A construction site fire operation safety supervision method using tower crane video image detection

Abstract

The present application relates to a kind of construction site fire operation safety supervision method using tower crane video image detection, comprising the following steps: extracting construction site fire operation information;Set up fire operation safety supervision area;Fire operation safety supervision tower crane hook video effective data acquisition;Fire operation safety supervision tower crane video image detection;Fire operation safety live judgment and risk early warning.The present application is a kind of construction site fire operation safety supervision method using tower crane video image detection, with camera pose data as prior condition, filters camera effective image, reduces image processing quantity and calculation cost;Image is automatically detected and safety discrimination using deep learning algorithm, improves safety supervision efficiency, realizes intelligent supervision of fire operation;Using construction site camera data, without increasing additional process, from low altitude angle, supervises the abnormal situation that may occur in construction site fire operation, enriches supervision means.

Description

Technical Field

[0001] This invention relates to the field of building construction, and specifically to a method for safety supervision of hot work operations on construction sites using tower crane video image detection. Background Technology

[0002] Hot work refers to any activity involving open flames, generating heat or sparks that may lead to fire or explosion, including electrical work performed in flammable and explosive environments. On construction sites, hot work is a significant cause of fires. Currently, the entire hot work process involves: the development and approval of a hot work plan, the requirements and inspection of equipment and personnel, standardized on-site management, hot work execution, and the removal of personnel and tools from the site. Regarding the supervision of hot work, regulations require dedicated personnel to oversee the process; however, supervision relies heavily on the experience of these personnel, leading to significant subjectivity and insufficient oversight. This makes it difficult to detect many potential hazards during the operation, and the entire hot work process cannot be fully recorded and traced. Therefore, relying on safety regulations or human experience to inspect for safety hazards on construction sites is no longer sufficient to meet the complex and dynamic requirements of hot work safety management.

[0003] With the practical application of IoT technology in construction engineering, various sensors and network-connected intelligent devices perceive the on-site environment, capture and transmit large amounts of data for early risk identification and safety management. The deployment of multiple sensors in IoT technology can not only prevent employees from being exposed to hazardous environments but also detect such situations before or during their occurrence; however, applications are mostly limited to staff remotely viewing video surveillance. Currently, some researchers are attempting to automatically identify abnormal situations during hot work operations using fixed cameras, but due to limitations in the distribution and field of view of fixed cameras, it is difficult to achieve automatic monitoring of any hot work location within the construction site, limiting its widespread application.

[0004] Existing technology CN216982008U discloses a safe operation intelligent terminal, including a work area enclosure assembly and a power supply enclosure. The power supply enclosure is located below the work area enclosure assembly, and the power supply enclosure and the work area enclosure assembly are interconnected. The work area enclosure assembly includes an enclosure fixing plate, a gas sensor enclosure is located above the enclosure fixing plate, a camera enclosure is located above the gas sensor enclosure, and a display screen enclosure is located above the enclosure fixing plate and behind the gas sensor enclosure and the camera enclosure. The gas sensor enclosure, the display screen enclosure, and the camera enclosure are interconnected. The terminal, equipped with a camera and a gas sensor, enables the detection and alarm of gases in the hot work area and the recording of hot work activities. The terminal allows for real-time monitoring of the content of flammable and explosive gases in the environment. However, its camera is fixed, making it difficult to automatically monitor any hot work location within the construction site. Summary of the Invention

[0005] To address the technical problems existing in the prior art, the present invention provides a...

[0006] To achieve the above objectives, the technical solution of the present invention is as follows:

[0007] A method for safety supervision of hot work operations at construction sites using tower crane video image detection includes the following steps:

[0008] Step 1: Extract hot work information from the construction site;

[0009] Step 2: Establish a safety supervision area for hot work operations;

[0010] Step 3: Obtaining valid video data of tower crane hooks for hot work safety supervision;

[0011] Step 4: Safety supervision of hot work operations, including tower crane video image detection;

[0012] Step 5: Determine the safety status and issue a risk warning for hot work operations.

[0013] As a preferred technical solution, step one includes collecting documents and materials related to hot work at the construction site; extracting text information from the documents and materials; and extracting key information about the hot work.

[0014] As a preferred technical solution, extracting key information about hot work operations includes: segmenting text data using an open-source word segmenter to obtain a segmented word set; pre-setting keywords for hot work operations and extracting the text segments containing these keywords to obtain text data strongly related to the hot work operations; setting event trigger words and event elements; selecting several hot work operation text data sets and using a data annotation platform to perform entity and relation annotations on the key information of the hot work operations to construct a small sample training set; converting the annotated data into word vectors and inputting them into an event extraction model for training; verifying the accuracy and recall of the model after small sample training; if both accuracy and recall are greater than a set threshold, the model training is complete; otherwise, repeat the first two steps until the model meets the requirements; and using the trained model to extract key information about the hot work operations, including "hot work operation time," "hot work operation location," "hot work operation level," and "hot work operation type."

[0015] As a preferred technical solution, step two includes:

[0016] Step 2.1: Based on the extracted hot work operation time information, filter out hot work operations that occurred before the current date;

[0017] Step 2.2: Based on the extracted hot work location information, set the spatial coordinates of the hot work point as (X,Y,Z);

[0018] Step 2.3: Set the radius R of the hot work influence range according to the hot work level and type;

[0019] Step 2.4: The circular area with the spatial coordinates (X,Y,Z) of the hot work point as the center and R as the radius is the hot work safety supervision area.

[0020] As a preferred technical solution, step three includes:

[0021] Step 3.1, calculate the camera's field of view.

[0022] Horizontal field of view:

[0023]

[0024] Vertical field of view:

[0025]

[0026] In the formula, f is the focal length of the camera, w is the width of the sensor, and h is the height of the sensor.

[0027] Step 3.2, calculate the pose parameter relationship of the camera on the tower crane:

[0028]

[0029] In the formula, x, y, z are the coordinate pose parameters of the camera, α, β, γ are the angular pose parameters of the camera; (x0, y0) is the rotation center of the tower crane, θ is the rotation angle of the tower crane at a certain moment; the ground height is defined as 0, H0 is the height of the camera above the ground, H is the height of the working surface above the ground; R0 is the distance between the camera and the rotation center of the tower crane, i.e., the rotation radius, and Δh = H0 - H is the height of the camera above the working surface.

[0030] Step 3.3: Calculate the range of values ​​for the rotation angle θ used by the tower crane to obtain valid data:

[0031]

[0032]

[0033] In the formula, X and Y are the coordinates of the hot work point, x0 and y0 are the rotation center of the tower crane, and R is the radius of the hot work influence range;

[0034] Step 3.4: Calculate the range of values ​​for the slewing radius R0 of the camera used by the tower crane to acquire valid data.

[0035]

[0036] In the formula, X and Y are the coordinates of the hot work point, x0 and y0 are the rotation center of the tower crane, and R is the radius of the hot work influence range;

[0037] Step 3.5: When both the tower crane rotation angle θ and the camera rotation radius R0 are within their respective ranges, capture the image data captured by the camera and automatically transmit and process it; otherwise, do not process the data captured by the camera.

[0038] As a preferred technical solution, step four includes setting the category of the object to be detected during hot work, and performing rapid analysis and feature detection on the transmitted image data based on a preset image deep learning model. The input is the image captured and transmitted back by the camera in a specific area, and the output is an image with detection box coordinate information. The output result includes the category of the detected object and the pixel coordinate information of the points in the detection box (x, y, z). i ,y i ).

[0039] As a preferred technical solution, step five includes:

[0040] Step 5.1: Preset the safe distance between different detection objects and the ignition point:

[0041] Step 5.2, determine the distance between different types of detection objects and hot work points: if the distance is less than the preset safe distance, issue a hazard warning and send feedback to the backend;

[0042] Step 5.3: Issue a hazard warning for the detected dangerous conditions and provide feedback to the back-end management personnel for on-site intervention.

[0043] As a preferred technical solution, step 5.2 includes:

[0044] Step 5.2.1, calculate the horizontal coverage length S1 of the camera:

[0045]

[0046] Step 5.2.2, calculate the longitudinal coverage length S2 of the camera:

[0047]

[0048] The rectangle with S1 as the length and S2 as the width represents the camera's monitoring range;

[0049] Step 5.2.3, combining the horizontal coverage length S1, the vertical coverage length S2, and the image size (w) of the camera. j ,h j This yields the true size of the image;

[0050] like The image's true size conversion ratio like The image's true size conversion ratio

[0051] Step 5.2.4: Based on the pixel position of the center point of the ignition point detection box and the pixel position information of the center points of other detection object boxes, determine the distance D of different detection objects from the ignition point:

[0052]

[0053] Step 5.2.5: Determine the actual distance between the object being tested and the ignition point compared to the preset safe distance. If the distance is greater than the preset safe distance, the placement of the object being tested is safe, and no hazard warning is issued; if the distance is less than the preset safe distance, the placement of the object being tested is unsafe, and proceed to step 5.3.

[0054] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0055] This invention proposes a method for safety supervision of hot work operations at construction sites using tower crane video images. It uses camera pose data as a priori condition to filter valid camera images, reducing the amount of image processing and computational costs. Deep learning algorithms are used for automatic image detection and safety assessment, improving safety supervision efficiency and achieving intelligent supervision of hot work operations. Furthermore, by utilizing camera data from the construction site, it monitors potential anomalies during hot work operations from a low-altitude perspective without adding extra procedures, enriching the means of supervising hot work operations at construction sites. Attached Figure Description

[0056] Figure 1 This is a flowchart of a method for safety supervision of hot work operations at construction sites using tower crane video image detection, according to the present invention.

[0057] Figure 2 This is a schematic diagram of the on-site arrangement and image acquisition method of the tower crane camera in a method for safety supervision of hot work operations at construction sites using tower crane video image detection, according to the present invention.

[0058] Figure 3 This is a schematic diagram of the hot work safety supervision area and the effective data range of the tower crane camera in a method for supervising hot work operations at construction sites using tower crane video images, according to the present invention.

[0059] In the image: 1. Tower crane; 2. Camera; 3. Hot work safety supervision area; 4. Camera monitoring range. Detailed Implementation

[0060] The technical solution of the present invention will be further described below with reference to specific embodiments:

[0061] like Figure 1As shown, a method for safety supervision of hot work operations on construction sites using video images from a tower crane 1 includes the following steps:

[0062] Step 1: Extract hot work information from the construction site;

[0063] Step 1.1: Collect relevant documents and materials related to hot work at the construction site;

[0064] Step 1.2: Extract text information from the document. If the document is a text document, extract the text information directly to obtain txt text data. If the document is an image or scanned document (containing printed or handwritten fonts), use text recognition technology to extract the text information to obtain txt text data.

[0065] Step 1.3: Extract key information for hot work operations;

[0066] Step 1.3.1: Use the Jieba library to perform word segmentation on the text data to obtain a word segmentation set;

[0067] Step 1.3.2: Preset keywords for hot work operations: hot work, electric welding, electric cutting, gas welding, gas cutting, cutting, welding, open flame, fire prevention, fire extinguishing, extract the text segments containing matching characters, and obtain text data strongly related to hot work operations;

[0068] Step 1.3.3: Set the event trigger word to "hot work", and the event elements to "hot work time", "hot work location", "hot work level", and "hot work type";

[0069] Step 1.3.4: Select several hot work operation text data, use the data annotation platform docano to perform entity annotation and relation annotation on the key information of the hot work operation, and make the defined entity annotation consistent with the event elements to construct a small sample training set;

[0070] Step 1.3.5: After annotation, export a JSON file on the doccano platform, divide the dataset into training set: validation set: test set ratio of 8:1:1, set the model learning rate to 0.00001, the batch size to 16, specify the task type as "event extraction", randomly shuffle the dataset, and input it into the Universal Event Extraction Model (UIE) for small sample training.

[0071] Step 1.3.6: Verify the accuracy and recall of the model after small sample training using a single-stage evaluation method. If both accuracy and recall are greater than 80%, the model training is complete; otherwise, repeat steps 1.3.4 and 1.3.5 until the model meets the requirements.

[0072] Step 1.3.7: Extract key information for hot work operations using the trained model. This key information includes "hot work operation time," "hot work operation location," "hot work operation level," and "hot work operation type." Taking the extraction of a hot work permit as an example, the extracted event information is as follows:

[0073] "Start a fire",

[0074] {"Hot work period": "7:00 AM - 11:00 AM, July 15, 2022"}

[0075] {"Hot work location": Southeast corner of the twin tower connecting corridor}

[0076] {"Types of Hot Work": "Electric Welding", "Cutting"}

[0077] Step 2, set up a hot work safety supervision area 3;

[0078] Step 2.1: Based on the extracted hot work operation time information, filter out hot work operations that occurred before the current date;

[0079] Step 2.2: Based on the extracted hot work location information: "Southeast corner of the twin tower corridor", set the spatial coordinates of the hot work point as (19485,2630,0);

[0080] Step 2.3: Based on the type of hot work: "electric welding" or "cutting", set the radius of influence of the hot work operation to 30m.

[0081] Step 2.4: Based on the location of the hot work point and the impact range of the hot work, obtain the hot work safety supervision area 3, such as... Figure 3 As shown, the working surface is the ground, with a center point of (19485, 2630) and a circular area with a radius of 30m;

[0082] Step 3: Obtain valid data from the video of the tower crane hook during hot work safety supervision;

[0083] Step 3.1, in this embodiment, as follows: Figure 2 As shown, camera 2 is fixedly mounted on the crane trolley and moves along the boom with the crane trolley; the focal length of camera 2 is f = 8.6mm, and the sensor size is w*h = 17.3mm*13mm. Then the field of view of tower crane 1 is;

[0084] Horizontal field of view:

[0085]

[0086] Vertical field of view:

[0087]

[0088] In the formula, f is the focal length of camera 2, w is the width of the sensor, and h is the height of the sensor.

[0089] Step 3.2, the rotation center of tower crane 1 is (x0, y0) = (19450, 2680). Calculate the pose parameters of camera 2 on tower crane 1: the three coordinate poses of camera 2 are x, y, z, the three angular poses are α, β, γ, the pitch angle of camera 2 is vertically downward γ0 = 90°, the ground height is 0, the height of camera 2 is H0 = 55m, the height of the working surface is H = 0, the height of camera 2 from the working surface is Δh = H0 - H = 55m, the rotation angle of the tower crane at a certain moment is θ (θ changes continuously as the tower crane works), and the distance between camera 2 and the rotation center of tower crane 1, i.e., the rotation radius, is R0 = 50m.

[0090]

[0091] Step 3.3: Based on the coordinates of the hot work safety supervision area 3 and the coordinates of the tower crane's slewing center, calculate the range of the rotation angle θ for the tower crane 1 to obtain valid data: θ∈(25.5°, 84.5°);

[0092] Step 3.4: Calculate the range of values ​​for the rotation radius R0 of camera 2 that acquires valid data from tower crane 1. The value of R0 is R0∈(31m,91m); where the boom length of tower crane 1 is 65m, so the actual range of values ​​for the rotation radius R0 of camera 2 is R0∈(31m,65m).

[0093] Step 3.5: When the tower crane 1 rotation angle θ and the camera 2 rotation radius R0 are both within the range, capture the image data captured by the camera 2 and automatically transmit and process it; otherwise, do not process the data captured by the camera 2.

[0094] Step 4: Safety supervision of hot work operations, including video image detection of tower crane 1;

[0095] Step 4.1: Set the categories of objects to be inspected during hot work operations, including open flames, flammable gas cylinders, flammable liquid cylinders, safety signs, and safety protective clothing;

[0096] Step 4.2: Based on the preset image deep learning model, perform rapid analysis and feature detection on the returned image data;

[0097] The preset image depth model is used to automatically identify preset detection objects. The input is an image captured and transmitted back by the camera in a specific area, and the output is an image with detection box coordinate information. The output result includes the detection object category and the pixel coordinate information of the points in the detection box (x, y, y). i ,y i ).

[0098] Taking the image data of a tower crane as an example, the output result is:

[0099] Detection object category: ignition point; pixel coordinates of the center point of the detection box: (934, 652);

[0100] Detection object category: combustible gas cylinder; pixel coordinates of the center point of the detection frame: (109, 964);

[0101] Detection object category: flammable liquid bottle; pixel coordinates of the center point of the detection frame: (659, 865).

[0102] Step 5: Determine the safety status and issue a risk warning for hot work operations.

[0103] Step 5.1: Preset the safe distance between different detection objects and the ignition point:

[0104] Within 30m of the ignition point, the system automatically detects whether flammable gas is present.

[0105] Within 15m of the ignition point, the system automatically detects whether flammable liquids are placed.

[0106] Within 10 meters of the hot work point, automatic detection is used to check whether there are obvious signs and safety markings, whether the clothing and personal protective equipment of the supervisors are in accordance with regulations, and whether there is a dedicated person monitoring the work.

[0107] Step 5.2, determine the distance between different types of detection objects and hot work points: if the distance is less than the preset safe distance, issue a hazard warning and send feedback to the backend;

[0108] Step 5.2.1, calculate the horizontal coverage length S1 of camera 2:

[0109]

[0110] Step 5.2.2, calculate the longitudinal coverage length S2 of camera 2:

[0111]

[0112] The rectangle with S1 as the length and S2 as the width is the camera's monitoring range 4;

[0113] Step 5.2.3, combining the horizontal coverage length S1, the vertical coverage length S2, and the image size (w) of camera 2. j ,h j This yields the true size of the image; where the image pixels are... (Unit: Pixel), sensor size is (w,h) = (17.3,13) (unit: mm), calculate the ratio of sensor size to image size. The actual size conversion ratio of the image

[0114] Step 5.2.4: Based on the pixel position of the center point of the ignition point detection box and the pixel position information of the center points of other detection object boxes, determine the distance D of different detection objects from the ignition point:

[0115] Calculate the distance between the combustible gas cylinder and the ignition point;

[0116] The distance between the flammable liquid bottle and the ignition point is calculated as follows:

[0117] Step 5.2.5: Determine the difference between the actual distance of the object being tested and the preset safe distance. If the distance between the combustible gas and the ignition point is 34.3m, which is greater than the preset safe distance of 30m, then the placement of the combustible gas cylinder is safe. If the distance between the combustible liquid and the ignition point is 13.5m, which is less than the preset safe distance of 15m, then the placement of the combustible liquid cylinder is unsafe.

[0118] Step 5.3: When the location of the flammable liquid bottle is detected, a hazard warning is issued and feedback is sent to the back-end management personnel for on-site intervention.

[0119] This embodiment is merely a further explanation of the present invention and is not intended to limit the present invention. Those skilled in the art can make non-inventive modifications to this embodiment as needed after reading this specification, but such modifications are protected by patent law as long as they are within the scope of the claims of the present invention.

Claims

1. A method for safety supervision of hot work operations on construction sites using tower crane video image detection, characterized in that, The method includes the following steps: Step 1: Extract hot work information from the construction site; Step 2: Establish a safety supervision area for hot work operations; Step 3: Obtaining valid video data of tower crane hooks for hot work safety supervision; Step 4: Safety supervision of hot work operations, including tower crane video image detection; Step 5: Assess the safety status and provide risk warnings for hot work operations; Step three includes: Step 3.1, calculate the camera's field of view. Horizontal field of view: ， Vertical field of view: ， In the formula, The focal length of the camera. The width of the sensor. The height is the sensor size; Step 3.2, calculate the pose parameter relationship of the tower crane's camera: ， In the formula, These are the coordinate pose parameters of the camera. These are the camera's angle and pose parameters; The center of rotation of the tower crane. Let the tower crane rotate at a given moment; define the ground height as 0. The height of the camera above the ground. The height of the working surface from the ground; The distance between the camera and the center of rotation of the tower crane is the rotation radius. The height of the camera from the work surface; Step 3.3: Calculate the angle at which the tower crane obtains valid data. The range of values ​​for: ， ， In the formula, Here are the coordinates of the hot work location. The center of rotation of the tower crane. The radius of the area affected by hot work; Step 3.4: Calculate the slewing radius of the camera used by the tower crane to acquire valid data. The range of values ， In the formula, Here are the coordinates of the hot work location. The center of rotation of the tower crane. The radius of the area affected by hot work; Step 3.5, when the tower crane turns... and camera rotation radius When all values ​​are within the specified range, the image data captured by the camera is automatically transmitted and processed; otherwise, the data captured by the camera is not processed.

2. The method for safety supervision of hot work operations on construction sites using tower crane video image detection according to claim 1, characterized in that, Step one includes collecting documents and materials related to hot work at the construction site; extracting text information from the documents and materials; and extracting key information about the hot work.

3. The method for safety supervision of hot work operations on construction sites using tower crane video image detection according to claim 2, characterized in that, Extracting key information for hot work operations includes: segmenting text data using an open-source word segmenter to obtain a word set; pre-setting keywords for hot work operations and extracting the text segments containing these keywords to obtain text data strongly related to the hot work operations; setting event trigger words and event elements; selecting several hot work operation text data sets and using a data annotation platform to perform entity and relation annotations on the key information of the hot work operations to construct a small sample training set; converting the annotated data into word vectors and inputting them into the event extraction model for training; verifying the accuracy and recall of the model after small sample training. If both accuracy and recall are greater than the set threshold, the model training is complete; otherwise, repeat the first two steps until the model meets the requirements; and using the trained model to extract key information for the hot work operations, including "hot work operation time," "hot work operation location," "hot work operation level," and "hot work operation type." 4. A method for safety supervision of hot work operations at construction sites using tower crane video image detection, as described in claim 1 or 3, characterized in that, Step two includes: Step 2.1: Based on the extracted hot work operation time information, filter out hot work operations that occurred before the current date; Step 2.2: Based on the extracted hot work location information, set the spatial coordinates of the hot work point as follows: ; Step 2.3: Set the radius of influence of the hot work operation based on the hot work operation level and type. ; Step 2.4, using the spatial coordinates of the hot work point With the center of the circle, The circular area with a radius of is the safety supervision area for hot work operations.

5. A method for safety supervision of hot work operations on construction sites using tower crane video image detection as described in claim 1, characterized in that, Step four includes setting the category of objects to be detected during hot work operations, and performing rapid analysis and feature detection on the returned image data based on a preset image deep learning model. The input is the image captured and returned by the camera in a specific area, and the output is an image with detection box coordinate information. The output result includes the category of the detected object and the pixel coordinate information of the points in the detection box. .

6. A method for safety supervision of hot work operations on construction sites using tower crane video image detection as described in claim 1, characterized in that, Step five includes: Step 5.1: Preset the safe distance between different detection objects and the ignition point: Step 5.2, determine the distance between different types of detection objects and hot work points: if the distance is less than the preset safe distance, issue a hazard warning and send feedback to the backend; Step 5.3: Issue a hazard warning for the detected dangerous conditions and provide feedback to the back-end management personnel for on-site intervention.

7. A method for safety supervision of hot work operations at construction sites using tower crane video image detection as described in claim 6, characterized in that, Step 5.2 includes: Step 5.2.1, calculate the horizontal coverage length of the camera. : ， Step 5.2.2: Calculate the longitudinal length that the camera can cover. : ， by For the length, The width of the rectangle represents the camera's monitoring range; Step 5.2.3, combined with the horizontal coverage length of the camera Length of longitudinal coverage and image size This yields the true size of the image; like The image's true size conversion ratio ;like The image's true size conversion ratio ; Step 5.2.4: Based on the pixel position of the center point of the ignition point detection frame and the pixel position information of the center points of other detection object frames, determine the distance of different detection objects from the ignition point. : ， Step 5.2.5: Determine the difference between the actual distance of the object to be tested and the preset safe distance. If the distance is greater than the preset safe distance, the placement of the object to be tested is safe and no danger warning is issued. If the distance is less than the preset safe distance, the placement of the object to be tested is unsafe and step 5.3 is executed.

Images