Real-time monitoring system and method for construction sites based on drone technology

Abstract

The present invention relates to the field of construction monitoring technology, specifically to a real-time monitoring system and method for construction sites based on drone technology. Specifically, the drone is equipped with a high-definition camera, an infrared thermal imager, and a laser radar to collect multimodal data from the construction site in real time. A deep learning algorithm is used for multimodal semantic segmentation, fusing RGB images, infrared thermal imaging, and laser point cloud projection data. Combined with YOLOv8 target detection and temporal behavior modeling technology, abnormal risks are dynamically identified. A context-aware comprehensive risk scoring model is constructed to quantify the danger level of the construction scene and trigger a graded early warning mechanism. Simultaneously, based on three-dimensional modeling technology, the LIO-SLAM algorithm is used to analyze tank vertex cloud data, detect the degree of structural inclination, and initiate an emergency response. The present invention achieves intelligent and precise construction monitoring through multi-sensor fusion, dynamic risk assessment, and closed-loop control.

Description

Technical Field

[0001] The present invention relates to the technical field of building construction monitoring, and in particular to a real-time monitoring system and method for a construction site based on drone technology. Background Art

[0002] Construction site safety monitoring is a core requirement of construction project management. Traditional manual inspections and single-sensor monitoring have bottlenecks such as limited coverage, poor real-time performance, and static risk assessment. Existing technologies mostly rely on fixed cameras or drones, which are difficult to adapt to the dynamic changes of construction scenarios, especially in night operations, high-altitude areas or complex equipment layouts, which are prone to monitoring blind spots. In addition, risk warning mechanisms are usually based on preset thresholds and lack comprehensive dynamic analysis of human behavior, equipment status and environmental factors.

[0003] Existing image processing technologies mostly use single-modal data (such as visible light or infrared), resulting in insufficient target segmentation accuracy. This makes it difficult to accurately identify construction workers' illegal operations (such as not wearing safety helmets and safety belts), equipment anomalies (such as lifting offset) and structural hazards (such as tank roof blowing up and tilting).

[0004] In addition, with the development of drone platforms and artificial intelligence technologies, there is an urgent need to build a full-link monitoring system covering data collection, intelligent analysis to active intervention to address the shortcomings of existing technologies in multimodal fusion, behavioral dynamic modeling and risk quantification decision-making. Summary of the Invention

[0005] The purpose of the present invention is to address the problems existing in the background technology and propose a real-time monitoring system and method for construction sites based on drone technology.

[0006] The technical solution of the present invention is a real-time monitoring system for construction sites based on drone technology, comprising:

[0007] The drone measurement module is equipped with a drone equipped with a high-definition camera, infrared thermal imager and lidar to collect RGB images, infrared thermal imaging data and lidar point cloud data of the construction site in real time;

[0008] The cloud platform management module includes a data storage and management unit and a data analysis and processing unit, which is used to store collected data and perform semantic segmentation, target detection, and dynamic risk assessment on multimodal data using deep learning algorithms;

[0009] Data storage and management unit: used to store construction site image data collected by drones;

[0010] Data analysis and processing unit: Real-time analysis of image data from the construction site;

[0011] Alert and feedback module: triggers a graded response mechanism based on risk assessment results.

[0012] Preferably, the data analysis and processing unit fuses RGB images, infrared images and laser point cloud projection images, uses the DeepLabv3+ network for pixel-level semantic segmentation, and optimizes the segmentation accuracy of high-risk areas through the weighted cross-entropy loss function; identifies and detects targets based on the YOLOv8 network, and combines the DeepSORT algorithm to achieve continuous target tracking; and extracts dynamic behavior features through temporal behavior modeling, and outputs risk probability values; combines risk probability, environmental context weight and preset thresholds to generate four levels of warning levels.

[0013] Preferably, the weighted cross entropy loss function is: ;

[0014] Among them, L seg is the loss function; y i is the true category label of pixel i; is the predicted probability output by the network; w c (i) is the category weight of the i-th category pixel.

[0015] Preferably, temporal behavior modeling includes a TSM network and a Transformer encoder, which fuses target trajectory, posture, and scene position information to output risk probability.

[0016] The technical solution of the present invention is a method for real-time monitoring of construction sites based on drone technology, which is applied to the above-mentioned real-time monitoring system for construction sites based on drone technology, and includes the following specific implementation steps:

[0017] S51, initialize the UAV flight path and mission parameters, and dynamically match the construction site scope and environmental factors;

[0018] S52, the drone collects and transmits multimodal data;

[0019] S53, performs adaptive histogram equalization and non-local mean denoising on RGB images, fuses infrared images with point cloud projections to generate 6-channel fusion features; performs semantic segmentation based on the DeepLabv3+ network and outputs a segmentation map; combines YOLOv8 and DeepSORT to achieve target detection and tracking, and outputs detection results; outputs risk probabilities through temporal behavior modeling, and generates comprehensive risk scores based on environmental context; triggers four levels of warning based on the comprehensive risk score and outputs a response level; and analyzes the degree of tank structure inclination based on LiDAR 3D modeling and point cloud analysis methods;

[0020] S54. Initiate a feedback mechanism based on the response level;

[0021] S55. Store the current multimodal data, segmentation map, detection results, comprehensive risk score, response level, and corresponding feedback measures.

[0022] Preferably, comprehensive risk score: ;

[0023] in, is the comprehensive risk score at time t; K is the total number of risk behavior types; β k is the basic weight of the kth risk category; is the risk probability of the kth behavior at time t; The context weight is adjusted according to the current job environment.

[0024] Preferably, the implementation steps of the laser radar three-dimensional modeling and point cloud analysis method are:

[0025] S71, instructing the UAV to obtain point cloud data and odometer data of the tank wall and tank top in the UAV coordinate system;

[0026] S72. Fuse the point cloud data and the odometer data using the LIO-SLAM algorithm, filter out the three-dimensional coordinates of the point cloud at the junction of the tank top and the tank wall in the world coordinate system, and calculate the tilt of the tank top:

[0027] Adopt LIO-SLAM algorithm fusion and pose optimization;

[0028] Select frames that are farther apart in time but closer in distance from the historical key frames as candidate closed-loop frames, extract the feature point set of the current frame, downsample, perform scantomap optimization, and use the iterative closest point algorithm to obtain the optimized pose;

[0029] Downsample the point cloud image to determine the boundary between the tank top and the tank wall. Fill the point cloud and remove the interference of the tank wall point cloud data to obtain a point cloud image containing only the tank top data.

[0030] Extract the point cloud on the circumference of the tank top edge, based on the maximum value z of the edge point cloud height max and the minimum value z min , we can get the tilt degree of the tank top ΔH=z max -z min ;

[0031] The inclination degree of the output tank top ΔH;

[0032] S73, if ΔH ≥ the set maximum allowable tank top tilt height difference threshold ΔH max , trigger the sound and light alarm and adjust the blowing speed and air intake volume.

[0033] Preferably, the implementation process of LIO-SLAM algorithm fusion and pose optimization is:

[0034] S81, calculate IMU pre-integral: calculate the velocity, position and rotation changes between adjacent frames:

[0035] ;

[0036] ;

[0037] ;

[0038] Where Δt ij Indicates time t i and t j time interval; is the estimated value of the speed; is the estimated value of the position; v i and v j t i and t j The speed at the moment; g is the acceleration due to gravity; is the estimated value of the rotation matrix; R i and R j t i and t j The rotation matrix at the moment;

[0039] S82, Feature Matching: Use the distance relationship between points and lines and points and surfaces to establish the pose solution equation:

[0040] ;

[0041] ;

[0042] ;

[0043] Among them, k, u, v, and w are all corresponding features; is the 3D coordinate of the kth line feature point in the i+1th frame; and are the coordinates of the two endpoints of the same line feature in the i-th frame; d ek Indicates the current time point The vertical distance to the previous timeline feature; d pk For the current time point The vertical distance to the plane at the previous moment; is the 3D coordinate of the kth plane feature point in the i+1th frame; 、 and are three non-collinear points defining the plane in the i-th frame; e represents the line feature; P represents the surface feature; ΔT i,i+1is the relative pose between the current key frame and the previous key frame; T i and T i+1 is the pose of the frame at time i and time i+1;

[0044] S83, d ek and d pk Construct the objective function and solve the optimal ΔT by minimizing the error i,i+1 , perform pose optimization.

[0045] Compared with the prior art, the above technical solution of the present invention has the following beneficial technical effects:

[0046] This paper designs a real-time construction site monitoring system and method based on drone technology. Using drones equipped with multimodal sensors (high-definition cameras, infrared thermal imagers, and lidar), this system enables all-weather three-dimensional data acquisition of construction scenes. Combined with the DeepLabv3+ multi-scale semantic segmentation model and the YOLOv8 target detection algorithm, the system significantly improves the accuracy of identifying people, equipment, and hazardous areas in images. It utilizes temporal behavioral modeling (TSM+attention network) and DeepSORT target tracking technology to dynamically analyze construction worker violations (such as not wearing helmets) and equipment abnormalities (such as hoisting offset). A context-aware dynamic risk assessment model is constructed through multi-dimensional feature fusion, achieving quantitative scoring of risk probability and a four-level warning classification system. LiDAR point cloud data and the LIO-SLAM algorithm are used to accurately calculate the tilt height difference of the tank roof. Combined with a closed-loop response mechanism (such as automatic operation suspension and manual reset), the system effectively prevents structural collapse accidents. Furthermore, through automatic flight path planning, real-time fusion of multi-source data, and graded alarm push, the system reduces manual inspection costs, improves the real-time and intelligent level of construction safety management, and provides a high-precision, full-process active safety monitoring solution for complex construction scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 This is a system architecture diagram of a real-time monitoring system for construction sites based on drone technology proposed by the present invention;

[0048] Figure 2 A schematic diagram of a point cloud image containing only tank top data. DETAILED DESCRIPTION

[0049] Example 1, as Figure 1 As shown, the present invention proposes a real-time monitoring system for construction sites based on drone technology, including: a drone measurement module, a cloud platform management module, and an alarm and feedback module.

[0050] Drone measurement module: A drone equipped with a high-definition camera, infrared thermal imager, and lidar is used to collect image data from the construction site;

[0051] Cloud platform management module: used to store and analyze collected construction site image data, including:

[0052] Data storage and management unit: used to store construction site image data collected by drones to ensure data security and traceability;

[0053] Data analysis and processing unit: Utilizes big data analysis and machine learning technologies to conduct real-time analysis of construction site image data, including but not limited to detecting safety hazards, personnel location, and equipment status;

[0054] Alarm and feedback module: When potential safety risks are detected (including but not limited to illegal operations and personnel crossing the boundary), an alarm will be automatically issued and real-time feedback will be provided to construction management personnel through mobile terminal devices.

[0055] In Example 2, a method for real-time monitoring of a construction site based on drone technology proposed in the present invention is applied to a real-time monitoring system for a construction site based on drone technology proposed in Example 1. The specific implementation steps are as follows:

[0056] S1. System initialization phase: To ensure that the drone can effectively cover the construction site and complete the monitoring mission, initialization is first performed, including but not limited to hardware equipment inspection, flight environment setting, and clear monitoring target definition. Specifically:

[0057] The user inputs basic information of the construction site and monitoring requirements (including but not limited to area scope, monitoring frequency, and specific targets) through a mobile terminal device;

[0058] Automatically generate flight paths based on the location and size of the construction site, and dynamically adjust them based on environmental factors (including but not limited to wind speed and weather conditions);

[0059] Set the drone's flight mode (including but not limited to cruise, fixed-point stay, etc.) and mission parameters (including but not limited to image acquisition).

[0060] S2. The drone measurement module instructs the drone to perform the task according to the predetermined flight mission and start on-site monitoring: After the drone takes off, it automatically flies according to the preset path, enters the monitoring area, and starts to synchronously collect real-time data of the construction site based on the equipped high-definition camera, infrared thermal imager, and lidar equipment;

[0061] Transmit construction site data to the cloud platform management module.

[0062] S3, the data analysis and processing unit, uses intelligent algorithms to process and analyze construction site data collected from drones in real time. It builds a full-process analysis mechanism from image data preprocessing to intelligent identification and risk assessment. Through deep learning and multi-dimensional information fusion, it achieves efficient and intelligent monitoring of construction sites and dynamic risk perception, identifying abnormal situations, including but not limited to potential safety hazards, equipment failures, and personnel violations. The specific implementation steps are as follows:

[0063] S31. Image Data Preprocessing and Multimodal Semantic Segmentation: Because drone-captured image data has large variations in perspective, large differences in target size, and high levels of noise, the original images must first be standardized and preprocessed. A multimodal semantic segmentation model is then introduced to fuse RGB images, infrared thermal imaging, and point cloud projections to accurately segment core targets within the construction area (including but not limited to personnel, equipment, warning lines, and danger zones). Specifically:

[0064] S3101 uses adaptive histogram equalization (CLAHE) to process RGB image channels to enhance local contrast, and introduces the non-local means denoising algorithm (NLM) to reduce noise while preserving edges. The drone sensor attitude parameters are matched with the image SIFT feature points, and the back-projection model is used to restore the distorted area.

[0065] ;

[0066] Among them, (x,y) represents the original coordinates; (x corr ,y corr ) represents the corrected coordinates; k1 and k2 represent the radial distortion coefficients, which are obtained by the UAV internal calibration;

[0067] S3102: Fusion of RGB images, infrared images, and laser point cloud projections to enhance image depth and thermal information expression and improve segmentation accuracy. Specifically:

[0068] (1) Match the timestamps of RGB, infrared and point cloud data, and perform two-dimensional projection of the point cloud to generate a grayscale depth map;

[0069] (2) Using the SURF algorithm to detect image key points, calculate the homography matrix H to achieve image spatial alignment: pʹ = H·p;

[0070] Where p is the pixel coordinate in the original image; pʹ is the coordinate after registration; H is the 3×3 homography matrix;

[0071] (3) Spatial registration and channel-level fusion of RGB images, infrared images, and depth images are performed to construct a multimodal feature tensor containing spatial, thermal, and geometric structure information (i.e., a 6-channel fused image);

[0072] S3103 uses the DeepLabv3+ network structure and combines the DeepLabv3+ network's multi-scale receptive field module (ASPP) to capture local and global structures. The input is a 6-channel fused image and the output is a pixel-level semantic segmentation map.

[0073] Use the improved weighted cross entropy as the loss function: ;

[0074] Among them, L seg is the loss function; y i is the true category label of pixel i; is the predicted probability output by the network; w c (i) is the category weight of the i-th category pixel, which is set based on experience;

[0075] For example: High-risk category (hoisting equipment) weight: >1.5; Medium-risk category (stacked objects, machinery) weight: ≈1; Low-risk category (open space) weight: <0.8;

[0076] S3104, output the segmented image;

[0077] S32. Based on the semantically segmented image regions, the system uses YOLOv8 for multi-target detection to accurately identify objects such as construction workers, helmets, equipment, and signage. To achieve intelligent recognition of dynamic behaviors, the system introduces a temporal behavior modeling module (TSM (Temporal Shift Module) + attention fusion network) to extract action features from consecutive frames. Specifically:

[0078] S3201, use the target detection network YOLOv8 (You Only Look Once version 8) as the main detector, input the image output after semantic segmentation into the YOLOv8 network, perform target recognition and classification, and output the position, category and confidence: any detection box B i =[x i ,y i ,w i ,h i ,c i ,s i ];

[0079] Among them, (x i ,y i ), w i 、h i 、c i 、s i They are the center point coordinates, width and height, category, and confidence of the bounding box respectively;

[0080] S3202, tracking the detected targets in the video frame sequence, using the DeepSORT algorithm, combined with appearance features and Kalman filtering, to complete target numbering and continuous tracking, and generate a trajectory sequence for each target;

[0081] S3203. Build a temporal modeling network: TSM (Temporal Shift Module) + TransformerEncoder. Output the target's key frame images, trajectory, and posture information over a period of time to the temporal modeling network and output the recognition results.

[0082] Based on this, we introduced multi-dimensional behavioral feature fusion: image + motion trajectory + scene location information, binding behaviors to semantic regions, and improving recognition confidence (for example, the behavior of "not wearing a helmet" must be within the "personnel region");

[0083] S3204. Convert identification results into quantitative probability values ​​to provide input for subsequent risk assessment:

[0084] ;

[0085] in, is the risk probability of the kth behavior at time t; is j behavioral impact features (including but not limited to: construction workers not wearing safety helmets, abnormal state of lifting equipment); is the weight coefficient; σ is the Sigmoid activation function, and the output value is normalized to [0,1];

[0086] S33. Quantify the identified risk behaviors in multiple dimensions and construct a comprehensive construction scenario risk scoring model to quantitatively reflect the current danger level and potential problems at the work site. Specifically:

[0087] S3301. Build a risk factor database: Define various risk events, including but not limited to "abnormal lifting equipment posture" or "obstacles piled in the construction area";

[0088] S3302. Extract the risk intensity corresponding to each type of behavior and weight it based on its behavior probability and environmental context;

[0089] S3303. Output overall risk score:

[0090] ;

[0091] in, is the comprehensive risk score at time t; K is the total number of risk behavior types; β k is the basic weight of the kth risk category; is the risk probability of the kth behavior at time t; Context weights adjusted based on the current operating environment (including but not limited to "nighttime operation", "near edge", and "enclosed space");

[0092] Based on this, static image information and dynamic behavior reasoning results are jointly modeled to achieve dynamic risk assessment under context awareness, and the probabilistic output is converted into globally usable quantitative results, which serve as the basis for system decision-making and response;

[0093] S34. Automatically initiate a hierarchical response mechanism based on the risk scoring results, specifically:

[0094] Pre-set thresholds and use the comprehensive risk score Mapped to four levels of response level Level(t):

[0095] ;

[0096] S35. The current risk scene image, segmentation map, detection results, scoring value, and response level are packaged and uploaded to the data storage and management unit, and the response level is sent to the alarm and feedback module.

[0097] S4. The alarm and feedback module activates the feedback mechanism based on the response level Level(t), including but not limited to: local voice broadcast (drone call); mobile terminal device push notification (construction manager); supervision platform warning synchronization.

[0098] S5. The data storage and management unit uses a built-in database to store the current risk scene image, segmentation map, detection results, scoring values, response levels, and corresponding feedback measures.

[0099] In a third embodiment, the present invention proposes a method for real-time monitoring of a construction site based on drone technology, which also includes a laser radar three-dimensional modeling and point cloud analysis method. The specific implementation steps are as follows:

[0100] A1. The drone measurement module instructs the drone to be equipped with a lidar, hover at a height of 15m, scan at a frequency of 10Hz, with a vertical viewing angle of -30° to +10° and a point cloud density of ≥10 points / m², ensuring that the top edge of the large storage tank is clearly visible;

[0101] Based on this: a coordinate system is established with the location of the drone as the origin, the direction facing the drone as the x-axis, the horizontal axis of the drone as the y-axis, and the vertical axis as the z-axis. The point cloud data and odometer data of the tank wall and tank top in the drone coordinate system are obtained, and the point cloud data and odometer data are transmitted to the cloud platform management module.

[0102] A2. The data analysis and processing unit fuses the point cloud data with the odometer data using the LIO-SLAM algorithm to obtain the position of these point cloud data in the world coordinate system. It then selects the three-dimensional coordinates of the point cloud at the junction of the tank roof and the tank wall in the world coordinate system. This allows the calculation of the tilt of the tank roof, specifically:

[0103] A21, using LIO-SLAM algorithm fusion and pose optimization, specifically:

[0104] (1) Calculate IMU pre-integration: Calculate the changes in velocity, position, and rotation between adjacent frames:

[0105] ;

[0106] ;

[0107] ;

[0108] Where Δt ij Indicates time t i and t j time interval; is the estimated value of the speed; is the estimated value of the position; v i and v j t i and t j The speed at the moment; g is the acceleration due to gravity; is the estimated value of the rotation matrix; R i and R j t i and t j The rotation matrix at the moment;

[0109] (2) Feature matching: Using the distance relationship between points and lines and points and surfaces, we can establish the pose solution equation:

[0110] ;

[0111] ; ;

[0112] Among them, k, u, v, and w are all corresponding features; is the 3D coordinate of the kth line feature point in the i+1th frame (current frame); and are the coordinates of the two endpoints of the same line feature in the i-th frame (previous frame); d ek Indicates the current time point The vertical distance to the previous timeline feature; d pk For the current time point The vertical distance to the plane at the previous moment; is the 3D coordinate of the kth plane feature point in the i+1th frame; 、 and are three non-collinear points defining the plane in the i-th frame; e represents the line feature; P represents the surface feature; ΔT i,i+1 is the relative pose between the current key frame and the previous key frame; T i and T i+1 is the pose of the frame at time i and time i+1;

[0113] Therefore: d ek and d pk Construct the objective function and solve the optimal ΔT by minimizing the error i,i+1 ;

[0114] It should be noted that LIO-SLAM is a 3D laser SLAM algorithm. SLAM, short for Simultaneous Localization and Mapping, is a technology that combines simultaneous mapping and localization. LIO-SAM is a tightly coupled lidar-inertial odometry framework based on factor graph optimization.

[0115] A22. Use the method of matching lidar scan frames with maps, select frames with longer time intervals and shorter distances from the historical key frames as candidate closed-loop frames, extract the feature point set of the current frame, downsample, perform scantomap optimization, and use the iterative closest point algorithm to obtain the optimized pose;

[0116] A23. Downsample the point cloud image to reduce the amount of computation while preserving the point cloud features. Determine the boundary between the tank top and the tank wall. Since the boundary is far from the LiDAR and the point cloud is sparse, it is necessary to fill the point cloud. Then remove the interference of the tank wall point cloud data to obtain a point cloud image containing only the tank top data (e.g. Figure 2 shown);

[0117] A24. Based on the top cloud of the tank, extract the point cloud on the circumference of the tank top edge. Since the number of sampling points on the edge is large enough, the maximum value of the edge point cloud height z can be used. max Subtract the minimum value z min Get the tilt of the tank top ΔH=z max -z min ;

[0118] A25. Transmit the tilt degree ΔH of the tank top to the alarm and feedback module.

[0119] A3. The alarm and feedback module receives the tank top tilt height difference ΔH. If ΔH ≥ ΔH max(The set maximum allowable tank top tilt height difference threshold meets the tank top specification requirements), triggering the sound and light alarm and transmitting the tank top balance data in real time, adjusting the blowing speed and air intake volume, and then adjusting the tank top posture.

[0120] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited thereto. Various changes can be made within the scope of knowledge possessed by those skilled in the art without departing from the spirit of the present invention.

Claims

1. A real-time monitoring system for construction sites based on drone technology, characterized in that: include: The drone measurement module is equipped with a drone equipped with a high-definition camera, infrared thermal imager and lidar to collect RGB images, infrared thermal imaging data and lidar point cloud data of the construction site; The cloud platform management module includes a data storage and management unit and a data analysis and processing unit, which is used to store collected data and perform semantic segmentation, target detection, and dynamic risk assessment on multimodal data using deep learning algorithms; Data storage and management unit: used to store construction site image data collected by drones; Data analysis and processing unit: performs real-time analysis of image data from the construction site: performs adaptive histogram equalization and non-local mean denoising on RGB images, and fuses infrared images with point cloud projection images to generate 6-channel fusion features; Perform semantic segmentation based on the DeepLabv3+ network and output segmentation maps; combine YOLOv8 and DeepSORT to achieve target detection and tracking, and output detection results; output risk probability through temporal behavior modeling, and generate a comprehensive risk score based on environmental context; Trigger four-level warnings based on the comprehensive risk score and output the response level; At the same time, the tilt degree of the tank structure is analyzed based on the LiDAR 3D modeling and point cloud analysis method; Alarm and feedback module: activates the feedback mechanism based on the response level of behavioral risk assessment and the instructions of structural tilt monitoring; The implementation steps of the lidar-based 3D modeling and point cloud analysis method are as follows: A1. Instruct the UAV to obtain point cloud data and odometer data of the tank wall and tank top in the UAV coordinate system; A2. Fuse the point cloud data with the odometry data using the LIO-SLAM algorithm, filter out the 3D coordinates of the point cloud at the junction of the tank roof and the tank wall in the world coordinate system, and calculate the tilt of the tank roof: Adopt LIO-SLAM algorithm fusion and pose optimization; Select frames that are farther apart in time but closer in distance from the historical key frames as candidate closed-loop frames, extract the feature point set of the current frame, downsample, perform scantomap optimization, and use the iterative closest point algorithm to obtain the optimized pose; Downsample the point cloud image to determine the boundary between the tank top and the tank wall. Fill the point cloud and remove the interference of the tank wall point cloud data to obtain a point cloud image containing only the tank top data. Extract the point cloud on the circumference of the tank top edge, based on the maximum value z of the edge point cloud height max and the minimum value z min , we can get the tilt degree of the tank top ΔH=z max -z min ; The inclination degree of the output tank top ΔH; A3. If ΔH ≥ the set maximum allowable tank top tilt height difference threshold ΔH max , trigger the sound and light alarm and adjust the blowing speed and air intake volume.

2. A construction site real-time monitoring system based on drone technology according to claim 1, characterized in that: The data analysis and processing unit integrates RGB images, infrared images, and laser point cloud data, uses the DeepLabv3+ network for pixel-level semantic segmentation, and optimizes the segmentation accuracy of high-risk areas through a weighted cross-entropy loss function. It identifies and detects targets based on the YOLOv8 network and combines the DeepSORT algorithm to achieve continuous target tracking. It extracts dynamic behavior features through temporal behavior modeling and outputs risk probability values. It combines risk probability, environmental context weights, and preset thresholds to generate four levels of warning levels.

3. A real-time monitoring system for construction sites based on drone technology according to claim 2, characterized in that: The weighted cross entropy loss function is: ； Among them, L seg is the loss function; y i is the true category label of pixel i; is the predicted probability output by the network; w c (i) is the category weight of the i-th category pixel.

4. The real-time monitoring system for construction sites based on drone technology according to claim 2 is characterized in that: Temporal behavior modeling includes the TSM network and the Transformer encoder, which integrates target trajectory, posture, and scene position information to output risk probability.

5. The real-time monitoring system for construction sites based on drone technology according to claim 1 is characterized in that: The implementation process of LIO-SLAM algorithm fusion and pose optimization is as follows: B1. Calculate IMU pre-integration: Calculate the changes in velocity, position, and rotation between adjacent frames: ； ； ； Where Δt ij Indicates time t i and t j time interval; is the estimated value of the speed; is the estimated value of the position; v i and v j t i and t j The speed at the moment; g is the acceleration due to gravity; is the estimated value of the rotation matrix; R i and R j t i and t j The rotation matrix at the moment; B2. Feature matching: Use the distance relationship between points and lines and points and surfaces to establish the pose solution equation: ； ； ； Among them, k, u, v, and w are all corresponding features; is the 3D coordinate of the kth line feature point in the i+1th frame; and are the coordinates of the two endpoints of the same line feature in the i-th frame; d ek Indicates the current time point The vertical distance to the previous timeline feature; d pk For the current time point The vertical distance to the plane at the previous moment; is the 3D coordinate of the kth plane feature point in the i+1th frame; 、 and are three non-collinear points defining the plane in the i-th frame; e represents the line feature; P represents the surface feature; ΔT i,i+1 is the relative pose between the current key frame and the previous key frame; T i and T i+1 is the pose of the frame at time i and time i+1; B3, d ek and d pk Construct the objective function and solve the optimal ΔT by minimizing the error i,i+1 , perform pose optimization.

6. A method for real-time monitoring of a construction site based on drone technology, which is applied to a real-time monitoring system for a construction site based on drone technology according to any one of claims 1 to 5, characterized in that: The specific implementation steps include the following: S61, initialize the UAV flight path and mission parameters, and dynamically match the construction site scope and environmental factors; S62, the drone collects and transmits multimodal data; S63, performing adaptive histogram equalization and non-local mean denoising on the RGB image, fusing the infrared image and the point cloud projection image to generate a 6-channel fusion feature; Perform semantic segmentation based on the DeepLabv3+ network and output segmentation maps; combine YOLOv8 and DeepSORT to achieve target detection and tracking, and output detection results; output risk probability through temporal behavior modeling, and generate a comprehensive risk score based on environmental context; Trigger four-level warnings based on the comprehensive risk score and output the response level; At the same time, the tilt degree of the tank structure is analyzed based on the LiDAR 3D modeling and point cloud analysis method; S64. Initiate feedback mechanisms based on the response level of behavioral risk assessment and the instructions of structural tilt monitoring; S65. Store the current multimodal data, segmentation map, detection results, comprehensive risk score, response level, and feedback measures.

7. The method for real-time monitoring of a construction site based on drone technology according to claim 6, characterized in that: Comprehensive risk score: ; in, is the comprehensive risk score at time t; K is the total number of risk behavior types; β k is the basic weight of the kth risk category; is the risk probability of the kth behavior at time t; The context weight is adjusted according to the current job environment.

Images