection height is dynamically matched with the platform's real-time operating height; the lateral coverage area covers the personnel safety boundary, ensuring that there are no blind spots in the dynamic monitoring of workers and the equipment's lifting mechanism.

[0012] In the preferred scheme, when there are two or more different types of equipment operating simultaneously at the construction site, the drone inspection path is planned according to the risk priority of the equipment. The risk priority is: manned operation equipment > high-altitude rotating equipment > ground operation equipment > mobile operation equipment. The inspection frequency of high-priority equipment is higher than that of low-priority equipment. For the same priority equipment, the inspection density increases according to the degree of overlap of the operation area, and the inspection points are more densely distributed in the overlapping area.

[0013] In the preferred embodiment, a dedicated coverage section for collaborative operation is set up along the drone inspection path for the collaborative operation equipment combination; The dedicated coverage area for collaborative operations must meet the following requirements: The coverage area includes the overlapping areas of equipment operations and the areas where coordinated actions are connected; The inspection interval is adapted to the accuracy requirements of collaborative operations, and the number of hovering points is matched with the number of collaborative devices to ensure that the key actions of each collaborative device can be captured. The hovering point view simultaneously covers the movement trajectories of key components of all collaborative devices, ensuring the continuity of collaborative actions and safety monitoring.

[0014] In the preferred embodiment, during step S6, when dynamically planning the UAV inspection path, the inspection strategy is adjusted synchronously based on the current operating status of the associated equipment. When the equipment is under load, extend the dwell time at the hover point to improve the accuracy of key component status recognition; When the equipment is in motion, the inspection path following speed is increased simultaneously to ensure full coverage of the movement route; When equipment is in collaborative operation mode, increase the inspection frequency of the dedicated coverage area for collaborative operation and strengthen the risk detection of collaborative actions.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) By identifying key components of the equipment in real time and calculating their real-time distance from the rotation center, the present invention can generate a dynamic electronic fence that changes with the working status, which completely changes the traditional fixed area monitoring mode, making the delineation of dangerous areas accurately match the actual physical extension range of the equipment, effectively avoiding monitoring blind spots and invalid alarms, and greatly improving the accuracy of safety warnings.

[0016] (2) This invention not only analyzes the current position of the equipment, but also integrates historical data and task type to predict its future trajectory. By calculating the collision time, minimum distance, and deviation from the preset trajectory between predicted trajectories, collision risks and trajectory anomalies can be quantified in advance, achieving risk-level early warning. This provides managers with a valuable reaction time window, preventing safety accidents from occurring in the first place.

[0017] (3) This invention dynamically plans differentiated UAV inspection paths for different types of equipment and their operating states, including path shape, hovering point, speed and viewing angle. This strategy achieves optimal allocation of monitoring resources, ensures that high-risk equipment and key actions are covered effectively and efficiently, and can intelligently schedule inspection resources according to risk priority in multi-equipment cross-operation scenarios, thereby comprehensively improving overall monitoring efficiency and system adaptability. Attached Figure Description

[0018] Figure 1 This is a flowchart of a construction equipment safety early warning and prediction method based on UAV dynamic perception.

[0019] Figure 2 This is a flowchart of the method in Example 2. Detailed Implementation

[0020] Example 1 A method for safety early warning and prediction of construction equipment based on UAV dynamic perception, such as Figure 1 As shown, it includes the following steps: S1. Collect real-time video data of the construction site and high-precision pose data of the drone itself through inspection drones; S2. Process video data, identify construction equipment, key components and construction personnel, and obtain the geographical coordinates of construction equipment, key components and construction personnel; S3. Generate dynamic electronic fences that change with the operation status based on key components; S4. Based on historical equipment data and task type, predict the future movement trajectory of construction equipment and analyze trajectory anomalies; S5. Calculate the collision risk level between construction equipment and optimize the risk assessment based on the degree of overlap beyond the permitted range; S6. Based on different types of construction equipment, dynamically plan the inspection path of the drone through differentiated inspection strategies; S7. Combine dynamic electronic fences with prediction results to conduct risk analysis and generate safety warnings related to construction equipment.

[0021] Preferably, step S3, the step of generating a dynamic electronic fence, includes: S31. Determine the geographical coordinates of the rotation center of the construction equipment through the key point detection model; S32. Identify the geographic coordinates of the tip of the boom of the construction equipment; S33. Calculate the real-time distance d between the center of rotation and the tip of the boom; S34. With the center of rotation as the center, and... A circular dynamic electronic fence is generated with radius [r, y], where [r, y] This is a preset safety margin.

[0022] Preferably, in step S33, the real-time distance is calculated using the great circle distance formula: The formula for great circle distance is: ; in, The average radius of the Earth; , These are the longitude and latitude of the center of rotation of the construction equipment. , These are the longitude and latitude of the boom tip, respectively. The difference in longitude between the two key points.

[0023] Preferably, step S4 further includes the following steps: S410. Based on the historical location and speed data of the construction equipment, construct the motion state vector of the construction equipment; identify the current task type of the construction equipment; when the task type is a special task, supplement the motion state vector with the estimated task duration, cooperating equipment ID and pre-positioning coordinate parameters of the overlapping area. S420. Using a time-series prediction model, the location of construction equipment at several points in time within a specific future period is predicted to form a predicted trajectory. S430. Calculate the distance between the predicted trajectories of different construction equipment at various future time points, determine the minimum predicted distance and expected collision time between the predicted trajectories, and calculate the collision risk level. When the collision risk level exceeds the set threshold, trigger a collision warning. S440. Compare the predicted trajectory of the construction equipment with the preset original work trajectory, and calculate the trajectory deviation. , ,when If the deviation exceeds the preset threshold, it is determined to be a trajectory deviation risk, and the risk type and level analysis process is initiated. S450. For specific tasks, calculate the activity frequency of the device in the overlapping region: Determine the start time of construction equipment entering the overlapping area and the end time of its departure from the overlapping area; Calculate the time it takes for the construction equipment to reach key points on the trajectory; Output the collision avoidance time window between construction equipment.

[0024] Preferably, in step S430, the formula for calculating the collision risk level is: ; in, Collision risk level, The minimum predicted distance is TTC, the predicted collision time is TTC, and the weighting coefficient is [missing information]. Overlap parameter ; The risk type and level process includes: The trajectory deviates from the risk level. The risk level is defined as ∈[5,10) meters. The risk level is defined as ∈[10,20) meters. ≥20 meters is classified as Level 3 risk; The risk level for exceeding the boundary is as follows: Level 1 risk is defined as the distance exceeding the boundary ∈ [1,3) meters, Level 2 risk is defined as the distance exceeding the boundary ∈ [3,5) meters, and Level 3 risk is defined as the distance exceeding the boundary ≥ 5 meters. Collision risk level is calculated using the collision risk level formula: when When the value is greater than 0.8, a level-two warning is triggered; when When the value is greater than 1.2, a Level 1 warning is triggered.

[0025] Preferably, the types of construction equipment in step S6 include high-altitude rotating equipment, ground operation equipment, mobile operation equipment, and manned operation equipment, and the dimensions of the differentiated inspection strategy include: inspection path shape, hovering point density, inspection height, key coverage area, and inspection speed, with each dimension strongly tied to the equipment's operating characteristics.

[0026] Preferably, the differentiated inspection strategy is as follows: For high-altitude rotating equipment: the inspection path is segmented and reciprocating along the length of the boom, with hovering points set according to the distribution of key risk points on the boom, and the hovering time is adapted to the identification requirements of key components; the inspection height avoids obstructing the boom luffing mechanism to ensure a clear monitoring view; and high-frequency fault risk points are given priority coverage. For ground-based equipment: the inspection path covers the core operating area of ​​the equipment in a grid pattern, and the density of hovering points is adapted to the bucket operating radius and track travel range; dedicated hovering points are added at the bucket lifting limit position to enhance the monitoring of key actions; the inspection perspective is close to the working surface to ensure clear capture of the bucket's movement and the contact state with the ground; Mobile operation equipment: The inspection path is dynamically adjusted to follow the equipment's operation line, prioritizing coverage of safety-sensitive areas; the inspection speed is adapted to the equipment's movement speed to avoid identification errors caused by excessive flight speed; fixed inspection points are set up at the unloading point to ensure that the entire unloading process is captured; For manned operation equipment: the inspection path is circular, surrounding the platform's lifting channel and the area where workers stand, and the inspection height is dynamically matched with the platform's real-time operating height; the lateral coverage area covers the personnel safety boundary, ensuring that there are no blind spots in the dynamic monitoring of workers and the equipment's lifting mechanism.

[0027] Preferably, when two or more different types of equipment are operating simultaneously at the construction site, the UAV inspection path is planned according to the risk priority of the equipment, with the risk priority being: manned operation equipment > high-altitude rotation equipment > ground operation equipment > mobile operation equipment; the inspection frequency of high-priority equipment is higher than that of low-priority equipment, and the inspection density of the same priority equipment increases with the degree of overlap of the operation area, and the inspection points are more densely distributed in the overlapping area.

[0028] Preferably, for combinations of collaborative operation equipment, the drone inspection path is set with a dedicated coverage section for collaborative operations; The dedicated coverage area for collaborative operations must meet the following requirements: The coverage area includes the overlapping areas of equipment operations and the areas where coordinated actions are connected; The inspection interval is adapted to the accuracy requirements of collaborative operations, and the number of hovering points is matched with the number of collaborative devices to ensure that the key actions of each collaborative device can be captured. The hovering point view simultaneously covers the movement trajectories of key components of all collaborative devices, ensuring the continuity of collaborative actions and safety monitoring.

[0029] Preferably, in step S6, when dynamically planning the UAV inspection path, the inspection strategy is adjusted synchronously based on the current operating status of the associated equipment: When the equipment is under load, extend the dwell time at the hover point to improve the accuracy of key component status recognition; When the equipment is in motion, the inspection path following speed is increased simultaneously to ensure full coverage of the movement route; When equipment is in collaborative operation mode, increase the inspection frequency of the dedicated coverage area for collaborative operation and strengthen the risk detection of collaborative actions.

[0030] Preferably, the dedicated coverage segment for collaborative operations quantifies the risk level of the collaborative operation area, providing a basis for allocating drone inspection resources. The specific formula is: ; in, The comprehensive risk weight for the dedicated coverage segment of collaborative operations; n is the total number of devices involved in collaborative operations; The inherent risk level of the i-th device; The collaborative importance of the i-th device; Risk coefficients for overlapping areas: no overlap = 1.0, partial overlap = 1.2-1.5, complete overlap = 1.8-2.0.

[0031] In another approach, a differentiated inspection strategy dynamically adjusts the inspection cycle, frequency, and route based on equipment risk and operational characteristics. Specifically, this includes: The basic inspection cycle for a single piece of equipment is determined based on the risk level and equipment type. The higher the risk, the shorter the cycle. The formula is as follows: ; in, The inspection cycle for a certain type of drone equipment, with a range of 5-30 minutes; The baseline inspection cycle refers to the basic cycle under no-risk conditions, and is generally taken as 10 minutes. For equipment type coefficient; The real-time risk level of the equipment is: High = 3, Medium = 2, Low = 1; This refers to the equipment operating status coefficient.

[0032] Based on the basic cycle, the inspection frequency is adjusted according to seasonal disturbances and site conditions, as shown in the following formula: ; in, The adjusted inspection frequency; Basic inspection frequency; m: number of influencing factors; is the frequency adjustment coefficient for the j-th factor.

[0033] Preferably, the differentiated inspection strategy includes a path optimization objective function, specifically: In multi-device scenarios, the formula for optimizing drone inspection paths, balancing adequate risk coverage and inspection efficiency, is as follows: ; in, The comprehensive cost of the inspection route, including time and energy consumption; Let be the inspection distance from the drone to the i-th device; Cost per unit distance for drones, including energy consumption and depreciation; The risk coverage weighting coefficient has a value of 1.2-1.5, emphasizing risk priority. Let the risk weight be that of the i-th device; Let be the risk coverage area of ​​the drone on the i-th device.

[0034] Another approach also includes an avoidance mechanism that adjusts the operation of construction equipment based on the behavioral intentions of construction personnel. The mechanism is implemented in the following steps: S81, dynamically collect the location sequence, movement vector, and site reporting information of construction personnel through drones, and combine them with a preset behavioral intention reasoning model to determine whether the personnel's behavioral intentions are clear; S82, when it is determined that the personnel's behavioral intentions are unclear, and when it is detected by high-precision geographic coordinates that the personnel have entered the dynamic electronic fence of any construction equipment without reporting as required, the equipment emergency avoidance process is triggered; S83, the construction equipment automatically adjusts its operation behavior to avoid collision risks according to its own type and current working status, and at the same time triggers site sound and light warnings and management personnel terminal alarms.

[0035] In step S81, the criteria for determining whether the behavioral intent is clear are as follows: The behavioral intent is clear: the matching degree between the personnel movement vector and the preset task target is greater than or equal to the set threshold, and the operation has been reported through the site terminal; Unclear behavioral intent: The personnel movement vector has no clear target direction, that is, the matching degree is less than the set threshold, or the speed of entering the dynamic electronic fence without reporting is greater than or equal to the set value, or the time spent in the fence exceeds the time required for normal operation; In step S82, the determination of "unauthorized entry" is: the personnel have not completed the work report for entering the protection range of the equipment in the site management system, and the geographical coordinates are within the dynamic electronic fence for more than 1 second.

[0036] In step S83, the specific emergency avoidance actions for different types of construction equipment are as follows: Aerial work platform: Immediately stop lifting and extension operations, lock the platform railings, and simultaneously lower the platform slowly to a safe height of ≤2 meters from the ground, while continuously emitting audible and visual alarms; Tower crane: Immediately stop the boom slewing, luffing and hook raising and lowering operations, slowly return the hook to the safe area directly below the center of rotation, and keep the boom stationary until personnel are evacuated; Tracked excavators: Immediately stop the bucket movement and slewing bearing rotation, raise the bucket to a safe height of ≥1.5 meters above the ground, lock the track brakes, and prevent accidental triggering of the traveling action; Wheel loader: Immediately stop the bucket opening and closing and steering actions, return the bucket to the horizontal closed position, slowly brake to a stop, and maintain a safe distance of ≥3 meters from intruders until personnel are evacuated.

[0037] Example 2 like Figure 2 As shown, a method for early warning and prediction of construction equipment safety based on UAV dynamic perception includes the following steps: S1. Collect real-time video data of the construction site and high-precision pose data of the drone itself through inspection drones; S2. Process video data, identify construction equipment, key components and construction personnel, and obtain high-precision geographic coordinates of construction equipment, key components and construction personnel; S3. Generate dynamic electronic fences that change with the operation status based on key components; S4. Based on historical equipment data and task type, predict the future movement trajectory of construction equipment and analyze trajectory anomalies; S5. Calculate the collision risk level between construction equipment and optimize the risk assessment based on the degree of overlap beyond the permitted range; S6. Dynamically plan the UAV inspection path with the goal of maximizing the coverage value function; S7. Combine dynamic electronic fences with prediction results to conduct risk analysis and generate safety warnings related to construction equipment.

[0038] Preferably, step S3, the step of generating a dynamic electronic fence, includes: S31. Determine the geographical coordinates of the rotation center of the construction equipment through the key point detection model; S32. Identify the geographic coordinates of the tip of the boom of the construction equipment; S33. Calculate the real-time distance d between the center of rotation and the tip of the boom; S34. With the center of rotation as the center, and... A circular dynamic electronic fence is generated with radius [r, y], where [r, y] This is a preset safety margin.

[0039] Preferably, in step S33, the real-time distance is calculated using the great circle distance formula: The formula for great circle distance is: ; in, The average radius of the Earth; , These are the longitude and latitude of the center of rotation of the construction equipment. , These are the longitude and latitude of the boom tip, respectively. The difference in longitude between the two key points.

[0040] Preferably, step S4 further includes the following steps: S410. Based on the historical location and speed data of the construction equipment, construct the motion state vector of the construction equipment; identify the current task type of the construction equipment; when the task type is a special task, supplement the motion state vector with the estimated task duration, cooperating equipment ID and pre-positioning coordinate parameters of the overlapping area. S420. Using a time-series prediction model, the location of construction equipment at several points in time within a specific future period is predicted to form a predicted trajectory. S430. Calculate the distance between the predicted trajectories of different construction equipment at various future time points, determine the minimum predicted distance and expected collision time between the predicted trajectories, and calculate the collision risk level. When the collision risk level exceeds the set threshold, trigger a collision warning.

[0041] Preferably, step S4 further includes the following steps: S440. Compare the predicted trajectory of the construction equipment with the preset original work trajectory, and calculate the trajectory deviation. , ,when If the deviation exceeds the preset threshold, it is determined to be a trajectory deviation risk, and the risk type and level analysis process is initiated. S450. For specific tasks, calculate the activity frequency of the device in the overlapping region: Determine the start time of construction equipment entering the overlapping area and the end time of its departure from the overlapping area; Calculate the time it takes for the construction equipment to reach key points on the trajectory; Output the collision avoidance time window between construction equipment.

[0042] Preferably, in step S430, the formula for calculating the collision risk level is: ; in, Collision risk level, The minimum predicted distance is TTC, the predicted collision time is TTC, and the weighting coefficient is [missing information]. Overlap parameter .

[0043] Preferably, the risk type and level process in step S440 includes: The trajectory deviates from the risk level. The risk level is defined as ∈[5,10) meters. The risk level is defined as ∈[10,20) meters. ≥20 meters is classified as Level 3 risk; The risk level for exceeding the boundary is as follows: Level 1 risk is defined as the distance exceeding the boundary ∈ [1,3) meters, Level 2 risk is defined as the distance exceeding the boundary ∈ [3,5) meters, and Level 3 risk is defined as the distance exceeding the boundary ≥ 5 meters. Collision risk level is calculated using the collision risk level formula: when When the value is greater than 0.8, a level-two warning is triggered; when When the value is greater than 1.2, a Level 1 warning is triggered.

[0044] Preferably, the coverage value function in step S6 is: ; Where V(P) is the coverage value function, and P is the coordinate of the point in the air. and The first The center point and risk weight of a dynamic electronic fence. This is the attenuation coefficient.

[0045] Preferably, step S6 further includes the following step: S61. Calculate the dynamic risk coverage map of the site based on the real-time location and range of the dynamic electronic fence and the preset risk weight. S62. With the goal of maximizing the coverage value function V(P), dynamically plan the flight trajectory, hovering position, and monitoring angle of the inspection drone; It also includes steps parallel to S62: S63. At the same time, by covering the equipment operation path with dynamic trajectory, abnormal equipment conditions can be captured in real time, improving the efficiency of detecting abnormal equipment trajectory conditions. Abnormal conditions include: trajectory deviation from the preset path and sudden speed changes. Coverage value function At the same time, the deviation between the current working path of the associated equipment and the inspection path of the drone is correlated, and higher weight is given to equipment areas with frequent fluctuations in historical trajectories or those prone to anomalies.

[0046] Preferably, when the task type in S410 is a special task, the time for the device to enter or leave the overlapping area is calculated, and the collision avoidance time window is output.

[0047] This embodiment has the following beneficial effects: (1) By identifying the coordinates of key components such as the rotation center and boom tip of the construction equipment, and combining them with real-time distance calculation, a dynamic electronic fence that changes with the operation status is generated. The preset safety margin is integrated in sync, which effectively solves the problem of false and missed warnings caused by changes in the equipment operation trajectory of static fences, and ensures that the safety boundary is accurately matched with the actual operation range of the equipment.

[0048] (2) Combining the historical location, speed data and current task type of the equipment, the future trajectory of the equipment is generated by the time-series prediction model. It can not only calculate the minimum predicted distance and the expected collision time between the equipment, but also determine the risk of trajectory deviation and range exceeding the boundary. It can also achieve graded early warning by quantifying the collision risk level, breaking through the limitation of relying solely on real-time distance to determine the risk, and identifying potential safety hazards in advance.

[0049] (3) With the goal of maximizing the coverage value function, the drone flight trajectory, hovering position and monitoring perspective are dynamically planned by combining the real-time location of the dynamic electronic fence, risk weight and equipment operation path. At the same time, higher weights are given to areas with frequent historical trajectory fluctuations and prone to anomalies, so as to achieve accurate coverage of equipment operation path, improve the efficiency of detecting equipment trajectory anomalies and give full play to the advantages of drone dynamic perception.

[0050] (4) By integrating the entire process of UAV dynamic data collection, key component identification, dynamic fence generation, trajectory prediction, risk classification and inspection path optimization, it effectively solves the problems of insufficient coverage of traditional manual inspection, poor adaptability of fixed monitoring, low accuracy of existing early warning and lack of trajectory prediction, and provides systematic technical support for the safety of personnel at construction sites, stable operation of equipment and construction progress.

[0051] Example 3 This embodiment provides a construction safety early warning and prediction method based on UAV dynamic perception, including the following steps: S1. Real-time video data of the construction site and high-precision position and attitude data of the drone itself are collected by the sensors carried by the inspection drone. S2. Process the real-time video data to identify at least one construction device and key components of the construction device in the site, and identify the construction personnel in the site, while obtaining their high-precision geographic coordinates. S3. Based on the identified key components of the construction equipment, dynamically calculate the real-time operating range of the construction equipment and generate a dynamic electronic fence that changes with the operating status of the equipment. S4. Based on the location, movement vector, and preset context information of construction personnel, predict the future movement trajectory or behavioral intention of construction personnel; based on the historical movement data of construction equipment, predict the future movement trajectory of construction equipment. S5. Based on the dynamic electronic fence and the future movement trajectory or behavioral intention predicted in the prediction step S4, perform spatiotemporal relationship analysis, and generate early warning information when potential risks are determined to exist.

[0052] Preferably, the specific steps for generating the dynamic electronic fence in step S3 include: S31. Determine the geographic coordinates of the rotation center of the construction equipment using a key point detection model. ; S32. Identify the geographical coordinates of the boom tip of the construction equipment. ; S33. Calculate the real-time distance d between the center of rotation and the tip of the boom; S34. With the center of rotation as the center, and... A circular dynamic electronic fence is generated with radius [r, y], where [r, y] This is a preset safety margin.

[0053] Preferably, in step S33, the real-time distance The calculation is obtained using the great circle distance formula: The formula for great circle distance is: ; in, The average radius of the Earth; , These are the longitude and latitude of the center of rotation of the construction equipment. , These are the longitude and latitude of the boom tip, respectively. The difference in longitude between the two key points.

[0054] Preferably, the specific steps in step S4 for predicting the behavioral intentions of construction workers include: S41. Obtain the real-time location sequence of construction workers and calculate their movement vectors. and the direction angle of movement θ; S42, Move the vector The system matches the movement of construction workers with several points of interest (POIs) on a pre-set site map, and calculates the consistency score between the movement direction of the construction workers and the direction of each POI. ; S43. Combining the current task status information, distance information from points of interest, and historical behavior pattern information of construction personnel, the confidence level of behavioral intention to go to each point of interest is calculated through weighted fusion. ; S44. Select confidence level The highest one or several behavioral intentions are used as the prediction results.

[0055] Preferably, when predicting the behavioral intentions of construction workers, the prediction dimensions are further refined through the following steps to improve the accuracy of intention prediction: S401. The current task status information of the personnel is further refined into the current task list, task order and task duration of the personnel; wherein, the task list is a list of tasks to be performed by the personnel, the task order is a preset logical sequence of tasks to be performed, and the task duration is the estimated time for each task to be performed. S402. Based on the movement direction angle θ of the construction personnel, match the construction equipment within the coverage area of ​​the movement direction angle and generate at least one location-related event; the location-related event refers to the behavioral event that matches the construction site scene, including: going to the target equipment for collaborative work, going to the facility to pick up parts, and going to the designated area to rest. Add location-related events as new points of interest to the preset list of points of interest on the site map; S403. When calculating the confidence level of a construction worker's intention to travel to each point of interest, the matching score between location-related events and points of interest is included in the weighted fusion calculation. The formula for calculating the confidence level of intention is as follows: ; in, From current location to point of interest The distance; The score is based on task relevance, and task relevance is calculated by combining the task list, task order, and task duration in S401. For scores based on historical behavioral patterns, Location-related events and points of interest Match score, , , , Let be the weight coefficient, and satisfy... ; S404. Based on the corrected intent confidence, select the 1 to 3 intents with the highest confidence as the final prediction results, and simultaneously output the location-related events and detailed task information as supplementary explanations of the prediction results to the risk assessment module.

[0056] Preferably, the confidence level of behavioral intent in step S43 The calculation formula is: ; in, This is the current location and point of interest of the construction workers. distance, It is a score based on the current task status information of the construction workers. It is a score based on historical behavioral pattern information. to These are the weighting coefficients, and their sum is 1.

[0057] Preferably, the specific steps for predicting the future trajectory of the device in step S4 include: S410. Based on the historical location and speed data of the construction equipment, construct the motion state vector of the construction equipment; identify the current task type of the construction equipment; when the task type is a special task, supplement the motion state vector with the estimated task duration, cooperating equipment ID and pre-positioning coordinate parameters of the overlapping area. S420. Using a time series prediction model, predict the location of construction equipment at several time points within a future period [t+1, t+T] to form a predicted trajectory; S430. Calculate the distance between the predicted trajectories of different devices at various future time points, and determine the minimum predicted distance between the predicted trajectories. and the estimated time to collision (TTC); S440. Compare the predicted trajectory of the equipment with the preset original work trajectory, and calculate the trajectory deviation ΔL. When ΔL exceeds the preset deviation threshold, it is determined to be a trajectory deviation risk, and the risk type and level analysis process is initiated. S450, For special task scenarios, calculate the activity frequency of the device in the overlapping area: Determine the start time of construction equipment entering the overlapping area and the end time of its departure from the overlapping area; Calculate the time it takes for the construction equipment to reach key points on the trajectory; Output the collision avoidance time window between construction equipment.

[0058] Preferably, after step S430, the method further includes: S440, Based on the minimum predicted distance in step S430 Based on the estimated time to collision (TTC), the collision risk level is calculated using a risk quantification formula: ; in, For collision risk level, the weighting coefficient Overlap parameter ,when When the set threshold is exceeded, a collision warning is triggered.

[0059] Preferably, the risk type and level in step S440 are as follows: The trajectory deviates from the risk level as follows: ΔL∈[5,10) meters is Level 1 risk, ΔL∈[10,20) meters is Level 2 risk, and ΔL≥20 meters is Level 3 risk. The risk level for exceeding the boundary is as follows: Level 1 risk is defined as the distance exceeding the boundary ∈ [1,3) meters, Level 2 risk is defined as the distance exceeding the boundary ∈ [3,5) meters, and Level 3 risk is defined as the distance exceeding the boundary ≥ 5 meters. Collision risk levels are calculated using the original formula: Risk > 0.8 triggers a Level 2 warning, and Risk > 1.2 triggers a Level 1 warning.

[0060] Preferably, the method further includes step S0: S01. Calculate the dynamic risk coverage map of the site based on the real-time location and range of the dynamic electronic fence and the preset risk weight. S02. With the goal of maximizing the coverage value function (V(P)), the flight trajectory, hovering position and monitoring perspective of the inspection drone are dynamically planned in real time. Meanwhile, by dynamically covering the equipment's operating path, abnormal equipment conditions can be captured in real time, improving the efficiency of detecting abnormal equipment trajectories. Abnormal conditions include: trajectory deviation from the preset path and sudden speed changes. Among them, the covering value function P is the coordinate of a point in the air. and The first The center point and risk weight of a dynamic electronic fence. The attenuation coefficient is... At the same time, the deviation between the current working path of the associated equipment and the inspection path of the drone is correlated, and higher weight is given to equipment areas with frequent fluctuations in historical trajectories or those prone to anomalies.

[0061] Preferably, in step S4, when predicting the behavioral intentions of construction workers, the system also receives reporting information actively submitted by construction workers through the terminal, and uses the reporting information to correct or confirm the prediction results, so as to reduce false alarms and system computational load.

[0062] Preferably, in step S5, potential risks include personnel intrusion into the dynamic electronic fence, predicted collisions between devices, or construction personnel deviating from the intended task.

[0063] This embodiment also provides a construction safety early warning and prediction system for implementing the above-mentioned construction safety early warning and prediction method based on UAV dynamic perception, including: The data acquisition module includes the inspection drone and its onboard visible light camera, RTK positioning unit or IMU unit; The data processing and recognition module is used to process video data, identify devices, personnel and their key points based on computer vision models, and perform high-precision geolocation. The dynamic fence calculation module is used to generate and update dynamic electronic fences in real time based on the identified key points of the equipment; The predictive analysis module includes a construction worker behavior intention prediction unit and an equipment trajectory prediction unit, which are used to predict worker intentions and equipment trajectories respectively. The risk assessment and early warning module is used to perform spatiotemporal collision risk analysis based on dynamic electronic fences and prediction results, and generate graded early warning information. The early warning execution and feedback module is used to distribute early warning information to various terminals and support personnel feedback.

[0064] Preferably, the data processing and recognition module further includes: The target detection unit uses the YOLO V9 target detection algorithm to perform preliminary localization and bounding box annotation of construction equipment and personnel in video frames; The key point detection unit, based on a regression model of a convolutional neural network, locates key points in the detected construction equipment image and outputs pixel coordinates of key points including the rotation center and the tip of the boom. The coordinate mapping unit, combined with the high-precision pose data of the UAV provided by the data acquisition module, accurately maps the pixel coordinates of key points to the geographic coordinate system through collinearity equations.

[0065] Preferably, the mathematical model for coordinate transformation performed by the coordinate mapping unit is as follows: ; in,( ( ) are the geographic coordinates of the target, () represents the GNSS coordinates of the UAV. Here is the rotation matrix calculated based on the IMU attitude angle, where (u, v) are the pixel coordinates of the key points. , ) represents the coordinates of the principal point. Focal length This is a scaling factor.

[0066] Preferably, the construction worker behavior intention prediction unit in the predictive analysis module includes: The movement vector calculation subunit is used to calculate the movement direction and speed based on the position data of construction workers in consecutive frames; The point-of-interest (POI) map database stores the geographic coordinates and type information of each POI at the construction site. The intent reasoning engine calculates the confidence level of construction workers' intent to go to each point of interest based on a weighted fusion algorithm. The reporting information interface is used to receive travel information proactively reported by construction workers through mobile terminals and to calibrate the inference results of the construction workers' behavioral intentions.

[0067] Preferably, the equipment trajectory prediction unit in the predictive analysis module uses a constant acceleration model or a long short-term memory network model to predict the future position sequence of the equipment and calculate the minimum predicted distance and expected collision time between the equipment.

[0068] Preferably, it also includes a path planning and control module, which is connected to the dynamic fence calculation module and the risk assessment and early warning module; The path planning and control module specifically includes: Receive the center position and risk weight of all dynamic electronic fences; With the goal of maximizing the coverage value function V(P), the model predictive control framework is used to solve the optimal flight path of the UAV in real time. Output control commands to the UAV flight control system in the data acquisition module to drive the UAV to perform dynamic adaptive inspection.

[0069] Preferably, the early warning execution and feedback module includes several early warning distribution channels, including: mobile application push, on-site sound and light alarm, and monitoring platform pop-up window; The early warning execution and feedback module also includes a feedback receiving mechanism to record personnel's confirmation of early warnings or appeals against false alarms.

[0070] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for safety early warning and prediction of construction equipment based on UAV dynamic perception, characterized by: Includes the following steps: S1. Collect real-time video data of the construction site and high-precision pose data of the drone itself through inspection drones; S2. Process video data, identify construction equipment, key components and construction personnel, and obtain the geographical coordinates of construction equipment, key components and construction personnel; S3. Generate dynamic electronic fences that change with the operation status based on key components; S4. Based on historical equipment data and task type, predict the future movement trajectory of construction equipment and analyze trajectory anomalies; S5. Calculate the collision risk level between construction equipment and optimize the risk assessment based on the degree of overlap beyond the permitted range; S6. Based on different types of construction equipment, dynamically plan the inspection path of the drone through differentiated inspection strategies; S7. Combine dynamic electronic fences with prediction results to conduct risk analysis and generate safety warnings related to construction equipment.

2. The method for early warning and prediction of construction equipment safety based on UAV dynamic perception as described in claim 1, characterized in that: Step S3, the steps for generating a dynamic electronic fence, include: S31. Determine the geographical coordinates of the rotation center of the construction equipment through the key point detection model; S32. Identify the geographic coordinates of the tip of the boom of the construction equipment; S33. Calculate the real-time distance d between the center of rotation and the tip of the boom; S34. With the center of rotation as the center, and... A circular dynamic electronic fence is generated with radius [r, y], where [r, y] This is a preset safety margin.

3. The method for early warning and prediction of construction equipment safety based on UAV dynamic perception according to claim 2, characterized in that: In step S33, the real-time distance is calculated using the great circle distance formula: The formula for great circle distance is: ； in, The average radius of the Earth; , These are the longitude and latitude of the center of rotation of the construction equipment. , These are the longitude and latitude of the boom tip, respectively. The difference in longitude between the two key points.

4. The method for early warning and prediction of construction equipment safety based on UAV dynamic perception according to claim 1, characterized in that: Step S4 also includes the following steps: S410. Based on the historical location and speed data of the construction equipment, construct the motion state vector of the construction equipment; identify the current task type of the construction equipment; when the task type is a special task, supplement the motion state vector with the estimated task duration, cooperating equipment ID and pre-positioning coordinate parameters of the overlapping area. S420. Using a time-series prediction model, the location of construction equipment at several points in time within a specific future period is predicted to form a predicted trajectory. S430. Calculate the distance between the predicted trajectories of different construction equipment at various future time points, determine the minimum predicted distance and expected collision time between the predicted trajectories, and calculate the collision risk level. When the collision risk level exceeds the set threshold, trigger a collision warning. S440. Compare the predicted trajectory of the construction equipment with the preset original work trajectory, and calculate the trajectory deviation. , ,when If the deviation exceeds the preset threshold, it is determined to be a trajectory deviation risk, and the risk type and level analysis process is initiated. S450. For specific tasks, calculate the activity frequency of the device in the overlapping region: Determine the start time of construction equipment entering the overlapping area and the end time of its departure from the overlapping area; Calculate the time it takes for the construction equipment to reach key points on the trajectory; Output the collision avoidance time window between construction equipment.

5. The method for early warning and prediction of construction equipment safety based on UAV dynamic perception according to claim 4, characterized in that: In step S430, the formula for calculating the collision risk level is: ； in, Collision risk level, The minimum predicted distance is TTC, the predicted collision time is TTC, and the weighting coefficient is [missing information]. Overlap parameter ; The risk type and level process includes: The trajectory deviates from the risk level. The risk level is defined as ∈[5,10) meters. The risk level is defined as ∈[10,20) meters. ≥20 meters is classified as Level 3 risk; The risk level for exceeding the boundary is as follows: Level 1 risk is defined as the distance exceeding the boundary ∈ [1,3) meters, Level 2 risk is defined as the distance exceeding the boundary ∈ [3,5) meters, and Level 3 risk is defined as the distance exceeding the boundary ≥ 5 meters. Collision risk level is calculated using the collision risk level formula: when When the value is greater than 0.8, a level-two warning is triggered; when When the value is greater than 1.2, a Level 1 warning is triggered.

6. The method for early warning and prediction of construction equipment safety based on UAV dynamic perception according to claim 1, characterized in that: The types of construction equipment in step S6 include high-altitude rotating equipment, ground operation equipment, mobile operation equipment, and manned operation equipment. The dimensions of the differentiated inspection strategy include: inspection path shape, hovering point density, inspection height, key coverage area, and inspection speed. Each dimension is strongly linked to the operating characteristics of the equipment.

7. The method for early warning and prediction of construction equipment safety based on UAV dynamic perception according to claim 6, characterized in that: The differentiated inspection strategy is as follows: For high-altitude rotating equipment: the inspection path is segmented and reciprocating along the length of the boom, with hovering points set according to the distribution of key risk points on the boom, and the hovering time is adapted to the identification requirements of key components; the inspection height avoids obstructing the boom luffing mechanism to ensure a clear monitoring view; and high-frequency fault risk points are given priority coverage. For ground-based equipment: the inspection path covers the core operating area of ​​the equipment in a grid pattern, and the density of hovering points is adapted to the bucket operating radius and track travel range; dedicated hovering points are added at the bucket lifting limit position to enhance the monitoring of key actions; the inspection perspective is close to the working surface to ensure clear capture of the bucket's movement and the contact state with the ground; Mobile operation equipment: Inspection routes are dynamically adjusted to follow the equipment's operation path, prioritizing coverage of safety-sensitive areas; The inspection speed is matched with the equipment movement speed to avoid identification errors caused by excessive flight speed; fixed stopping inspection points are set at the unloading point to ensure that the entire unloading process is captured; For manned operation equipment: the inspection path is circular, surrounding the platform's lifting channel and the area where workers stand, and the inspection height is dynamically matched with the platform's real-time operating height; the lateral coverage area covers the personnel safety boundary, ensuring that there are no blind spots in the dynamic monitoring of workers and the equipment's lifting mechanism.

8. The method for safety early warning and prediction of construction equipment based on UAV dynamic perception according to claim 7, characterized in that: When two or more different types of equipment are operating simultaneously at a construction site, the drone inspection path is planned according to the risk priority of the equipment. The risk priority is: manned operation equipment > high-altitude rotating equipment > ground operation equipment > mobile operation equipment. The inspection frequency of high-priority equipment is higher than that of low-priority equipment. For equipment of the same priority, the inspection density increases with the degree of overlap of the operation area, and the inspection points are more densely distributed in the overlapping areas.

9. A method for safety early warning and prediction of construction equipment based on UAV dynamic perception according to claim 7, characterized in that: For combinations of equipment used in collaborative operations, dedicated coverage sections for collaborative operations are set up along the drone inspection routes; The dedicated coverage area for collaborative operations must meet the following requirements: The coverage area includes the overlapping areas of equipment operations and the areas where coordinated actions are connected; The inspection interval is adapted to the accuracy requirements of collaborative operations, and the number of hovering points is matched with the number of collaborative devices to ensure that the key actions of each collaborative device can be captured. The hovering point view simultaneously covers the movement trajectories of key components of all collaborative devices, ensuring the continuity of collaborative actions and safety monitoring.

10. The method for early warning and prediction of construction equipment safety based on UAV dynamic perception according to claim 7, characterized in that: In step S6, when dynamically planning the UAV inspection path, the inspection strategy is adjusted synchronously based on the current operating status of the associated equipment: When the equipment is under load, extend the dwell time at the hover point to improve the accuracy of key component status recognition; When the equipment is in motion, the inspection path following speed is increased simultaneously to ensure full coverage of the movement route; When equipment is in collaborative operation mode, increase the inspection frequency of the dedicated coverage area for collaborative operation and strengthen the risk detection of collaborative actions.

Images