A construction site hazard source identification system and method based on drone

Abstract

The present invention discloses a construction site hazard source identification system and method based on unmanned aerial vehicles, and relates to the technical field of building construction. The construction site hazard source identification system includes a construction data acquisition module, a hazard source identification module, a risk prediction module and a risk warning module; the construction data acquisition module is used to collect remote sensing image data of the construction site using an unmanned aerial vehicle group equipped with remote sensing equipment; the hazard source identification module is used to identify the hazard sources at the construction site based on the hazard source characteristics; the risk prediction module is used to classify the identified hazard sources and predict the probability of occurrence of hazard sources at different levels. The present invention analyzes the human body motion trajectory through optical flow technology and pyramid model, can accurately track and analyze the specific movement path and behavior pattern of personnel at the construction site, and by comparing the actual behavior trajectory with the preset safety distance, can judge whether the personnel have entered the dangerous area to provide effective safety warnings.

Description

Technical Field

[0001] The present invention relates to the technical field of building construction, and in particular to a construction site hazard source identification system and method based on an unmanned aerial vehicle. Background Art

[0002] Construction refers to the entire process of building a building and its ancillary facilities from foundation to completion in accordance with design drawings and technical standards and through various building materials and construction technologies. Construction site hazard identification is the systematic identification, evaluation and control of potential risk factors that may cause personal injury, property loss or environmental damage during the construction process, which is essential to ensure the safety of the construction site. There are many types of hazard sources at the construction site, including dangers of working at heights, people entering dangerous areas, and dangers of foundation pit slope collapse.

[0003] Pit slope collapse is a serious danger that construction sites may face, especially when deep pit construction is carried out. If the pit design does not fully consider geological conditions and load requirements, it will cause insufficient slope stability and the risk of pit slope collapse. Failure to identify the danger of slope collapse in time may lead to sudden collapse events, which will not only cause harm to on-site workers, but also damage on-site machinery and other facilities, resulting in economic losses and affecting the construction progress.

[0004] Currently, no effective solution has been proposed for the problems in the related technologies. Summary of the invention

[0005] In view of the shortcomings of the prior art, the present invention provides a construction site hazard source identification system and method based on drones, which has the advantage of identifying the hazard of slope collapse, and thus can solve the problem of being unable to identify the hazard of slope collapse.

[0006] In order to achieve the above advantages of identifying the danger of slope collapse, the specific technical solutions adopted by the present invention are as follows:

[0007] According to one aspect of the present invention, a construction site hazard source identification system based on a drone is provided, the construction site hazard source identification system comprising a construction data acquisition module, a hazard source identification module, a risk prediction module and a risk warning module;

[0008] The construction data acquisition module is used to collect remote sensing image data of the construction site using the remote sensing equipment carried by the drone group and pre-process the remote sensing image data;

[0009] The hazard source identification module is used to extract hazard source features from the pre-processed remote sensing image data and identify the hazard sources at the construction site based on the hazard source features;

[0010] The risk prediction module is used to classify the identified hazards and predict the probability of occurrence of hazards of different levels;

[0011] The risk warning module is used to formulate corresponding emergency response measures based on the probability of occurrence of different levels of hazardous sources.

[0012] Furthermore, the hazard source identification module includes a hazard area calibration module, a hazard source feature extraction module, an abnormal behavior identification module and a foundation pit collapse identification module;

[0013] The dangerous area calibration module is used to extract continuous image sequences from the processed remote sensing image data, and to segment and calibrate specific dangerous areas from each construction image;

[0014] The hazard source feature extraction module is used to extract the key features of the hazard source from the construction image, where the key features include human motion features and foundation pit excavation features;

[0015] The abnormal behavior recognition module is used to analyze the safety distance from the human movement characteristics to the specific dangerous area at the construction site, and judge whether there is abnormality in the human behavior based on the safety distance;

[0016] The foundation pit collapse identification module is used to take the foundation pit excavation features as input to perform surface fitting and point cloud reconstruction in sequence to identify the safety status of the foundation pit.

[0017] Furthermore, the dangerous area demarcation module includes: extracting a continuous image sequence from the processed remote sensing image data, and segmenting and demarcating a specific dangerous area from each construction image;

[0018] Extracting a continuous image sequence from the processed remote sensing image data, and segmenting the construction area from each construction image in the continuous image sequence using image segmentation technology;

[0019] Construct a 3D view based on the construction area, and combine the terrain data of the construction area to display the mapping relationship between construction terrain changes and construction activities in real time in the 3D view;

[0020] Based on the mapping relationship between construction terrain changes and construction activities, specific dangerous areas are marked in the three-dimensional view, where specific dangerous areas include construction excavation areas, elevated operation areas and lifting operation areas.

[0021] Furthermore, the abnormal behavior recognition module analyzes the safety distance from the human movement characteristics to the specific dangerous area in the construction site, and determines whether the human behavior is abnormal based on the safety distance, including:

[0022] Performing differential processing on each construction image in the continuous image sequence to obtain a processed frame difference image, and screening the frame difference image into a first image and a second image;

[0023] Performing image binarization processing on the first image and the second image respectively, and performing logical operation bit by bit to obtain the human body movement area in the construction site;

[0024] A pyramid model is established based on the human body motion area, and the optical flow estimation value is initialized at the lowest layer of the pyramid model; the optical flow value of each layer of the pyramid model is calculated layer by layer from the lowest layer of the pyramid model;

[0025] Add the optical flow values ​​of each layer of the pyramid model and locate the human feature points in each frame difference image to identify the human motion trajectory;

[0026] Obtain a feasible path for the human motion trajectory to reach a specific dangerous area, and calculate the distance from the human motion trajectory on the feasible path to the specific dangerous area;

[0027] The distance from the human body's motion trajectory to a specific dangerous area is compared with the preset safety distance, and based on the comparison result, it is determined whether there is any abnormality in the human behavior.

[0028] Furthermore, obtaining a feasible path for the human motion trajectory to reach a specific dangerous area, and calculating the distance from the human motion trajectory to the specific dangerous area on the feasible path includes:

[0029] Combined with the terrain changes in the construction area, the movement path of human body to the specific dangerous area is identified from the 3D view;

[0030] The motion paths within the preset range are relaxed, and a distribution factor is introduced to dynamically distribute the search weight of each motion path according to the search tree structure;

[0031] Based on the search weight of the motion path, the trust value of each motion path is evaluated respectively, and the priority function is set for each motion path node;

[0032] The motion path node with the smallest priority is selected to update the priority function. Based on the updated priority function, the feasible path from the human motion trajectory to the specific dangerous area is determined and the distance is calculated.

[0033] Furthermore, the foundation pit collapse identification module takes the foundation pit excavation features as input and performs surface fitting and point cloud reconstruction in sequence to identify the safety status of the foundation pit, including:

[0034] Filter out foundation pit images from construction images, and convert foundation pit feature points in the foundation pit images into point cloud data;

[0035] The point cloud data in each foundation pit image are registered using the iterative closest point algorithm to ensure that all point cloud data are aligned in the same coordinates;

[0036] The surface fitting technology is used to fit the excavation features of the foundation pit, and the geometric model of the foundation pit is reconstructed based on the surface fitting results;

[0037] Based on the preset local deformation index of the foundation pit, the excavation characteristics of the foundation pit are taken as input to identify the safety status of the foundation pit using the reconstructed foundation pit geometric model.

[0038] Furthermore, the surface fitting technology is used to perform surface fitting on the excavation features of the foundation pit, and the geometric model of the foundation pit is reconstructed based on the surface fitting results, including:

[0039] The radial basis function is used to separate the excavation features of the foundation pit from the foundation pit image to generate the first layer of surface;

[0040] The features after removing the pit excavation features from the pit image are used as the second layer of surfaces, and the first layer of surfaces and the second layer of surfaces are reconstructed respectively;

[0041] Calculate the non-pit excavation feature points at the junction of the first layer surface and the second layer surface, and obtain the influence factors of the non-pit excavation feature points in the first layer surface and the second layer surface;

[0042] Based on the influencing factors, the first layer surface and the second layer surface are fitted to form the foundation pit excavation surface, and the foundation pit excavation surface is transformed into three-dimensional space to realize the reconstruction of the foundation pit geometric model.

[0043] Furthermore, the risk prediction module includes a hazard source classification module, a risk probability prediction module and a risk assessment module;

[0044] The hazard source classification module is used to classify the identified hazard sources into different hazard levels, where the hazard sources include abnormal behavior and foundation pit collapse;

[0045] The risk probability prediction module is used to predict the probability of occurrence of dangerous sources of different hazard levels;

[0046] The risk assessment module is used to evaluate the impact of a hazard after it actually occurs based on the predicted results of the hazard occurrence probability.

[0047] Furthermore, the expression of the influence factor of the non-excavation feature point in the first layer surface and the second layer surface is:

[0048] ;

[0049] In the formula, a i Represents the influence factor of non-pit excavation feature points in the first layer surface; a 0i Represents the influence factor of non-excavation feature points in the second layer surface; n iIndicates the number of non-excavation feature points in the first layer surface; n 0 indicates the number of non-excavation feature points in the second layer surface; Y i It represents the average distance from the non-excavation feature point to the center of the non-excavation feature point in the first layer surface; Y 0 represents the average distance from the non-pit excavation feature point to the center of the non-pit excavation feature point in the second layer surface.

[0050] Furthermore, the expression of the pyramid model is:

[0051] ;

[0052] In the formula, G T ( i , j ) indicates the pyramid T Layer i Row, No. j Column images, G T The pyramid T Layer image; G T-1 The pyramid T - 1 layer image; i represents a row in an image; j represents the columns in the image; ω represents the weight function; m represents the row in the weight function; n Represents the columns in the weight function.

[0053] According to another aspect of the present invention, a method for identifying hazardous sources at a construction site based on a drone is provided. The method for identifying hazardous sources at a construction site based on a drone comprises the following steps:

[0054] S1. Use the UAV group to carry remote sensing equipment to collect remote sensing image data of the construction site and pre-process the remote sensing image data;

[0055] S2, extracting hazard source features from the pre-processed remote sensing image data, and identifying hazard sources at the construction site based on the hazard source features;

[0056] S3. Classify the identified hazards and predict the probability of occurrence of hazards of different levels;

[0057] S4. Develop corresponding emergency response measures based on the probability of occurrence of different levels of hazardous sources.

[0058] Compared with the prior art, the present invention provides a construction site hazard source identification system and method based on drones, which has the following beneficial effects:

[0059] (1) The present invention analyzes the human motion trajectory through optical flow technology and pyramid model, and can accurately track and analyze the specific movement paths and behavior patterns of personnel at the construction site. By comparing the actual behavior trajectory with the preset safety distance, it can determine whether the personnel have entered the dangerous area, thereby providing effective safety warnings. At the same time, by introducing allocation factors and priority functions, the search weights and priorities of each motion path can be dynamically adjusted to ensure that the most likely dangerous behaviors and areas are monitored.

[0060] (2) The present invention extracts point cloud data from construction images and aligns the point cloud data, thereby ensuring that data from different image sources are accurately aligned in the same coordinate system. At the same time, radial basis functions and surface fitting techniques are used to accurately reconstruct the three-dimensional geometric model of the foundation pit. Thus, by constructing a three-dimensional geometric model of the foundation pit, the safety status of the foundation pit can be evaluated in real time, potential collapse risks can be discovered in time, and early warning of foundation pit collapse or other structural problems can be given.

[0061] (3) The present invention accurately extracts the characteristics of hazard sources from the pre-processed remote sensing image data, and classifies the hazard sources according to their characteristics and impact range, while predicting the probability of occurrence of hazard sources of different levels, thereby facilitating risk management and resource allocation. Timely warning and emergency response help prevent and reduce construction accidents and protect the lives of personnel and construction equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0063] Figure 1 is a principle block diagram of a construction site hazard source identification system based on a drone according to an embodiment of the present invention;

[0064] Figure 2 The present invention is a flowchart of a method for identifying construction site hazards based on a drone according to an embodiment of the present invention.

[0065] In the figure:

[0066] 1. Construction data collection module; 2. Hazard source identification module; 3. Risk prediction module; 4. Risk warning module. DETAILED DESCRIPTION

[0067] To further illustrate each embodiment, the present invention provides drawings, which are part of the disclosure of the present invention and are mainly used to illustrate the embodiments. They can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. With reference to these contents, ordinary technicians in this field should be able to understand other possible implementation methods and advantages of the present invention.

[0068] According to an embodiment of the present invention, a construction site hazard source identification system and method based on an unmanned aerial vehicle are provided.

[0069] The present invention is further described with reference to the accompanying drawings and specific embodiments. Figure 1 As shown, according to the construction site hazard source identification system based on drones according to an embodiment of the present invention, the construction site hazard source identification system includes a construction data acquisition module 1, a hazard source identification module 2, a risk prediction module 3 and a risk warning module 4;

[0070] The construction data acquisition module 1 is used to collect remote sensing image data of the construction site using a drone group equipped with remote sensing equipment, and pre-process the remote sensing image data.

[0071] It should be noted that the construction data acquisition module 1 includes the following steps when using the unmanned aerial vehicle group to carry remote sensing equipment to collect remote sensing image data of the construction site and preprocess the remote sensing image data:

[0072] Flight planning and mission setting: Design the UAV’s flight route based on the size and characteristics of the construction site to ensure full coverage of the required monitoring area; set the UAV’s flight altitude, speed and camera parameters to ensure the required image resolution and data quality.

[0073] Data collection: The drone flies along a predetermined route, uses the remote sensing equipment on board (such as high-resolution cameras, multispectral sensors, etc.) to capture images, and records the data captured by the remote sensing equipment and related flight log information, such as timestamp, location coordinates, and flight altitude.

[0074] Data transmission: After the drone completes its flight mission, it downloads the image data stored in the device to the computing system and transmits the data to the data processing center via wireless transmission or physical storage devices.

[0075] Data preprocessing: Perform geometric and photometric correction on the captured images to eliminate distortion caused by drone movement, camera angle changes, etc.; improve image quality and enhance the visual effects of specific frequency bands to better identify specific features; remove noise from the image to improve image clarity and analysis accuracy.

[0076] The hazard source identification module 2 is used to extract hazard source features from the pre-processed remote sensing image data, and identify hazard sources at the construction site based on the hazard source features.

[0077] Among them, the hazard source identification module 2 includes a hazard area calibration module, a hazard source feature extraction module, an abnormal behavior identification module and a foundation pit collapse identification module;

[0078] The dangerous area calibration module is used to extract continuous image sequences from the processed remote sensing image data, and to segment and calibrate specific dangerous areas from each construction image.

[0079] The dangerous area calibration module extracts a continuous image sequence from the processed remote sensing image data, and segments and calibrates specific dangerous areas from each construction image, including:

[0080] A continuous image sequence is extracted from the processed remote sensing image data, and the construction area is segmented from each construction image in the continuous image sequence using image segmentation technology.

[0081] It should be noted that the image segmentation technology used to segment the construction area from each construction image in the continuous image sequence includes:

[0082] Threshold segmentation: The image is divided into different areas according to the threshold of the pixel value. It is suitable for scenes with high image contrast and obvious difference between the target and the background.

[0083] Region-based segmentation: Grouping regions with similar attributes (such as color, texture) in an image together, which includes region growing and region merging.

[0084] Edge-based segmentation: Segmentation is performed based on the edge information in the image, and gradient operators (such as Sobel and Canny operators) are usually used to detect edges.

[0085] Construct a 3D view based on the construction area, and combine the terrain data of the construction area to display the mapping relationship between construction terrain changes and construction activities in real time in the 3D view.

[0086] It should be noted that the construction of a three-dimensional view based on the construction area and the real-time display of the mapping relationship between the construction terrain changes and the construction activities in the three-dimensional view in combination with the terrain data of the construction area include:

[0087] 3D model construction: Use terrain data to create a 3D terrain model and fuse the terrain data with construction activity data to ensure accurate alignment of the two in 3D space.

[0088] 3D visualization: Provide visualization of 3D models using ArcGIS 3D, Unity 3D or other 3D visualization software;

[0089] Scene setup: Set lighting, texture, perspective, etc. in the 3D model to create a realistic 3D scene, and import the 3D terrain model and construction element model into the 3D scene for scene layout.

[0090] Real-time data integration: Integrate real-time monitoring data streams into 3D scenes, such as dynamically updating machinery positions and construction progress, mapping construction activities in 3D scenes, and distinguishing different types of construction activities through visual means such as colors and markings.

[0091] Based on the mapping relationship between construction terrain changes and construction activities, specific dangerous areas are marked in the three-dimensional view, where specific dangerous areas include construction excavation areas, elevated operation areas and lifting operation areas.

[0092] It should be noted that the construction excavation area mainly refers to the area where earthwork excavation is carried out in the construction site. These areas may be used for building foundations, pipeline laying and basement construction. The excavation depth and scope vary according to project requirements and may involve shallow excavation or deep foundation pit operations. Foundation pit collapse refers to the collapse of the soil wall or supporting structure of the foundation pit during the foundation pit excavation and construction process. This phenomenon usually occurs in deep foundation pit projects.

[0093] The hazard source feature extraction module is used to extract the key features of the hazard source from the construction image, where the key features include human motion features and foundation pit excavation features;

[0094] The abnormal behavior recognition module is used to analyze the safety distance from the human movement characteristics to the specific danger area at the construction site, and judge whether there is any abnormality in the human behavior based on the safety distance.

[0095] Among them, the abnormal behavior recognition module analyzes the safety distance from the human movement characteristics to the specific dangerous area in the construction site, and judges whether there is abnormality in the human behavior based on the safety distance, including:

[0096] Differential processing is performed on each construction image in the continuous image sequence to obtain a processed frame difference image, and the frame difference image is screened into a first image and a second image.

[0097] It should be noted that performing differential processing on each construction image in the continuous image sequence to obtain a processed frame difference image, and screening the frame difference image into a first image and a second image includes:

[0098] Differential processing: Calculate the difference between two adjacent frames in a continuous image sequence by subtracting the pixel values ​​of adjacent frames to obtain a differential image; take the absolute value of the differential result to ensure that all pixel values ​​are non-negative.

[0099] Classification and screening: Use the connected component labeling algorithm to identify and label each independent area in the binary frame difference image, take the characteristics of each labeled area, such as area size, shape, position, etc., and select specific areas as the area of ​​interest according to the set standards (such as area threshold, position, etc.), and classify them into the first image and the second image.

[0100] The first image and the second image are subjected to image binarization processing respectively, and logical operations are performed bit by bit to obtain the human body movement area in the construction site.

[0101] It should be noted that the first image and the second image are subjected to image binarization processing respectively, and logical operations are performed bit by bit to obtain the human body movement area in the construction site, which includes:

[0102] Image binarization: Convert the first image and the second image from color to grayscale, and apply threshold processing to the grayscale image to convert the image into a binary image. Usually, pixels exceeding a certain brightness threshold are set to 1 (white) and others are set to 0 (black).

[0103] Bitwise logical operation: Select the logical operation based on the image comparison goal. For example, use the "exclusive OR" (XOR) operation to identify the changed area between two images; perform a bitwise XOR operation on the first image and the second image. In the XOR operation, the result is 1 (indicating change) only when the brightness of the corresponding pixels is different, which helps to highlight the moving human body.

[0104] A pyramid model is established based on the human body motion area, and the optical flow estimation value is initialized at the lowest layer of the pyramid model; the optical flow values ​​of each layer of the pyramid model are calculated layer by layer from the lowest layer of the pyramid model.

[0105] It should be noted that the pyramid model is a multi-resolution image representation method that creates a series of reduced images by gradually reducing the resolution to form a pyramid structure. The image at the top layer has the lowest resolution, while the bottom layer maintains the original resolution. The pyramid model allows the algorithm to quickly capture global features at a lower resolution, and then gradually fine-tune to the high-resolution layer, thereby improving computational efficiency and processing speed.

[0106] A pyramid model is established based on the human body motion area, and the optical flow estimation value is initialized at the lowest layer of the pyramid model; the optical flow values ​​of each layer of the pyramid model are calculated layer by layer from the lowest layer of the pyramid model, including:

[0107] Starting from the original resolution, the image resolution is gradually reduced, usually by half at each layer, until a predetermined number of layers or a minimum resolution is reached; each layer is a scaled-down version of the previous layer, and these layers are stored in a pyramid structure for subsequent optical flow calculations.

[0108] Optical flow estimation starts at the bottom layer (lowest resolution) of the pyramid, where the optical flow is initialized to zero or a preliminary estimate is made based on the prediction model, and an optical flow algorithm (such as the Lucas-Kanade method or the Horn-Schunck algorithm) is used to calculate the motion between pixels at each layer.

[0109] After calculating the optical flow of each layer, the estimated value is upsampled to a higher resolution level, refined and adjusted at that level, and this process is repeated at each level, gradually optimizing the optical flow vector until the top layer (original resolution) is reached.

[0110] Among them, the expression of the pyramid model is:

[0111] ;

[0112] In the formula, G T ( i , j ) indicates the pyramid T Layer i Row, No. j Column images, G T The pyramid T Layer image; G T-1 The pyramid T - 1 layer image; i represents a row in an image; j represents the columns in the image; ω represents the weight function; m represents the row in the weight function; n Represents the columns in the weight function.

[0113] Add the optical flow values ​​of each layer of the pyramid model and locate the human feature points in each frame difference image to identify the human motion trajectory;

[0114] Obtain a feasible path for the human motion trajectory to reach a specific dangerous area, and calculate the distance from the human motion trajectory on the feasible path to the specific dangerous area.

[0115] Wherein, obtaining a feasible path for the human motion trajectory to reach a specific dangerous area and calculating the distance from the human motion trajectory on the feasible path to the specific dangerous area includes:

[0116] Combined with the terrain changes in the construction area, the movement path of human body to the specific dangerous area is identified from the 3D view;

[0117] The motion paths within the preset range are relaxed, and a distribution factor is introduced to dynamically distribute the search weight of each motion path according to the search tree structure.

[0118] It should be noted that the motion paths within the preset range are relaxed, and the allocation factor is introduced to dynamically allocate the search weights of each motion path according to the search tree structure, including:

[0119] Initialization and path relaxation: Set a preset range of the motion path based on historical data or predetermined motion patterns, allowing the path to have a certain degree of flexibility within the preset range to adapt to possible changes or unforeseen environmental factors.

[0120] Build a search tree: Based on the preset range and relaxation conditions of the motion path, a search tree is constructed. Each node represents a motion decision or path change point, and defines a distribution factor, which is used to dynamically adjust the search weight of each path according to the possibility and importance of the path.

[0121] Weight allocation: The search weight of each path is calculated using the allocation factor and other relevant parameters (such as the risk assessment and historical efficiency of the path), and the weight of each node in the search tree is dynamically adjusted based on real-time data and environmental feedback.

[0122] The trust value of each motion path is evaluated based on the search weight of the motion path, and a priority function is set for each motion path node.

[0123] It should be noted that the priority function is a method used in the search algorithm to determine the processing order. It assigns a value to each node, which represents the priority of the node. In the search and decision-making based on the motion path, the priority function helps the algorithm to effectively select the best or most likely path.

[0124] The priority function usually includes the following key parameters:

[0125] (1) Cost: The known cost from the starting point to the current node.

[0126] (2) Heuristic estimate (if applicable): the estimated cost from the current node to the goal.

[0127] (3) Risk assessment: the risks that the path may involve, such as the probability of encountering obstacles or safety risks.

[0128] (4) Historical data: Past experience or historical efficiency data helps improve path selection.

[0129] (5) Application of priority function: trust value evaluation of motion path.

[0130] When setting the priority function, take the above key parameters into consideration and create a function to dynamically adjust the weight of each node in the search tree. For example, in a navigation system for an autonomous vehicle, the trust value evaluation includes:

[0131] Parameter definition: Define various parameters that affect the trust value, such as road information, traffic conditions, weather conditions, etc.

[0132] Calculate initial trust value: Based on the known information at startup, calculate an initial trust value for each path node, which can be based on risk assessment, estimated time of arrival (ETA) and historical data.

[0133] Construction of priority function: Define a function that combines the node's cost, risk assessment, and other relevant parameters (such as historical traffic efficiency) to calculate the priority of each node.

[0134] For example, the function might be f(n) = g(n) + h(n) + r(n), where g(n) is the actual cost from the starting point to node n; h(n) is a heuristic estimate of the cost from node n to the goal; and r(n) is the risk estimate.

[0135] The motion path node with the smallest priority is selected to update the priority function. Based on the updated priority function, the feasible path from the human motion trajectory to the specific dangerous area is determined and the distance is calculated.

[0136] It should be noted that when determining the feasible path from the human body's motion trajectory to a specific dangerous area and calculating the distance, using Euclidean distance as a measurement method is a common and intuitive choice. By calculating and monitoring the distance from the human body to the dangerous area in real time, potential safety accidents can be prevented, and monitoring resources can be concentrated by selecting the shortest or most likely path to improve monitoring efficiency; distance data can be used to assist decision-making, such as arranging timely patrols or adjusting the on-site layout to avoid accidents.

[0137] The distance from the human body's motion trajectory to a specific dangerous area is compared with the preset safety distance, and based on the comparison result, it is determined whether there is any abnormality in the human behavior.

[0138] The foundation pit collapse identification module is used to take the foundation pit excavation features as input to perform surface fitting and point cloud reconstruction in sequence to identify the safety status of the foundation pit.

[0139] Among them, the foundation pit collapse identification module takes the foundation pit excavation features as input and performs surface fitting and point cloud reconstruction in sequence to identify the safety status of the foundation pit, including:

[0140] Filter out foundation pit images from construction images, and convert foundation pit feature points in the foundation pit images into point cloud data;

[0141] The iterative closest point algorithm is used to align the point cloud data in each foundation pit image to ensure that all point cloud data are aligned in the same coordinates.

[0142] It should be noted that using the Iterative Closest Point (ICP) algorithm to align the point cloud data in each foundation pit image is an effective method to ensure that all point cloud data are aligned in the same coordinate system. The ICP algorithm is mainly used to minimize the distance between two point sets so that one point set is aligned with the other point set as much as possible. The following are the steps for point cloud data alignment using the ICP algorithm:

[0143] Select target and source point clouds: From multiple point cloud sets, select one point cloud as the target (fixed) and the others as sources (need to be moved and rotated to match the target), and provide an initial pose estimate.

[0144] Iterative registration process: In each iteration, find the nearest point in the target point cloud for each point in the source point cloud; calculate the distance between all point pairs and define an error metric (usually the squared difference); minimize the total error between all paired points by rotating and translating the source point cloud, which is done by solving a least squares problem; update the pose (i.e., position and orientation) of the source point cloud, and repeat the nearest point matching and error minimization process.

[0145] End condition: When the iteration reaches the preset maximum number or when the error between point clouds drops below a certain threshold, the algorithm stops and checks the quality of the final point cloud alignment to ensure that all point cloud data are correctly aligned.

[0146] The surface fitting technology is used to fit the excavation features of the foundation pit, and the foundation pit geometric model is reconstructed based on the surface fitting results.

[0147] Among them, the surface fitting technology is used to perform surface fitting on the excavation features of the foundation pit, and the geometric model of the foundation pit is reconstructed based on the surface fitting results, including:

[0148] The radial basis function is used to separate the foundation pit excavation features from the foundation pit image to generate the first layer of surface.

[0149] It should be noted that the radial basis function is a function used in many applications, especially in interpolation and neural networks. RBF (Radial Basis Function) is a real-valued function centered at the origin and whose influence range is adjusted as the distance changes. The radial basis function is used to separate the foundation pit excavation features from the foundation pit image, and the first layer of the surface generated includes:

[0150] Parameter configuration: Set the extended parameter ϵ of RBF, which affects the smoothness and fitting accuracy of the model.

[0151] Interpolation calculation: Use RBF to interpolate the selected feature points to generate a three-dimensional surface that describes the excavation characteristics of the foundation pit.

[0152] Surface optimization: adjust parameters and center point positions to optimize the surface to better conform to the actual pit characteristics.

[0153] The features after removing the pit excavation features from the pit image are used as the second layer of surfaces, and the first layer of surfaces and the second layer of surfaces are reconstructed respectively;

[0154] The non-pit excavation feature points at the junction of the first layer surface and the second layer surface are calculated, and the influence factors of the non-pit excavation feature points in the first layer surface and the second layer surface are obtained.

[0155] Among them, the expression of the influence factor of the non-excavation feature point in the first layer surface and the second layer surface is:

[0156] ;

[0157] In the formula, a i Represents the influence factor of non-pit excavation feature points in the first layer surface; a 0i Represents the influence factor of non-excavation feature points in the second layer surface; n i Indicates the number of non-excavation feature points in the first layer surface; n 0 indicates the number of non-excavation feature points in the second layer surface; Y i It represents the average distance from the non-excavation feature point to the center of the non-excavation feature point in the first layer surface; Y 0 represents the average distance from the non-pit excavation feature point to the center of the non-pit excavation feature point in the second layer surface.

[0158] Based on the influencing factors, the first layer surface and the second layer surface are fitted to form the foundation pit excavation surface, and the foundation pit excavation surface is transformed into three-dimensional space to realize the reconstruction of the foundation pit geometric model.

[0159] Based on the preset local deformation index of the foundation pit, the excavation characteristics of the foundation pit are taken as input to identify the safety status of the foundation pit using the reconstructed foundation pit geometric model.

[0160] The risk prediction module 3 is used to classify the identified hazards and predict the probability of occurrence of hazards of different levels.

[0161] Among them, the risk prediction module 3 includes a hazard source level classification module, a risk probability prediction module and a risk assessment module;

[0162] The hazard source classification module is used to classify the identified hazard sources into different hazard levels, where the hazard sources include abnormal behavior and foundation pit collapse;

[0163] The risk probability prediction module is used to predict the probability of occurrence of hazardous sources of different hazard levels.

[0164] The risk assessment module is used to evaluate the impact of a hazard after it actually occurs based on the predicted results of the hazard occurrence probability.

[0165] It should be noted that the impact of the actual occurrence of a hazard source is assessed based on the prediction results of the hazard source occurrence probability, including:

[0166] Select a prediction model based on data characteristics and problem features. Prediction models include decision trees, random forests, neural networks, etc.; use training data sets to train the model and adjust parameters to achieve optimal prediction performance; test the stability and generalization ability of the model through cross-validation technology; map the occurrence probability predicted by the model to a specific risk level; determine the threshold for assigning the predicted probability to different risk levels.

[0167] The risk warning module 4 is used to formulate corresponding emergency response measures based on the probability of occurrence of different levels of dangerous sources.

[0168] Low risk level: Set up early warning notifications, such as notifying relevant personnel of risk status via SMS or email.

[0169] Medium risk level: Activate the automatic monitoring system to enhance monitoring and detection.

[0170] High risk level: Real-time sound and light alarm, activate the emergency broadcast system, and quickly notify all personnel.

[0171] Suppose a chemical plant has set up different emergency response levels based on the probability of occurrence of hazardous sources:

[0172] Low risk (less than 10% probability):

[0173] Measures: Increase the frequency of on-site safety inspections and strengthen monitoring of raw material storage areas.

[0174] Implementation: Carry out an additional safety inspection per shift to ensure that all monitoring equipment is operating properly.

[0175] Medium risk (probability 10%-50%):

[0176] Measures: Activate the accident warning system and conduct evacuation drills in some areas.

[0177] Implementation: Conduct evacuation drills for potential spill areas monthly to ensure employees are familiar with emergency evacuation routes.

[0178] High risk (probability greater than 50%):

[0179] Measures: Comprehensively check the safety facilities of the production line and implement full evacuation in emergency situations.

[0180] Implementation: Immediately organize a safety check for all employees and immediately fix any problems found; ensure that the emergency evacuation plan and emergency contact list are updated and effective.

[0181] It should be noted that the corresponding emergency response measures based on the probability of occurrence of different levels of dangerous sources include:

[0182] like Figure 2 As shown, according to another embodiment of the present invention, a method for identifying hazardous sources at a construction site based on a drone is also provided. The method for identifying hazardous sources at a construction site based on a drone comprises the following steps:

[0183] S1. Use the UAV group to carry remote sensing equipment to collect remote sensing image data of the construction site and pre-process the remote sensing image data;

[0184] S2, extracting hazard source features from the pre-processed remote sensing image data, and identifying hazard sources at the construction site based on the hazard source features;

[0185] S3. Classify the identified hazards and predict the probability of occurrence of hazards of different levels;

[0186] S4. Develop corresponding emergency response measures based on the probability of occurrence of different levels of hazardous sources.

[0187] In summary, with the help of the above-mentioned technical scheme of the present invention, the present invention analyzes the human motion trajectory through optical flow technology and pyramid model, can accurately track and analyze the specific movement path and behavior pattern of personnel at the construction site, and by comparing the actual behavior trajectory with the preset safety distance, can determine whether the personnel have entered the dangerous area to provide effective safety warnings, and at the same time, by introducing allocation factors and priority functions, can dynamically adjust the search weight and priority of each motion path to ensure that the most likely dangerous behaviors and areas are monitored; the present invention extracts point cloud data from the construction image and aligns the point cloud data, which can ensure that data from different image sources are in the same coordinate system. The three-dimensional geometric model of the foundation pit can be accurately reconstructed by using radial basis function and surface fitting technology. Therefore, by constructing the three-dimensional geometric model of the foundation pit, the safety status of the foundation pit can be evaluated in real time, and the potential collapse risk can be discovered in time, so as to give early warning of foundation pit collapse or other structural problems. The present invention accurately extracts the characteristics of the hazard source from the pre-processed remote sensing image data, and classifies the hazard source according to the characteristics and influence range of the hazard source, and predicts the occurrence probability of hazard sources of different levels, so as to help risk management and resource allocation. Timely warning and emergency response are helpful to prevent and reduce the occurrence of construction accidents and protect the lives of personnel and construction equipment.

[0188] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A construction site hazard identification system based on drones, characterized in that: The construction site hazard source identification system includes a construction data acquisition module, a hazard source identification module, a risk prediction module and a risk warning module; The construction data acquisition module is used to collect remote sensing image data of the construction site using the remote sensing equipment carried by the drone group, and pre-process the remote sensing image data; The hazard source identification module is used to extract hazard source features from the pre-processed remote sensing image data, and identify hazard sources at the construction site based on the hazard source features; The risk prediction module is used to classify the identified hazards and predict the probability of occurrence of hazards of different levels; The risk warning module is used to formulate corresponding emergency response measures based on the probability of occurrence of different levels of dangerous sources; The hazard source identification module includes a hazard area calibration module, a hazard source feature extraction module, an abnormal behavior identification module and a foundation pit collapse identification module; The dangerous area demarcation module is used to extract a continuous image sequence from the processed remote sensing image data, and to segment and demarcate specific dangerous areas from each construction image; The hazard source feature extraction module is used to extract key features of the hazard source from the construction image, wherein the key features include human motion features and foundation pit excavation features; The abnormal behavior recognition module is used to analyze the safety distance from the human movement characteristics to the specific dangerous area in the construction site, and determine whether the human behavior is abnormal based on the safety distance; The foundation pit collapse identification module is used to take the foundation pit excavation features as input to perform surface fitting and point cloud reconstruction in sequence to identify the safety status of the foundation pit; The foundation pit collapse identification module takes the foundation pit excavation features as input and performs surface fitting and point cloud reconstruction in sequence to identify the safety status of the foundation pit, including: Filter out foundation pit images from construction images, and convert foundation pit feature points in the foundation pit images into point cloud data; The point cloud data in each foundation pit image are registered using the iterative closest point algorithm to ensure that all point cloud data are aligned in the same coordinates; The surface fitting technology is used to fit the excavation features of the foundation pit, and the geometric model of the foundation pit is reconstructed based on the surface fitting results; Based on the preset local deformation index of the foundation pit, the excavation characteristics of the foundation pit are used as input to identify the safety status of the foundation pit using the reconstructed foundation pit geometric model. The surface fitting of the foundation pit excavation features by using the surface fitting technology and reconstructing the foundation pit geometric model based on the surface fitting result comprises: The radial basis function is used to separate the excavation features of the foundation pit from the foundation pit image to generate the first layer of surface; The features after removing the pit excavation features from the pit image are used as the second layer of surfaces, and the first layer of surfaces and the second layer of surfaces are reconstructed respectively; Calculate the non-pit excavation feature points at the junction of the first layer surface and the second layer surface, and obtain the influence factors of the non-pit excavation feature points in the first layer surface and the second layer surface; Based on the influencing factors, the first layer surface and the second layer surface are fitted to form a foundation pit excavation surface, and the foundation pit excavation surface is transformed into a three-dimensional space to achieve reconstruction of the foundation pit geometric model; The expression of the influence factor of the non-foundation pit excavation feature point in the first layer surface and the second layer surface is: ; In the formula, a i Represents the influence factor of non-pit excavation feature points in the first layer surface; a 0i Represents the influence factor of non-excavation feature points in the second layer surface; n i Indicates the number of non-excavation feature points in the first layer surface; n 0 indicates the number of non-excavation feature points in the second layer surface; Y i It represents the average distance from the non-pit excavation feature point to the center of the non-pit excavation feature point in the first layer surface; Y 0 represents the average distance from the non-excavation feature point to the center of the non-excavation feature point in the second layer surface; The abnormal behavior recognition module analyzes the safety distance from the human movement characteristics to the specific dangerous area in the construction site, and determines whether the human behavior is abnormal based on the safety distance, including: Performing differential processing on each construction image in the continuous image sequence to obtain a processed frame difference image, and screening the frame difference image into a first image and a second image; Performing image binarization processing on the first image and the second image respectively, and performing logical operation bit by bit to obtain the human body movement area in the construction site; A pyramid model is established based on the human body motion area, and the optical flow estimation value is initialized at the lowest layer of the pyramid model; the optical flow value of each layer of the pyramid model is calculated layer by layer from the lowest layer of the pyramid model; Add the optical flow values ​​of each layer of the pyramid model and locate the human feature points in each frame difference image to identify the human motion trajectory; Obtain a feasible path for the human motion trajectory to reach a specific dangerous area, and calculate the distance from the human motion trajectory on the feasible path to the specific dangerous area; Compare the distance from the human body's motion trajectory to a specific dangerous area with the preset safety distance, and determine whether the person's behavior is abnormal based on the comparison result; The step of obtaining a feasible path for the human motion trajectory to reach the specific dangerous area and calculating the distance from the human motion trajectory on the feasible path to the specific dangerous area includes: Combined with the terrain changes in the construction area, the movement path of human body to the specific dangerous area is identified from the 3D view; The motion paths within the preset range are relaxed, and a distribution factor is introduced to dynamically distribute the search weight of each motion path according to the search tree structure; Based on the search weight of the motion path, the trust value of each motion path is evaluated respectively, and the priority function is set for each motion path node; The motion path node with the smallest priority is selected to update the priority function. Based on the updated priority function, the feasible path from the human motion trajectory to the specific dangerous area is determined and the distance is calculated.

2. The construction site hazard source identification system based on drone according to claim 1 is characterized in that: The dangerous area demarcation module extracts a continuous image sequence from the processed remote sensing image data, and segments and demarcates a specific dangerous area from each construction image, including: Extracting a continuous image sequence from the processed remote sensing image data, and segmenting the construction area from each construction image in the continuous image sequence using image segmentation technology; Construct a 3D view based on the construction area, and combine the terrain data of the construction area to display the mapping relationship between construction terrain changes and construction activities in real time in the 3D view; Based on the mapping relationship between construction terrain changes and construction activities, specific dangerous areas are marked in the three-dimensional view, where the specific dangerous areas include construction excavation areas, elevated operation areas and lifting operation areas.

3. The construction site hazard source identification system based on drone according to claim 2 is characterized in that: The risk prediction module includes a hazard source level classification module, a risk probability prediction module and a risk assessment module; The hazard source classification module is used to classify the identified hazard sources into hazard levels, wherein the hazard sources include abnormal behavior and foundation pit collapse; The risk probability prediction module is used to predict the probability of occurrence of risk sources of different risk levels; The risk assessment module is used to assess the impact of a hazard after it actually occurs based on the prediction result of the hazard occurrence probability.

4. The construction site hazard source identification system based on drone according to claim 3 is characterized in that: The expression of the pyramid model is: ; In the formula, G T ( i , j ) indicates the pyramid T Layer i Row, No. j Column images, G T The pyramid T Layer image; G T-1 The pyramid T - 1 layer image; i represents a row in an image; j represents the columns in the image; ω represents the weight function; m represents the row in the weight function; n Represents the columns in the weight function.

5. A method for identifying hazardous sources at a construction site based on an unmanned aerial vehicle, used to implement the system for identifying hazardous sources at a construction site based on an unmanned aerial vehicle as claimed in any one of claims 1 to 4, characterized in that: The construction site hazard source identification method based on drone includes the following steps: S1. Use the UAV group to carry remote sensing equipment to collect remote sensing image data of the construction site and pre-process the remote sensing image data; S2, extracting hazard source features from the pre-processed remote sensing image data, and identifying hazard sources at the construction site based on the hazard source features; S3. Classify the identified hazards and predict the probability of occurrence of hazards of different levels; S4. Develop corresponding emergency response measures based on the probability of occurrence of different levels of hazardous sources.

Images