Hoisting safety monitoring method and system based on large model

Abstract

The invention provides a hoisting safety monitoring method and system based on a large model, and the method comprises the steps: a monitoring end obtains a hoisting construction image in a hoisting process in real time, inputs the hoisting construction image into a front-end model deployed at the monitoring end, obtains a hoisting state recognition result, and transmits the hoisting state recognition result to the monitoring end; judging whether the hoisting state recognition result is hoisting operation or not, and if yes, sending a first hoisting state image in hoisting operation in the hoisting construction image to a cloud; and the cloud end receives the first hoisting state image, inputs the first hoisting state image into a multi-mode large model to obtain a first hoisting safety identification result, judges whether the first hoisting safety identification result shows that hoisting hidden dangers exist or not, and pushes the hoisting hidden dangers to a monitoring end if the first hoisting safety identification result shows that the hoisting hidden dangers exist. According to the invention, the computing power resource requirement of the front-end equipment can be reduced, and the multi-feature hidden danger identification accuracy can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of engineering construction, and in particular to a large-model-based hoisting safety monitoring method and system. Background Art

[0002] There are many hoisting operations on construction sites, which run almost throughout the entire construction process. The main problems with hoisting operations are as follows: the continuous operation time of hoisting is long, and the on-site safety management personnel cannot always keep an eye on the hoisting operations to see if there are any hidden dangers, which can easily lead to omissions. Common safety hazards in hoisting operations include: (1) There is no square wood or steel plate under the legs of the truck crane; (2) Long and short materials are hoisted together; (3) Crooked and slanted hoisting; (4) Scattered materials are not hoisted in a bucket; (5) Single-point hoisting; (6) The hoisted objects overflow; (7) People stand on the hoisted objects; (8) There are people under the hoisted objects without safety helmets.

[0003] The existing safety monitoring method for hoisting operations usually places a monitoring device at the construction site and uniformly deploys a hoisting hidden danger identification algorithm model on the monitoring device, identifies and outputs hidden dangers in real time at the hoisting site, and handles them in a timely manner when safety hazards occur. However, the above safety monitoring method has the problem of high equipment computing resource requirements and low accuracy in multi-feature hidden danger identification. Summary of the invention

[0004] In order to solve the above-mentioned problems in the prior art, the present invention provides a large-model-based lifting and hoisting safety monitoring method and system, which reduces the computing power resource requirements of the front-end equipment and can improve the accuracy of multi-feature hidden danger identification.

[0005] In order to achieve the above object, the technical solution adopted by the present invention is: In a first aspect, the present invention provides a method for monitoring the safety of hoisting and lifting based on a large model, comprising the steps of: S1. The monitoring end obtains the hoisting construction image in the lifting process in real time, inputs the hoisting construction image into the front-end model deployed on the monitoring end, obtains the hoisting state recognition result, determines whether the hoisting state recognition result is a hoisting operation, and if so, sends the first hoisting state image in the hoisting operation in the hoisting construction image to the cloud; S2. The cloud receives the first hoisting status image, and inputs the first hoisting status image into the multimodal large model to obtain a first hoisting safety identification result, and determines whether the first hoisting safety identification result indicates that there is a hoisting hidden danger. If so, the hoisting hidden danger is pushed to the monitoring end.

[0006] The beneficial effects of the present invention are: the feature identification of hoisting hazards is dispersed to the front-end model deployed at the construction site for real-time collection and analysis of the single feature of the hoisting state, and the multi-modal large model deployed in the cloud only performs multi-feature hazard analysis on images in the hoisting state without real-time full analysis of hoisting hazards, which not only reduces the computing power resource requirements of the front-end equipment, but also reduces the overall computing power resource pressure. Moreover, after the front-end model first identifies and analyzes the hoisting state, the cloud-based multi-modal large model is used to identify the hazards existing in the hoisting state, thereby improving the accuracy of multi-feature hazard identification.

[0007] Optionally, the method further comprises the steps of: Collect pictures of hidden dangers in lifting, and label the pictures of hidden dangers in lifting, wherein the labels include description of hidden danger characteristics, correct operation requirements and hidden danger types; The marked lifting hazard pictures are input as prompt words into the multimodal big model for training to obtain the trained multimodal big model, and the trained multimodal big model is deployed in the cloud. The multimodal big model includes a general big model for lifting hazard and a hazard classification neural network. The general big model for lifting hazard identifies lifting safety hazards on the lifting hazard pictures. If lifting hazards are identified, the input content is vectorized and output to the hazard classification neural network, and the hazard category is output through the hazard classification neural network.

[0008] The step S2 comprises the steps of: The cloud receives the first hoisting status image, and inputs the first hoisting status image into the multimodal large model to obtain a first hoisting safety identification result, and determines whether the first hoisting safety identification result indicates that there is a hoisting hidden danger. If so, the hidden danger category of the hoisting hidden danger is pushed to the monitoring end.

[0009] According to the above description, positive and negative example comparisons are formed through the description of hidden danger features and correct operation requirements, and the lifting hidden dangers are classified through the hidden danger classification neural network, thereby increasing the robustness of the multimodal large model and avoiding the problem of divergence in the feedback of the multimodal large model recognition results.

[0010] Optionally, in step S2, pushing the hidden danger category of the lifting hidden danger to the monitoring end includes the steps of: When a hoisting hazard is identified and the hazard category is obtained, the first hoisting status image is segmented into a second hoisting status image which only excludes the area where the hazard category is located, and the second hoisting status image is re-input into the multimodal large model to obtain a second hoisting safety identification result, and the first hoisting safety identification result and the second hoisting safety identification result are summarized, and the summarized hazard category is pushed to the monitoring end.

[0011] Optionally, if the second hoisting safety identification result still contains hidden danger categories, the second hoisting state image is segmented and identified again until the hoisting safety identification result of the segmented image contains no hidden dangers.

[0012] Optionally, step S2 further includes the steps of: When the first hoisting safety identification result and the second hoisting safety identification result are inconsistent, the hidden danger categories obtained by the two identifications are pushed to the monitoring end together.

[0013] Optionally, the method further comprises the steps of: When the second hoisting safety identification result still has safety hazards, the first hoisting state image and the summarized hazard categories are sent to the multimodal large model for model updating to obtain an updated multimodal large model.

[0014] According to the above description, it can be seen that the accuracy of hoisting safety identification is improved through secondary identification, and is used to update the multimodal large model to further improve the recognition accuracy of the multimodal large model.

[0015] Optionally, the step S1 further includes the steps of: If the lifting status identification result is a lifting operation, it is analyzed whether the pre-lifting work is complete. If it is not complete, a reminder message is issued. The pre-lifting work includes the acceptance of the lifting equipment required for the lifting operation, the certification of the lifting operators and the lifting warning.

[0016] Optionally, the hoisting construction image is stored in real time in a monitoring terminal.

[0017] Optionally, the method further comprises the steps of: S3. The monitoring end receives the lifting hidden danger and links the on-site broadcasting device to perform voice broadcast reminder of the lifting hidden danger.

[0018] Optionally, the hoisting hidden dangers may also be sent to a terminal of an on-site safety manager.

[0019] In a second aspect, the present invention provides a large-scale model-based hoisting safety monitoring system, comprising: The monitoring end is used to obtain the hoisting construction image in the lifting process in real time, input the hoisting construction image into the front-end model deployed on the monitoring end, obtain the hoisting state recognition result, judge whether the hoisting state recognition result is a hoisting operation, and if so, send the first hoisting state image in the hoisting operation in the hoisting construction image to the cloud; The cloud end is used to receive the first hoisting status image, and input the first hoisting status image into the multimodal large model to obtain a first hoisting safety identification result, and determine whether the first hoisting safety identification result indicates that there is a hoisting hidden danger. If so, the hoisting hidden danger is pushed to the monitoring end.

[0020] Among them, the technical effect corresponding to the large-model-based lifting and hoisting safety monitoring system provided by the second aspect refers to the relevant description of the large-model-based lifting and hoisting safety monitoring method provided by the first aspect. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 A schematic diagram of the main flow of a large-model-based hoisting safety monitoring method according to an embodiment of the present invention; Figure 2 A schematic diagram of a framework of a large-model-based hoisting safety monitoring system according to an embodiment of the present invention; Figure 3 It is a schematic diagram of the framework of the monitoring terminal involved in the embodiment of the present invention.

[0022] Description of reference numerals: 1. A hoisting safety monitoring system based on a large model; 2. Monitoring terminal; 3. Cloud. DETAILED DESCRIPTION

[0023] In order to better understand the above technical solution, exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided to enable a clearer and more thorough understanding of the present invention and to fully convey the scope of the present invention to those skilled in the art. Embodiment 1

[0024] This embodiment is applicable to application scenarios that require lifting and hoisting, including but not limited to on-site management of lifting and hoisting operations in various types of projects such as building construction, road and bridge engineering, rail engineering, airport engineering, municipal engineering, and water transport engineering. The existing lifting hidden danger identification algorithm model has the problem of high equipment computing power resource requirements and low multi-feature hidden danger identification accuracy. This embodiment disassembles the lifting hidden danger features, and the front-end model and the multi-modal large model on the cloud perform division of labor and collaborative identification, which not only reduces the computing power resource requirements of the front-end equipment, but also improves the accuracy of multi-feature hidden danger identification. See the following description for details.

[0025] Please refer to Figure 1 A method for monitoring the safety of hoisting and lifting based on a large model comprises the following steps: S01. Collect pictures of hidden dangers in lifting, and label the pictures of hidden dangers in lifting, wherein the labels include descriptions of hidden danger characteristics, correct operation requirements, and hidden danger types.

[0026] Among them, pictures containing safety hazards in lifting and hoisting are collected in advance, and the hazard characteristics, correct operation requirements and hazard types are described for these pictures, so as to complete the labeling of lifting hazard pictures.

[0027] S02. Input the marked lifting hidden danger pictures as prompt words into the multimodal big model for training, and obtain the trained multimodal big model. The trained multimodal big model is deployed in the cloud. The multimodal big model includes a general big model for lifting hidden dangers and a hidden danger classification neural network. The general big model for lifting hidden dangers identifies lifting safety hazards in lifting hidden danger pictures. If lifting hidden dangers are identified, the input content is vectorized and output to the hidden danger classification neural network, and the hidden danger category is output through the hidden danger classification neural network.

[0028] Among them, when the existing general large model of lifting hidden dangers performs intelligent identification and analysis of hidden danger problems, the large model recognition results have the problem of divergent feedback and poor robustness. This embodiment 1 adds the positive and negative example comparison of hidden danger feature description and correct operation requirements as prompt words to the multimodal large model, which helps to concentrate the output feedback of the multimodal large model, thereby increasing the robustness of the large model in the vertical field of intelligent identification of lifting hidden dangers, avoiding the problem of divergent feedback of the large model recognition results, and enabling the large model to accurately analyze the hidden danger type when performing image feature recognition.

[0029] Among them, the hidden danger classification neural network is composed of a convolutional layer, a fully connected layer, and a softmax weight function. After training with lifting hidden danger pictures and corresponding safety specification pictures, a classification model is obtained that can identify the hidden danger categories of lifting and hoisting.

[0030] Among them, the hidden danger categories include but are not limited to: mixed lifting of different objects, single-point lifting, irregular crane legs, asymmetrical or tilted lifting objects, loose binding of lifting objects, not using a lifting bucket for scattered materials, and mixed lifting of long and short materials, etc.

[0031] S1. The monitoring end obtains the hoisting construction image in the lifting process in real time, inputs the hoisting construction image into the front-end model deployed on the monitoring end, obtains the hoisting state recognition result, and determines whether the hoisting state recognition result is a hoisting operation. If so, the first hoisting state image in the hoisting operation in the hoisting construction image is sent to the cloud.

[0032] During on-site hoisting, the side robot starts working as the monitoring end to monitor the hoisting process in real time. By capturing the hoisting construction images in real time, it is possible to determine whether the hook is off the ground or suspended in the air. If the hook is off the ground or suspended in the air, it is judged as a hoisting operation, and the hoisting status image is pushed to the multimodal large model in the cloud to identify whether there are any hidden dangers in the hoisting process. In this way, the front-end model deployed on the monitoring end does not require high computing resources, because it only needs to identify whether it is hoisting.

[0033] Among them, the hoisting construction images are saved in real time in the monitoring end. At this time, when the cloud-based multimodal large model service is down, the front-end hoisting status recognition algorithm model can still operate normally to record the hoisting operation process. When the cloud-based multimodal large model service is restored, it can resume the breakpoint to analyze and determine the hidden dangers of the hoisting operation.

[0034] In this embodiment, step S1 further includes the steps of: If the lifting status identification result is lifting operation, it is analyzed whether the preparatory work for lifting is complete. If not, a reminder message is issued. The preparatory work includes the acceptance of lifting equipment required for the lifting operation, the certification of lifting operators and lifting warning.

[0035] Among them, the reminder information can be linked to the on-site broadcasting device for voice broadcast to remind on-site workers to stop it in time. At the same time, the reminder information will be pushed to the terminal of the relevant management personnel for disposal to achieve a closed loop.

[0036] S2. The cloud receives the first hoisting status image, and inputs the first hoisting status image into the multimodal large model to obtain a first hoisting safety identification result, and determines whether the first hoisting safety identification result indicates that there is a hoisting hidden danger. If so, the hoisting hidden danger is pushed to the monitoring end.

[0037] Among them, the general large model of hoisting hidden dangers in the multimodal large model first identifies whether there are safety hazards. When there are hoisting hidden dangers, the hidden danger classification neural network identifies and classifies the hidden danger categories, and finally outputs the first hoisting safety identification result. Since the result is limited to the standard description used during training, its feedback result will be more concentrated. Moreover, the multimodal large model in the cloud only needs to perform multi-feature hidden danger analysis on images in the hoisting state instead of real-time full analysis of hoisting hidden dangers, thereby reducing the pressure on computing resources.

[0038] At the same time, the unified deployment of the lifting hidden danger identification algorithm model has a high recognition accuracy rate for single-feature hidden dangers, but a low recognition accuracy rate for hidden dangers that are integrated with multiple features. Most of the hidden dangers in the lifting and hoisting construction process are caused by the integration of multiple features, rather than single-feature hidden dangers. Among them, single features refer to: lifting status, personnel position, not wearing a safety helmet, lifting conditions, bundling conditions, etc. Multi-feature fusion refers to: lifting status + people standing under the lifting object, lifting status + workers not wearing safety helmets, lifting status + people standing on the lifting object, lifting status + irregular bundling of lifting objects, etc. Since the lifting status is disassembled into the front-end model for identification, the back-end model reduces the types of features that need to be identified, thereby improving the pertinence and accuracy of hidden danger identification and analysis.

[0039] In this embodiment, step S2 includes the steps of: S21. The cloud receives a first hoisting state image, and inputs the first hoisting state image into a multimodal large model to obtain a first hoisting safety identification result, and determines whether the first hoisting safety identification result indicates that there is a hoisting hazard.

[0040] S22. If yes, segment the first hoisting status image into a second hoisting status image which excludes only the area where the hidden danger category is located, re-input the second hoisting status image into the multimodal large model to obtain a second hoisting safety identification result, summarize the first hoisting safety identification result and the second hoisting safety identification result, and push the summarized hidden danger category to the monitoring end.

[0041] Among them, when a safety hazard is found, the area where the safety hazard appears on the first hoisting state image is cut off and removed, and the remaining second hoisting state image is retained, that is, other image information that is considered to be safe is re-identified. If the second hoisting safety identification result is safe, the original hazard category is pushed to the monitoring end. If the second hoisting safety identification result still shows safety hazards, two hazard categories are output to the monitoring end.

[0042] Wherein, step S22 is further defined as follows: If the second hoisting safety identification result still contains hidden danger categories, the second hoisting state image is segmented and identified again until the hoisting safety identification result of the segmented image contains no hidden dangers.

[0043] That is, the first hoisting state image is continuously segmented and identified to ensure that all potential safety hazards can be accurately identified.

[0044] S23: When there are still potential safety hazards in the second hoisting safety identification result, the first hoisting state image and the summarized potential safety hazard categories are sent to the multimodal large model for model updating to obtain an updated multimodal large model.

[0045] Among them, the multimodal large model identifies safety hazards in the entire image, but performs secondary identification based on possible omissions in the identification process, thereby improving the accuracy and comprehensiveness of lifting safety identification, and is used to update the multimodal large model to further improve the recognition accuracy of the multimodal large model.

[0046] S3. The monitoring end receives the hidden dangers of hoisting, links the on-site broadcasting device to make voice broadcast reminders of the hidden dangers of hoisting, and at the same time, sends the hidden dangers of hoisting to the terminal of the on-site safety management personnel.

[0047] Among them, when step S2 determines that there is a hidden danger in the lifting operation, the on-site broadcasting device is linked to make a voice broadcast to remind the on-site workers to correct it in time; at the same time, the hidden danger information is converted into a hidden danger task and pushed to the relevant management personnel for disposal and closing the loop.

[0048] In this embodiment, based on the above safety hazard identification results, a hoisting hazard analysis report is generated in the form of a week, month or time period to assist project safety management personnel in carrying out targeted hazard education and training.

[0049] The beneficial effects of the present invention are: the feature identification of hoisting hazards is dispersed to the front-end model deployed at the construction site for real-time collection and analysis of the single feature of the hoisting state, and the multi-modal large model deployed in the cloud only performs multi-feature hazard analysis on images in the hoisting state without real-time full analysis of hoisting hazards, which not only reduces the computing power resource requirements of the front-end equipment, but also reduces the overall computing power resource pressure. Moreover, after the front-end model first identifies and analyzes the hoisting state, the cloud-based multi-modal large model is used to identify the hazards existing in the hoisting state, thereby improving the accuracy of multi-feature hazard identification. Embodiment 2

[0050] Please refer to Figure 2 A large-model-based lifting and hoisting safety monitoring system 1 includes a monitoring terminal 2 and a cloud 3, wherein the monitoring terminal 2 is used to implement step S1 in embodiment 1, and the cloud 3 is used to implement steps S2 and S3 in embodiment 1.

[0051] like Figure 3 As shown, the monitoring terminal 2 includes a lifting state analysis module that deploys a front-end model and performs data analysis, a 5G communication module for communicating with the cloud, terminals, etc., a high-precision positioning module for positioning, a data storage module for real-time storage of lifting construction images, a broadcast voice module for prompting, a display module for displaying operation information and prompt information, a camera module for acquiring lifting construction images, and a pan / tilt module according to the camera module. Among them, the on-site broadcasting device in Example 1 can be a broadcast voice module, or other on-site broadcasting devices, or a combination of the two.

[0052] Since the system / device described in the above embodiments of the present invention is a system / device used to implement the method of the above embodiments of the present invention, a person skilled in the art can understand the specific structure and deformation of the system / device based on the method described in the above embodiments of the present invention, and thus will not be described in detail here. All systems / devices used in the method of the above embodiments of the present invention belong to the scope of protection of the present invention.

[0053] It should be understood by those skilled in the art that embodiments of the present invention may be provided as methods, apparatuses or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0054] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (apparatus) and computer program products according to embodiments of the present invention. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions.

[0055] It should be noted that in the claims, any reference numerals placed between brackets shall not be construed as limiting the claims. The word "comprising" does not exclude the presence of components or steps not listed in the claims. The word "a" or "an" preceding a component does not exclude the presence of a plurality of such components. The invention may be implemented by means of hardware comprising several different components and by means of a suitably programmed computer. In the claims enumerating several means, several of these means may be embodied by the same hardware. The use of the words first, second, third, etc., is for convenience of expression only and does not indicate any order. These words may be understood as part of the component name.

[0056] In addition, it should be noted that, in the description of this specification, the description of the terms "one embodiment", "some embodiments", "embodiment", "example", "specific example" or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, unless they are contradictory.

[0057] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments after knowing the basic creative concept. Therefore, the claims should be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present invention.

[0058] Obviously, those skilled in the art can make various modifications and variations to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention should also include these modifications and variations.

Claims

1. A method for monitoring the safety of hoisting and lifting based on a large model, characterized in that: Includes steps: S1. The monitoring end obtains the hoisting construction image in the lifting process in real time, inputs the hoisting construction image into the front-end model deployed on the monitoring end, obtains the hoisting state recognition result, determines whether the hoisting state recognition result is a hoisting operation, and if so, sends the first hoisting state image in the hoisting operation in the hoisting construction image to the cloud; S2. The cloud receives the first hoisting status image, and inputs the first hoisting status image into the multimodal large model to obtain a first hoisting safety identification result, and determines whether the first hoisting safety identification result indicates that there is a hoisting hidden danger. If so, the hoisting hidden danger is pushed to the monitoring end.

2. A method for monitoring the safety of hoisting and lifting based on a large model according to claim 1, characterized in that: Also includes the steps: Collect pictures of hidden dangers in lifting, and label the pictures of hidden dangers in lifting, wherein the labels include description of hidden danger characteristics, correct operation requirements and hidden danger types; The marked lifting hidden danger pictures are input as prompt words into the multimodal big model for training to obtain the trained multimodal big model, and the trained multimodal big model is deployed in the cloud, wherein the multimodal big model includes a general big model for lifting hidden dangers and a hidden danger classification neural network, and the general big model for lifting hidden dangers identifies lifting safety hidden dangers on the lifting hidden danger pictures. If the lifting hidden danger is identified, the input content is vectorized and output to the hidden danger classification neural network, and the hidden danger category is output through the hidden danger classification neural network; The step S2 comprises the steps of: The cloud receives the first hoisting status image, and inputs the first hoisting status image into the multimodal large model to obtain a first hoisting safety identification result, and determines whether the first hoisting safety identification result indicates that there is a hoisting hidden danger. If so, the hidden danger category of the hoisting hidden danger is pushed to the monitoring end.

3. A large model-based hoisting safety monitoring method according to claim 2, characterized in that: In step S2, the hidden danger category of the lifting hidden danger is pushed to the monitoring end, including the steps of: When a hoisting hazard is identified and the hazard category is obtained, the first hoisting status image is segmented into a second hoisting status image which only excludes the area where the hazard category is located, and the second hoisting status image is re-input into the multimodal large model to obtain a second hoisting safety identification result, and the first hoisting safety identification result and the second hoisting safety identification result are summarized, and the summarized hazard category is pushed to the monitoring end.

4. A method for monitoring the safety of hoisting and lifting based on a large model according to claim 3, characterized in that: If the second hoisting safety identification result still contains hidden danger categories, the second hoisting state image is segmented and identified again until the hoisting safety identification result of the segmented image contains no hidden dangers.

5. The method for monitoring the safety of hoisting and lifting based on a large model according to claim 3 is characterized in that: Also includes the steps: When the second hoisting safety identification result still has safety hazards, the first hoisting state image and the summarized hazard categories are sent to the multimodal large model for model updating to obtain an updated multimodal large model.

6. A method for monitoring the safety of hoisting and lifting based on a large model according to any one of claims 1 to 5, characterized in that: The step S1 further comprises the steps of: If the lifting status identification result is a lifting operation, it is analyzed whether the pre-lifting work is complete. If it is not complete, a reminder message is issued. The pre-lifting work includes the acceptance of the lifting equipment required for the lifting operation, the certification of the lifting operators and the lifting warning.

7. A method for monitoring the safety of hoisting and lifting based on a large model according to any one of claims 1 to 5, characterized in that: The hoisting construction image is stored in real time in the monitoring terminal.

8. A method for monitoring the safety of hoisting and lifting based on a large model according to any one of claims 1 to 5, characterized in that: Also includes the steps: S3. The monitoring end receives the lifting hidden danger and links the on-site broadcasting device to perform voice broadcast reminder of the lifting hidden danger.

9. A method for monitoring the safety of hoisting and lifting based on a large model according to claim 8, characterized in that: The lifting hidden dangers will also be sent to the terminal of the on-site safety manager.

10. A large-scale model-based hoisting safety monitoring system, comprising: The monitoring end is used to obtain the hoisting construction image in the lifting process in real time, input the hoisting construction image into the front-end model deployed on the monitoring end, obtain the hoisting state recognition result, judge whether the hoisting state recognition result is a hoisting operation, and if so, send the first hoisting state image in the hoisting operation in the hoisting construction image to the cloud; The cloud end is used to receive the first hoisting status image, and input the first hoisting status image into the multimodal large model to obtain a first hoisting safety identification result, and determine whether the first hoisting safety identification result indicates that there is a hoisting hidden danger. If so, the hoisting hidden danger is pushed to the monitoring end.

Images