Building construction high-place operation protection system and method based on vision and language

Abstract

The invention relates to a building construction high-place operation protection system and method based on vision and language, and belongs to the field of image recognition. The system comprises a data acquisition and preprocessing module, a remote configuration module, a support identification module, a displacement identification module, a personnel identification module and a safety belt identification module. Through cooperation of multiple systems, real-time monitoring and safety management of the supporting device and personnel are achieved. The existence and the installation state of the supporting device are identified through the image identification technology, the displacement of the supporting device is detected through SIFT and RANSAC algorithms, whether a person is in a dangerous area or not and whether the person wears a safety belt or not is detected through the biological identification technology, and the wearing condition and correctness of the safety belt are detected through the NFC and visual analysis technology. The system has the advantages of automation, high efficiency, accuracy, multi-system cooperation, real-time monitoring, intelligent learning, datamation management and the like, the construction safety is effectively improved, and the occurrence probability of safety accidents is reduced.

Description

Technical Field

[0001] The invention belongs to the field of image recognition, and relates to a system and method for protecting construction work at heights based on vision and language. Background Art

[0002] With the rapid development of the construction industry, high-rise buildings and complex structures are constantly emerging, and working at heights has become an indispensable part of the construction process. However, working at heights is also accompanied by huge safety risks. Once an accident occurs, the consequences are often disastrous.

[0003] Traditional construction work protection at height mainly relies on manual inspection and experience judgment, which has the following problems:

[0004] Manual inspections require a lot of time and manpower, are inefficient, and it is difficult to cover all operating areas and time periods.

[0005] Manual inspections are greatly affected by factors such as personnel experience and emotions, and are prone to misjudgments and omissions, making it difficult to ensure the reliability and effectiveness of protective measures.

[0006] Manual inspections cannot monitor the situation at the construction site in real time, making it difficult to detect safety hazards and take measures in a timely manner, which can easily lead to accidents.

[0007] In recent years, with the rapid development of computer vision and deep learning technology, it has become possible to apply machine vision technology to the protection of high-altitude operations in construction. The existing machine vision protection system mainly relies on image recognition technology to identify the existence of support devices and the location of personnel, which has improved the protection efficiency to a certain extent, but it still has shortcomings in the following aspects:

[0008] Due to the complex construction site environment and large changes in lighting conditions, existing image recognition technology is difficult to accurately identify the type and installation status of the support device, which can easily lead to misjudgment and omission.

[0009] The existing system has a slow processing speed and is unable to achieve real-time monitoring and early warning, making it difficult to meet the needs of dynamically changing construction sites.

[0010] The existing system lacks intelligent learning capabilities, making it difficult to adapt to different construction environments and types of operations, and it is difficult to achieve personalized protection.

[0011] In order to solve the above problems, the present invention provides a system and method for high-altitude construction work protection based on machine vision technology and large language model, aiming to improve the protection efficiency and accuracy and reduce the probability of safety accidents. Summary of the invention

[0012] In view of this, an object of the present invention is to provide a system and method for protecting construction work at heights based on vision and language.

[0013] In order to achieve the above object, the present invention provides the following technical solutions:

[0014] A visual and language-based construction height operation protection system, including:

[0015] Data acquisition and preprocessing module, used to acquire construction site images and preprocess them;

[0016] Remote configuration module for setting safety standards and configuration parameters;

[0017] A support identification module, used to identify the presence of support devices and their installation status;

[0018] A displacement identification module is used to detect the displacement of the support device;

[0019] Personnel identification module, used to identify whether a person is in a hazardous area and whether he or she is wearing a seat belt;

[0020] The seat belt recognition module is used to identify the wearing condition and correctness of the seat belt.

[0021] Furthermore, the data acquisition and preprocessing module includes:

[0022] An image acquisition device, used for acquiring images of the construction site;

[0023] The image preprocessing device is used to perform grayscale processing, image enhancement, frame segmentation and other preprocessing operations on the collected image. Further, the remote configuration module includes:

[0024] A cloud configuration system for setting safety standards and configuration parameters, and sending the configuration to on-site monitoring equipment;

[0025] A self-learning system for training image recognition models based on labeled data and preset parameters.

[0026] Furthermore, the support identification module includes:

[0027] A support detection device, used to identify whether a support device exists in the image;

[0028] The support installation status analysis device is used to analyze whether the installation status of the support device meets the safety standards.

[0029] Furthermore, the displacement identification module includes:

[0030] A support displacement detection device, used to detect the displacement of the support device;

[0031] The support stability analysis device is used to analyze the stability of the support structure.

[0032] Furthermore, the personnel identification module includes:

[0033] A person detection device, used to detect whether there is a person in the image;

[0034] The seat belt wearing identification device is used to identify whether a person is wearing a seat belt.

[0035] Further, the seat belt recognition module includes:

[0036] Seat belt image recognition device, used to identify the wearing condition and correctness of seat belts;

[0037] Seat belt NFC identification device, used to identify the wearing condition and correctness of the seat belt.

[0038] A method for protecting construction work at height based on vision and language, comprising the following steps:

[0039] Collect construction site images and preprocess them;

[0040] Set the training parameters of the image recognition model according to the preset security standards and configuration parameters;

[0041] Using the trained image recognition model, identify whether there is a support device in the image and its installation status;

[0042] Detect the displacement of the support device and analyze the stability of the support structure;

[0043] Detect whether there is a person in the image and identify whether the person is wearing a seat belt;

[0044] Identify the wearing and correctness of seat belts.

[0045] Furthermore, the step of collecting and preprocessing the construction site image includes:

[0046] Using an image acquisition device to collect images of the construction site;

[0047] Perform preprocessing operations such as grayscale processing, image enhancement, and frame segmentation on the collected images.

[0048] Further, the step of identifying whether there is a support device in the image and its installation status includes:

[0049] Using a support detection device to identify whether a support device exists in the image;

[0050] The support installation status analysis device is used to analyze whether the installation status of the support device meets the safety standards.

[0051] The beneficial effects of the present invention are:

[0052] (1) Through the combination of machine vision technology and large language models, real-time monitoring and safety management of support devices and personnel are achieved, which improves work efficiency and accuracy and reduces the cost of manual inspections and the impact of subjective factors.

[0053] (2) The present invention integrates multiple modules such as data acquisition, remote configuration, support identification, displacement identification, personnel identification and safety belt identification, realizing all-round protection for high-altitude operations and effectively reducing the probability of safety accidents.

[0054] (3) The system can monitor the situation at the construction site in real time, detect potential safety hazards in a timely manner, and issue early warning information so that timely measures can be taken to avoid accidents.

[0055] (4) Through the self-learning system and big data analysis, the system can continuously learn and optimize, improve recognition accuracy and protection capabilities, and adapt to different construction environments and types of operations.

[0056] (5) The system can record and store relevant data, facilitate data analysis and accident tracing, and provide data support for construction safety management.

[0057] Other advantages, objectives and features of the present invention will be described in the following description to some extent, and to some extent, will be obvious to those skilled in the art based on the following examination and study, or can be taught from the practice of the present invention. The objectives and other advantages of the present invention can be realized and obtained through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be described in detail below in conjunction with the accompanying drawings, wherein:

[0059] Figure 1 is a system diagram of the present invention;

[0060] Figure 2 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0061] The following describes the embodiments of the present invention by specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments only illustrate the basic concept of the present invention in a schematic manner, and the following embodiments and features in the embodiments can be combined with each other without conflict.

[0062] Among them, the drawings are only used for illustrative explanations, and they only represent schematic diagrams rather than actual pictures, and should not be understood as limitations on the present invention. In order to better illustrate the embodiments of the present invention, some parts of the drawings may be omitted, enlarged or reduced, and do not represent the size of actual products. For those skilled in the art, it is understandable that some well-known structures and their descriptions in the drawings may be omitted.

[0063] The same or similar numbers in the drawings of the embodiments of the present invention correspond to the same or similar parts; in the description of the present invention, it should be understood that if the terms "upper", "lower", "left", "right", "front", "rear", etc. indicate the orientation or position relationship, they are based on the orientation or position relationship shown in the drawings, which is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operate in a specific orientation. Therefore, the terms describing the position relationship in the drawings are only used for illustrative purposes and cannot be understood as limiting the present invention. For ordinary technicians in this field, the specific meanings of the above terms can be understood according to specific circumstances.

[0064] 1. Data acquisition and preprocessing module

[0065] (1) Image acquisition: Deploy a dome camera with a resolution of at least 720P (1280x720 pixels), and 1080P (1920x1080 pixels) is recommended to clearly capture the details of the support device. The camera collects video streams in real time to ensure that every frame of the image can be analyzed.

[0066] (2) Image processing: Segment the collected video data to obtain continuous image frames. Use the OpenCV library to enhance the image and improve the image quality, especially the edges and details of the support device: convert the image into a grayscale image through the grayscale algorithm, use Gaussian blur or median filtering to reduce image noise, and use histogram equalization to improve image contrast. Finally, use the Canny edge detection algorithm to extract the edge information of the support device.

[0067] 2. Remote configuration module:

[0068] (1) Cloud operation and maintenance configuration system

[0069] ① Set the marking content: including the type of support device and the position and status of key parts (such as guardrails, safety nets, embedded anchor rings, etc.).

[0070] 1) Guardrails: including the position and status of key parts of the guardrails, such as the upper rod, lower rod, pillars, and footboards.

[0071] a. Horizontal bar position: mark the distance between the upper bar and the lower bar and the ground, ensuring that the upper bar is 1.2 meters from the ground and the lower bar is 0.5 meters from the ground.

[0072] b. Toeboard position: Mark the height of the toeboard and ensure it is not less than 180 mm.

[0073] c. Spacing between pillars: Mark the spacing between pillars to ensure it does not exceed 2 meters.

[0074] d. Connection point: Mark the connection point between the railing and the support to ensure that the connection is firm and reliable.

[0075] 2) Safety net: Mark the installation location of the safety net and ensure it is set below the high-risk area.

[0076] a. Mesh size: Mark the mesh size and ensure it is no larger than 10cm×10cm.

[0077] b. Fixing points: Mark the fixing points of the safety net to ensure it is firm and reliable.

[0078] 3) Safety rope: including the position and status of each key part of the safety belt (shoulder strap and waist belt), and marking multiple key points for each image (such as the upper point of the shoulder strap, the lower point of the shoulder strap, the left point of the waist belt, and the right point of the waist belt).

[0079] a. Shoulder Strap Position: Mark where the shoulder straps go across your shoulders, ensuring they are positioned correctly under your chest.

[0080] b. Belt Position: Mark where the belt fits just above the hips, making sure it fits snugly against the body.

[0081] c. Connection points: mark the four points of the seat belt

[0082] 4) Opening cover: including the position and status of key parts such as the position, fixing points, and materials of the opening cover.

[0083] a. Opening cover position: Mark the position of the cover to ensure it fits tightly against the edge of the opening.

[0084] b. Fixing point: Mark the fixing point and fixing method of the cover to ensure it is firm and reliable.

[0085] c. Material status: Mark the appearance of the cover material to ensure there is no deformation, cracks or corrosion.

[0086] ②Set the labeling method: After loading the image, use the mouse to draw a rectangular box on the image to surround the object to be labeled (such as support device). In the pop-up dialog box, enter the label name, such as "support device", "safety rope", etc.

[0087] ③Background parameter configuration

[0088] 1) Job type setting (users can select different job types in the system, and the system applies corresponding safety standards according to the job type:

[0089] a. Working on the edge: protective measures such as guardrails and safety nets need to be set up.

[0090] b. Opening operations: It is necessary to set up opening covers, guardrails and other protective measures.

[0091] 2) Threshold setting

[0092] a. Falling height reference plane: Set the falling height threshold for edge work. By default, a dense mesh needs to be set if the falling height is greater than 2 meters or above.

[0093] b. Guardrail height: The default upper rail height is 1.2 meters and the lower rail height is 0.5 meters. Users can adjust according to needs.

[0094] c. Pillar spacing: The default pillar spacing does not exceed 2 meters, and users can adjust it according to actual conditions.

[0095] d. Height of footboard: The default height is not less than 180 mm.

[0096] e. Safety net mesh size: The default mesh size is no larger than 10cm×10cm.

[0097] f. Opening size: The user sets the thresholds for the length of the short side of the vertical opening, the length of the short side of the vertical opening, and the length of the short side of the non-vertical opening.

[0098] 3) Rule settings

[0099] Users can set specific rules based on standards such as "Construction Safety Inspection Standards (JGJ59-2011)", "Safety Net Standards (GB5725-1997)", "Fall Protection Safety Belts (GB6095-2021)", and "Technical Specifications for Safety of Height Operations in Construction (JGJ80)".

[0100] 4) Other setting requirements: (1) Ensure that the rectangular box accurately surrounds the target object and does not include too much background; (2) The rectangular box should completely cover all important parts of the target object (such as the connection points and fixing points of the support device); (3) Multi-angle annotation: Images under different angles and lighting conditions need to be annotated to ensure that the model can generalize.

[0101] 5) Judgment criteria: When the marked rectangular frame accurately surrounds the support device and meets the preset label, the support device is judged to be installed. If the rectangular frame fails to cover the support device or the label is wrong, the support device is judged not to be installed or installed incorrectly.

[0102] 6) System integration: Call the model in the video surveillance software for real-time image analysis. Send the surveillance video stream to the server through the API interface, and the server performs image processing and support recognition and returns the recognition results.

[0103] (2) Self-learning system: Use convolutional neural network (CNN) for image recognition and choose verified models such as ResNetet-50 or VGG.

[0104] ① Set annotation parameters: Annotate the specific parameters of the support device, including position (bounding box coordinates) and type (support device category).

[0105] ②Image annotation:

[0106] 1) After loading the image, use the mouse to draw a rectangular box on the image to surround the object to be marked (such as the support device).

[0107] 2) Set the label name: Enter the label name in the pop-up dialog box, such as "support device", "safety rope", etc.

[0108] 3) Set the annotation scene

[0109] ③Training model:

[0110] 1) Use the PyTorch framework for model training.

[0111] 2) Load the labeled data into the dataset and perform data augmentation (such as random cropping, rotation, and flipping).

[0112] 3) Define the loss function (cross entropy loss) and optimizer (Adam optimizer).

[0113] 4) Use the back-propagation algorithm to optimize the model parameters and update the model weights by minimizing the loss function.

[0114] ④Large language model assisted self-learning:

[0115] The large language model processes multimodal data, translates irregular instructions into task language and extracts effective features, helping agents filter invalid information and accelerating downstream neural network learning. It then analyzes data to identify patterns and generate optimization strategies, designs implicit and explicit reward functions, and generates reward signals. The model self-corrects based on new data and continuously optimizes.

[0116] 3.Support identification module

[0117] (1) Support identification (detecting whether there is a support device in the image)

[0118] ① Implementation method:

[0119] 1) Use a pre-trained convolutional neural network (CNN) model such as ResNet-50 or VGG.

[0120] 2) Process the input image to identify and mark the presence of the support device.

[0121] 3) Identification standard: If there is a support device in the rectangular frame, it is determined that the support device exists; otherwise, it is determined that it does not exist.

[0122] ②Steps:

[0123] 1) Image input: Get real-time images from the video surveillance system.

[0124] 2) Image processing: Use pre-trained CNN models to analyze images.

[0125] 3) Support detection: Locate the support device in the image and draw a rectangular box.

[0126] 4) Result output: Output the support detection results and mark the position of the support device.

[0127] (2) Identify whether the support is installed correctly (check whether the support device is installed correctly according to the set safety standards)

[0128] ① Implementation method:

[0129] 1) Use a pre-trained convolutional neural network (CNN) model, such as ResNet-50 or VGG. If a support device exists, extract features through the convolution layer, and the classification layer outputs the type of support device.

[0130] 2) Call the numerical thresholds set in the background, such as the height of the support, the height of the footboard, etc.

[0131] 3) Conduct a detailed analysis of the detected support devices to determine whether their installation meets the preset standards.

[0132] 4) Identification standard: compare according to the set threshold value. If the installation parameters of the support device are within the threshold range, it is determined that the installation is correct; otherwise, it is determined that the installation is incorrect.

[0133] ②Steps:

[0134] 1) Image input: Get real-time images from the video surveillance system.

[0135] 2) Support detection: Use the pre-trained CNN model to identify support devices and mark their locations.

[0136] 3) Parameter comparison:

[0137] a. Support installation position identification:

[0138] a) Pixel point recognition: Extract the position coordinates of the support device through image processing technology.

[0139] b) Define coordinate points: Define the range of coordinate points in the standard installation position. For example, the installation point range of the safety rope can be defined as a rectangular area from (x1, y1) to (x2, y2).

[0140] c) Actual scene comparison: Compare the standard installation position and angle requirements with the actual image. For example, if the support device should be 1.2 to 1.5 meters away from the wall, the actual distance in the image can be calculated by the distance between the pixels and compared.

[0141] d) Position judgment: If the identified support device coordinate point is within the defined threshold, the installation position is judged to be accurate; if it exceeds the threshold, the installation position is judged to be inaccurate.

[0142] b. Support installation angle and direction identification:

[0143] a) Hough transform detection of direction line: Use the Hough transform algorithm to detect the direction line of the support device and calculate its angle with the standard direction.

[0144] b) Benchmarking angle: Use the standard direction template to set the angle range of the standard direction (e.g. the safety rope should be arranged horizontally, and the allowable deviation angle is ±5°).

[0145] c) Comparison process: The identified direction line is compared with the standard direction. If it is within the allowable range, it is judged to meet the standard, otherwise it is judged to be non-compliant.

[0146] c. Identification of support installation stability:

[0147] a) Image recognition: Use image processing algorithms to determine the edges and connection points of the support device and detect whether there are signs of looseness.

[0148] b) Edge detection: The edge of the support device is extracted using the Canny edge detection algorithm, and the clarity and integrity of the edge are analyzed.

[0149] c) Connection point analysis: Check whether the connection point has displacement or deformation, and determine whether the installation is stable.

[0150] d) Algorithm implementation: Use OpenCV library for edge detection and connection point analysis.

[0151] e) Result output: Output the support installation inspection results, mark the parts that do not meet the standards and issue an alarm.

[0152] 4. Displacement recognition module

[0153] This patent provides a support displacement monitoring method, which combines the camera with the equipment hardware to determine in real time whether the support structure has been displaced. The camera configured in the system scans the site regularly. Each scan stays for 5-10 seconds according to the size of the facade opening, and records an optimal picture. It is expanded according to the specific range and records the support information of the location. The camera records around the edge of the entire facade opening based on the preset point setting method to ensure that objects at the edge are always above 50 pixels.

[0154] In the overall image processing process, the SIFT (Scale-Invariant Feature Transform) algorithm is used to detect feature points of the identified images, and the RANSAC (Random Sample Consensus) algorithm is used for matching, thereby ensuring high accuracy during the image recording process. Subsequently, the Homography-based stitching algorithm is used to integrate multiple images recorded by the camera into a panoramic image showing the condition of a certain facade according to the feature values ​​of the images, and the image is uploaded to the cloud platform, where the support position is calculated:

[0155] (1) Case 1: The YOLO algorithm model is used to detect the crossbar and the left and right walls. When it is found that the crossbar of the support surface does not overlap with the left and right walls, it is judged that the support has shifted left and right.

[0156] (2) Case 2: Through target recognition, the topmost horizontal bar in the camera is identified. Through the near and far focus adjustment of the camera, the horizontal bar and the bottom position are framed in the same picture at the same time, and the pixel position height of the horizontal bar from the bottom is calculated in the picture. The camera tilt angle and the position of the camera from the target point are compositely calculated to obtain the distance between the top horizontal bar and the support surface. If the top of the horizontal bar support is lower than 1.2-1.5m from the ground, it is judged that the support has been displaced in height;

[0157] (3) Case 3: Use the Faster R-CNN (Region-based Convolutional Neural Network) model to accurately locate and identify the vertical bars at the facade opening. If only one vertical bar is found on the support surface, it is determined that the support has vertical bar displacement;

[0158] When the above three situations occur, it is judged that the support is displaced, and subsequent judgments are made based on whether there is someone there.

[0159] 5. Personnel identification module

[0160] (1) Obtain real-time images through the camera.

[0161] (2) Use pre-trained biometric models (such as YOLO or SSD) to identify humans in real time.

[0162] ① Real-time discrimination: The model will detect whether there is a human body in each frame of the image.

[0163] ② Avoid misjudgment: Set the confidence threshold to 0.9 (i.e. 90%). Only detection results with a confidence level higher than 90% will be considered as people outside the fence. Furthermore, the range is set to 3 consecutive frames. If a human body is detected in 3 consecutive frames of images, it is confirmed that there is a person.

[0164] (3) If no one is detected, the countdown is triggered.

[0165] ①The countdown is set in the system (or software) and is implemented through the timer function of the software.

[0166] ② Use the system’s timer tool or the timer module in the programming language (such as the time module in Python).

[0167] ③After the countdown ends: the system automatically performs the preset return inspection operation at the end of the countdown (using image processing technology to determine whether the support device has returned to its position by the difference method of the previous and next frame images):

[0168] 1) Previous and next frame comparison: Compare the current frame image with the previously stored reference frame image.

[0169] 2) Image difference: Calculate the absolute difference between two frames of images to obtain a differential image.

[0170] 3) Threshold processing: Binarize the difference image to obtain a binary image of the changed area.

[0171] 4) Change detection: By detecting the changed area in the binary image, determine whether there is a significant change. If the area of ​​the changed area is smaller than the set threshold, the support device is considered to be in place, otherwise it is considered not to be in place.

[0172] When there is no one working at the facade opening, the system will start a countdown on site according to the specified time configured in the cloud. At the same time, the camera will be zoomed in to a position that just covers the facade opening for real-time monitoring until someone is found in the area. If no one appears on the screen after the specified time and the support has not been restored normally, an early warning will be recorded and sent to the cloud server together with the video, and a rectification list will be generated.

[0173] (4) If there are people present, the seat belt recognition system will be called to perform seat belt wearing recognition

[0174] 6. Seat belt recognition module

[0175] (1) Seat belt shoulder belt recognition (image recognition)

[0176] ① Model selection: The ResNet model is used for seat belt wearing recognition. This model is different from the ResNet model used for support recognition mentioned above in terms of tasks. This model is used to detect seat belt wearing conditions, while the former is used to detect support devices.

[0177] ② Image annotation: Use tools such as LabelImg to annotate the collected images. The annotation content includes the position and status of each key part of the seat belt (shoulder belt and waist belt), and annotates multiple key points for each image (such as the upper point of the shoulder belt, the lower point of the shoulder belt, the left point of the waist belt, and the right point of the waist belt). The annotation method is as follows:

[0178] 1) Shoulder Strap Position: Mark where the shoulder straps go across your shoulders, making sure they are positioned correctly under your chest.

[0179] 2) Belt position: Mark where the belt fits just above the hips, making sure it fits snugly against the body.

[0180] 3) Connection points: Mark the four points of the seat belt.

[0181] The marking requirements are as follows:

[0182] 1) Accurate positioning: Set 200px to ensure that the rectangular box accurately surrounds the target object and does not include too much background.

[0183] 2) Covering the target: The rectangular frame should completely cover all important parts of the target object (such as the connection points and fixing points of the support device).

[0184] 3) Multi-angle annotation: Images at different angles and lighting conditions need to be annotated to ensure that the model can generalize.

[0185] Use a deep learning model for training, with the input being the labeled image and the output being the coordinates and status of each key point. Input the preprocessed image into the trained model to obtain the coordinates and status of each key point. The judgment criteria are as follows (refer to JGJ59-2011):

[0186] 1) Correct shoulder strap position: The shoulder strap should cross the shoulders and be fixed under the chest without sliding or shifting.

[0187] 2) Correct belt position: The belt should be fixed above the hips and adjusted to the appropriate tightness.

[0188] 3) Correct wearing status: The safety belt is not twisted or knotted during use, and all connection points are firm and reliable.

[0189] When the marked rectangle accurately surrounds the support device and meets the preset label, it is considered to be correctly identified. If the rectangle does not cover the support device or the label is wrong, it is considered to be unrecognizable.

[0190] (2) Abnormal identification of embedded anchor ring (NFC identification)

[0191] An NFC card recognition device is configured on the seat belt buckle and can be linked to the Internet through a 4G module. At the same time, an NFC tag card is posted on the hook.

[0192] When the on-site communication recognizes that a person is wearing a safety belt and working at height, the safety belt buckle warning mechanism will be activated. If the NFC card recognition device cannot recognize the corresponding NFC tag card during the operation, the safety belt will issue an alarm to remind the on-site personnel that there is an abnormality in wearing the safety belt. At the same time, the camera will be pulled to monitor the personnel. The real-time recognition system will record it, and at the same time, the on-site camera will be linked to track the person's work status and notify the person in time to make corrections.

[0193] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solution, which should be included in the scope of the claims of the present invention.

Claims

1. A visual and language-based construction height operation protection system, characterized by: include: Data acquisition and preprocessing module, used to acquire construction site images and preprocess them; Remote configuration module for setting safety standards and configuration parameters; A support identification module, used to identify the presence of support devices and their installation status; A displacement identification module is used to detect the displacement of the support device; Personnel identification module, used to identify whether a person is in a hazardous area and whether he or she is wearing a seat belt; The seat belt recognition module is used to identify the wearing condition and correctness of the seat belt.

2. The visual and language-based construction height operation protection system according to claim 1 is characterized by: The data acquisition and preprocessing module includes: An image acquisition device, used for acquiring images of the construction site; The image preprocessing device is used to perform preprocessing operations such as grayscale processing, image enhancement, frame segmentation, etc. on the collected images.

3. The visual and language-based construction height operation protection system according to claim 1 is characterized by: The remote configuration module includes: A cloud configuration system for setting safety standards and configuration parameters, and sending the configuration to on-site monitoring equipment; A self-learning system for training image recognition models based on labeled data and preset parameters.

4. The visual and language-based construction height operation protection system according to claim 1 is characterized by: The support identification module includes: A support detection device, used to identify whether a support device exists in the image; The support installation status analysis device is used to analyze whether the installation status of the support device meets the safety standards.

5. The visual and language-based construction height operation protection system according to claim 1 is characterized by: The displacement recognition module comprises: A support displacement detection device, used to detect the displacement of the support device; The support stability analysis device is used to analyze the stability of the support structure.

6. The visual and language-based construction height operation protection system according to claim 1 is characterized by: The personnel identification module comprises: A person detection device, used to detect whether there is a person in the image; The seat belt wearing identification device is used to identify whether a person is wearing a seat belt.

7. The visual and language-based construction height operation protection system according to claim 1 is characterized by: The seat belt recognition module comprises: Seat belt image recognition device, used to identify the wearing condition and correctness of seat belts; Seat belt NFC identification device, used to identify the wearing condition and correctness of the seat belt.

8. A method for protecting construction work at height based on vision and language, characterized by: The following steps are involved: Collect construction site images and preprocess them; Set the training parameters of the image recognition model according to the preset security standards and configuration parameters; Using the trained image recognition model, identify whether there is a support device in the image and its installation status; Detect the displacement of the support device and analyze the stability of the support structure; Detect whether there is a person in the image and identify whether the person is wearing a seat belt; Identify the wearing and correctness of seat belts.

9. The method for protecting construction work at heights based on vision and language according to claim 8 is characterized by: The steps of collecting construction site images and preprocessing them include: Using an image acquisition device to collect images of the construction site; Perform preprocessing operations such as grayscale processing, image enhancement, and frame segmentation on the collected images.

10. The method for protecting construction work at heights based on vision and language according to claim 8 is characterized by: The step of identifying whether there is a support device in the image and its installation status comprises: Using a support detection device to identify whether a support device exists in the image; The support installation status analysis device is used to analyze whether the installation status of the support device meets the safety standards.

Images