A high-position camera-based construction site inspection method and device

Abstract

The application discloses a kind of based on high camera construction site inspection method and device, the method comprises: obtaining target image by being in the high camera of controlled inspection state and being shot;Wherein, the high camera is located at the preset height of construction site;The target image is divided into multiple grid images, and whether at least one preset characteristic factor exists in each grid image is detected in turn using multimodal large model, and the main grid image containing characteristic factor is obtained;Obtain the hidden danger identification model corresponding to the characteristic factor contained in the main grid image, and identify the main grid image using the obtained hidden danger identification model, and obtain the risk information existing in the main grid image;Risk inspection list is generated based on the risk information of each main grid image;The present application can realize high-precision multi-target recognition under large-scale monitoring based on high camera, and automatically generate risk inspection list.

Description

Technical Field

[0001] This invention relates to the field of image recognition technology, and in particular to a construction site inspection method and apparatus based on a high-position camera. Background Technology

[0002] In construction scenarios, safety inspections are a core component of ensuring safe production on construction sites, and their efficiency and accuracy in risk identification directly impact the effectiveness of construction safety management. Currently, construction site inspection and monitoring primarily rely on two methods: manual on-site inspections and fixed-point video surveillance. However, manual inspections are limited by personnel's field of vision and pathways, failing to achieve continuous coverage of the entire construction site. Furthermore, the results of hazard identification are highly subjective and lack real-time responsiveness. While fixed monitoring can achieve continuous recording, the fixed camera angle makes it difficult to balance field of view and resolution. In wide-angle mode, the pixel ratio of distant targets is too small to support refined identification, and the system lacks automatic image content understanding capabilities. Hazard detection still relies on manual monitoring, resulting in a high rate of missed detections.

[0003] In recent years, some studies have attempted to apply general object detection models to construction site monitoring images, but existing solutions generally suffer from the following technical bottlenecks: First, the inherent contradiction between panoramic monitoring and detail recognition is difficult to reconcile, and insufficient target resolution in wide-angle images leads to a sharp drop in recognition accuracy; second, the recognition dimension is singular, with most solutions focusing only on the recognition of single violations such as safety helmets, lacking joint perception and professional analysis of multiple types of objects; third, the model output results are disconnected from business management, and cannot automatically generate a risk list that can guide on-site inspections.

[0004] Therefore, how to achieve high-precision multi-target identification under large-scale monitoring and automatically generate structured risk information is a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0005] The purpose of this invention is to provide a construction site inspection method and device based on a high-position camera, which can achieve high-precision multi-target identification under large-area monitoring and automatically generate structured risk information.

[0006] According to one aspect of the present invention, a construction site inspection method based on a high-position camera is provided, the method comprising:

[0007] Acquire target images captured by a high-position camera in a controlled inspection state; wherein the high-position camera is located at a predetermined high point on the construction site; The target image is divided into multiple grid images, and a multimodal large model is used to sequentially detect whether there is at least one preset feature factor in each grid image, so as to obtain a main grid image containing the feature factor; Obtain a hazard identification model corresponding to the feature factors contained in the main grid image, and use the obtained hazard identification model to identify the main grid image to obtain the risk information existing in the main grid image; A risk inspection checklist is generated based on the risk information from each main grid image.

[0008] Optionally, acquiring the target image captured by the high-position camera in a controlled inspection state includes: According to the preset inspection task, the parameters of the high-position camera are adjusted to the target position and target focal length; Multiple initial images are captured continuously using the adjusted high-position camera, and the current position and current focal length of the high-position camera after capturing the images are obtained. The perceptual hash value of each initial image is calculated using an image hashing algorithm, and the distance between each initial image is calculated based on the perceptual hash value of each initial image. The system determines whether the current position and the target position, as well as the current focal length and the target focal length, are consistent. If all distance values ​​are less than a preset distance threshold and the current position and the target position, as well as the current focal length and the target focal length, are consistent, the system determines that the high-position camera is in a controlled inspection state and uses any one of the multiple initial images as the target image.

[0009] Optionally, the step of acquiring a hazard identification model corresponding to the feature factors contained in the main grid image, and using the acquired hazard identification model to identify the main grid image to obtain the risk information existing in the main grid image, includes: The position information of the main grid image is obtained, and the high-position camera is adjusted according to the position information to obtain a high-definition close-up image corresponding to the main grid image; Identify regions of interest corresponding to various feature factors from the high-definition close-up images; wherein, the feature factors include at least one of the following: human body, lifting machinery, construction equipment; Various hazard identification models corresponding to the various feature factors contained in the high-definition close-up image are obtained, and the obtained hazard identification models are used to identify the corresponding areas of interest in order to obtain the corresponding risk information.

[0010] Optionally, the step of acquiring various hazard identification models corresponding to various feature factors contained in the high-definition close-up image, and using the acquired hazard identification models to identify the corresponding areas of interest in order to obtain the corresponding risk information, includes: Identify the human body attention area in the high-definition close-up image, calculate the image proportion of each human body attention area in the high-definition close-up image, and retain the crane machinery attention area whose image proportion is greater than a first preset proportion value; The safety hazard identification model is used to identify the retained human body areas of interest, and the corresponding human image information is obtained. The system performs posture detection, target object detection, and behavior detection on the facial image information, and determines whether the detection results conform to preset business rules. If not, it forms personnel risk information based on the facial image information and the detection results.

[0011] Optionally, the step of acquiring various hazard identification models corresponding to various feature factors contained in the high-definition close-up image, and using the acquired hazard identification models to identify the corresponding areas of interest in order to obtain the corresponding risk information, includes: Identify the areas of interest for lifting machinery in the high-definition close-up image, calculate the image proportion of each area of ​​interest for lifting machinery in the high-definition close-up image, and retain the areas of interest for lifting machinery whose image proportion is greater than a second preset proportion value; The crane machinery hazard identification model is used to identify the retained crane machinery areas of interest, thereby obtaining the type of crane machinery and the surrounding scene within the preset range of the crane machinery. The system acquires security devices located within the surrounding environment and determines whether the placement and / or type of the security devices conform to preset business rules. If not, crane risk information is generated based on the type of crane machinery, the placement location of the safety equipment, and the equipment type.

[0012] Optionally, the step of acquiring various hazard identification models corresponding to various feature factors contained in the high-definition close-up image, and using the acquired hazard identification models to identify the corresponding areas of interest in order to obtain the corresponding risk information, includes: Identify the areas of interest for construction machinery in the high-definition close-up image, calculate the image proportion of each area of ​​interest for construction machinery in the high-definition close-up image, and retain the areas of interest for construction machinery whose image proportion is greater than a third preset proportion value; The construction machinery hazard identification model is used to identify the areas of concern for the retained construction machinery, thereby obtaining the type of construction machinery and personnel characteristics within the preset range of the construction machinery; If the personnel characteristic information conforms to the preset business rules, it is determined that the construction equipment is in use, and construction equipment risk information is generated based on the type of construction equipment, the personnel characteristic information, and the usage status.

[0013] Optionally, generating a risk inspection list based on the risk information of each main grid image includes: For a main grid image, a personnel risk report is generated based on the personnel risk information of the human body attention area; wherein, the personnel risk report includes: human image information, risk type, high-definition close-up image of the risk type, location of the risk type, and time when the risk type was detected; The project construction stage is determined based on the type of lifting machinery, the type of construction equipment, and the surrounding environment. A machinery risk report is then generated based on the risk information of the lifting machinery and the construction equipment. The machinery risk report includes: the project construction stage, the lifting machinery and construction equipment present in the surrounding environment, and the existing risk information. Based on the personnel risk report and the machinery risk report, comprehensive risk information of the main grid image is generated, and the comprehensive risk information of all main grid images is integrated to form the risk inspection list.

[0014] To achieve the above objectives, the present invention also provides a construction site inspection device based on a high-position camera, the device comprising: The acquisition module is used to acquire target images captured by a high-position camera in a controlled inspection state; wherein the high-position camera is located at a preset high point of the construction site; The detection module is used to divide the target image into multiple grid images, and use a multimodal large model to sequentially detect whether there is at least one preset feature factor in each grid image, so as to obtain a main grid image containing the feature factor; The identification module is used to acquire a hazard identification model corresponding to the feature factors contained in the main grid image, and to use the acquired hazard identification model to identify the main grid image to obtain the risk information present in the main grid image; The generation module is used to generate a risk inspection list based on the risk information of each master grid image.

[0015] To achieve the above objectives, the present invention also provides a computer device, which specifically includes: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the construction site inspection method based on a high-position camera described above.

[0016] To achieve the above objectives, the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the construction site inspection method based on a high-position camera described above.

[0017] This invention provides a construction site inspection method and device based on a high-position camera. By acquiring target images captured by a high-position camera under controlled inspection, it ensures that the acquired images originate from a stable inspection process executed according to preset tasks without human intervention, thus guaranteeing the continuity and reproducibility of subsequent analysis from the source. A coarse-grained detection strategy using grid partitioning and a multimodal large model is employed to quickly locate the main grid image containing characteristic factors in the target image, achieving rapid target screening under a large field of view. Then, a hazard identification model corresponding to the characteristic factors is invoked to perform refined analysis of the main grid image, achieving accurate identification. Finally, a risk inspection list is generated, transforming scattered identification results into a structured risk list that can guide on-site management. This transforms the inspection work from indiscriminate general inspection to targeted key checks, significantly improving the accuracy and efficiency of construction site inspections. Attached Figure Description

[0018] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 This is a schematic diagram of an optional process for a construction site inspection method based on a high-position camera, as provided in Embodiment 1. Figure 2 This is a schematic diagram of an optional component structure of the construction site inspection device based on a high-position camera provided in Embodiment 2; Figure 3 This is a schematic diagram of an optional hardware structure for the computer device provided in Embodiment 3. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without inventive effort are within the scope of protection of this invention.

[0020] Example 1 This invention provides a construction site inspection method based on a high-position camera, such as... Figure 1 As shown, the method specifically includes the following steps: Step S101: Acquire the target image captured by a high-position camera in a controlled inspection state; wherein the high-position camera is located at a preset high point of the construction site.

[0021] In this embodiment, high-position cameras are installed at predetermined high points on the construction site, such as on-site installation on tower cranes or rooftops, or using drone-borne cameras, and connected to the video management platform according to the GB / T28181 standard protocol. Users publish inspection tasks for the construction site on the video platform. The video platform accepts the inspection tasks and allocates high-risk cameras to the inspection area, conducting inspections on a regular schedule and according to frequency requirements. Within the inspection area, the high-risk cameras automatically rotate to a predetermined position, control their focal length, and capture video footage from the current viewpoint, simultaneously obtaining parameters such as the camera's current focal length, gimbal azimuth angle, and pitch angle. Using the continuously captured video footage as initial images, and after confirming a controlled inspection state, the target image is selected from the multiple initial images. This ensures that the acquired images originate from a stable inspection process executed according to the preset task, without human intervention or equipment malfunction, guaranteeing the continuity and reproducibility of subsequent image analysis from the source, and resolving issues such as inspection interruptions and invalid data caused by uncertain camera status.

[0022] Step S102: Divide the target image into multiple grid images, and use a multimodal large model to sequentially detect whether there is at least one preset feature factor in each grid image, so as to obtain a main grid image containing the feature factor.

[0023] In this embodiment, the entire target image captured by the high-position camera is uniformly divided into grids without overlap or omission. The division rules can be preset according to the actual field of view requirements and recognition accuracy requirements of construction site inspection. Preferably, a 3×3 nine-grid division method is adopted to decompose the large field of view target image into multiple independent grid images, ensuring that the entire construction site scene corresponding to the target image is divided into each grid, realizing full coverage of subsequent detection and avoiding missed detection of feature factors due to the large image range. At the same time, three core elements, namely human body, lifting machinery, and construction equipment, are preset feature factors for this detection. These feature factors are the core regulatory objects of construction site safety inspection, defining clear recognition targets and directions for multimodal large-scale model detection work, and adapting to the safety management business needs of the construction site. Subsequently, a multimodal large-scale model is invoked to perform independent image content understanding and feature recognition on each of the divided grid images in a predetermined order. Relying on its image analysis and content understanding capabilities, the multimodal large-scale model performs a full-domain scan and feature matching of the visual content within each grid image, determining whether at least one preset feature element exists within that grid image, such as a human body, lifting machinery, or construction equipment. The detection process for each grid image is independent, avoiding interference between the image content of different grids and ensuring the accuracy of feature element determination. Grid images that are determined by the multimodal large-scale model to contain at least one preset feature element are designated as the core grid images for analysis. The position coordinates and grid number of each core grid image in the original target image are simultaneously recorded, providing precise spatial positioning data for subsequent high-position cameras to focus and magnify the area to obtain high-definition close-up images. Grid images that do not detect any preset feature elements are deemed to have no analytical value and are not subject to further refined identification and analysis. This reduces unnecessary computation and processing costs, significantly improving the overall analysis efficiency of construction site inspection images.

[0024] Step S103: Obtain the hazard identification model corresponding to the feature factors contained in the main grid image, and use the obtained hazard identification model to identify the main grid image to obtain the risk information existing in the main grid image.

[0025] In this embodiment, after determining the main grid image containing human bodies, lifting machinery, or construction equipment, a corresponding dedicated hazard identification model is matched for each main grid image based on the preset association mapping relationship between feature factors and hazard identification models. Different types of feature factors correspond to suitable professional hazard identification models, achieving accurate matching between feature factors and hazard identification models and avoiding the situation where a single model generally identifies different types of feature factors. Subsequently, the main grid image is input into the matched hazard identification model, which performs refined feature analysis and risk detection on the target content within the main grid image, conducts targeted identification and judgment based on the attributes of feature factors, and finally outputs specific risk information related to feature factors in the main grid image. This risk information clearly includes core content such as the hazard type and specific location of the risk corresponding to the feature factors within the main grid image, ensuring that the risk information of each main grid image is accurately extracted.

[0026] Step S104: Generate a risk inspection list based on the risk information of each main grid image.

[0027] In this embodiment, after completing the risk information identification and extraction of all main grid images containing preset feature factors, the process proceeds to the stage of generating a risk inspection list based on the risk information of each main grid image. This stage is a crucial connecting step from intelligent risk identification at the construction site to the implementation of safety inspections, and it is also the core process of transforming scattered grid risk data into standardized and implementable inspection guidance. First, the risk information of all main grid images identified during this automatic inspection is integrated. The risk information covers all specific risk information corresponding to the three preset feature factors: human body, lifting machinery, and construction equipment. This includes the type of personnel risk detected in each main grid image, the risk details of lifting machinery and construction equipment, and information such as the actual spatial location of the construction site corresponding to each risk, the risk detection time, and high-definition close-up image evidence, ensuring that the risk information of each grid is completely summarized. Subsequently, combining the on-site management zoning standards of construction sites with the operational needs of safety inspections, all integrated main grid image risk information was systematically sorted and categorized. It can be classified according to dimensions such as pre-set inspection zones, risk types (personnel risks, machinery and equipment risks), and associated work scenarios. Simultaneously, each risk information is correlated and matched with the project risk database to align with the actual safety management requirements of construction sites. Based on the standardized risk data after sorting and categorizing, a unified risk inspection list is generated. This list clearly lists the specific construction site area corresponding to each main grid image, the types of characteristic factors existing within the area, the specific risk information detected, the corresponding evidence information, and the location / time of the risk, comprehensively presenting the risk distribution, specific hazard types, and corresponding details throughout the entire construction site. The generated risk inspection list will be directly pushed to project safety officers and other on-site safety management personnel, providing them with clear and targeted guidance for carrying out on-site safety inspections. This will enable them to focus on the risk areas corresponding to each main grid image for targeted on-site verification and hazard rectification, replacing the traditional indiscriminate, blind inspection model. At the same time, the risk inspection list will also serve as an important basis for subsequent hazard handling and rectification review, providing data support for the closed-loop advancement of construction site safety management.

[0028] In this embodiment, by acquiring target images captured by a high-position camera under controlled inspection, it is ensured that the acquired target images originate from a stable inspection process executed according to preset tasks without human intervention, thus guaranteeing the continuity and reproducibility of subsequent analysis from the source. A coarse-grained detection strategy of grid division and multimodal large model is adopted to quickly locate the main grid image with characteristic factors in the target image, realizing rapid target screening under a large field of view. Then, the hazard identification model corresponding to the characteristic factors is called to perform fine analysis on the main grid image, realizing accurate identification. Finally, a risk inspection list is generated, transforming the scattered identification results into a structured risk list that can guide on-site management, so that the inspection work is transformed from indiscriminate full-area inspection to targeted key inspection, significantly improving the accuracy and work efficiency of construction site inspection.

[0029] Specifically, the step S101 of acquiring the target image captured by the high-position camera in a controlled inspection state includes: Step A1: According to the preset inspection task, adjust the parameters of the high-position camera to the target position and target focal length; Step A2: Use the adjusted high-position camera to continuously capture multiple initial images, and obtain the current position and current focal length of the high-position camera after capturing the images; Step A3: Calculate the perceptual hash value of each initial image using an image hashing algorithm, and calculate the distance between each initial image based on the perceptual hash value of each initial image. Step A4: Determine whether the current position and the target position, and the current focal length and the target focal length are all consistent; if all distance values ​​are less than the preset distance threshold and the current position and the target position, and the current focal length and the target focal length are consistent, then determine that the high-position camera is in a controlled inspection state, and take any one of the multiple initial images as the target image.

[0030] In this embodiment, after receiving the inspection task from the video platform, the high-position camera adjusts to the area to be inspected according to the target focal length and target position in the task, and continuously captures multiple initial images. After capturing, the current focal length and current position are obtained and compared with the target focal length and target position given in the inspection task. If they match, the high-position camera is in a controlled state; if they do not match, it indicates that another user is using the high-position camera, which will cause the automatic completion of the inspection task issued by the platform to fail. In this case, the current inspection task will be paused, waiting for the high-position camera to be in a controlled inspection state before restarting or calling another high-position camera to execute the inspection task. At the same time, a hash algorithm is used to calculate the perceptual hash value of each initial image, and the perceptual hash value is used to calculate the distance value between each initial image. If the distance value is less than a preset distance threshold, it indicates that the high-position camera has been in the current inspection area and has not been affected by other factors to deviate from its current position. By employing dual verification through the hash distance between initial images and the built-in physical parameters of the high-position camera, this method is more robust than a single verification approach, effectively identifying uncontrolled states caused by human manipulation, network jitter, and mechanical hysteresis. Furthermore, the perceptual hash algorithm quantifies image content stability into comparable numerical distances, providing an objective and reproducible technical standard for image stability determination and avoiding subjective threshold settings.

[0031] Specifically, step S103, which involves obtaining a hazard identification model corresponding to the feature factors contained in the main grid image and using the obtained hazard identification model to identify the main grid image to obtain risk information present in the main grid image, includes: Step B1: Obtain the position information of the main grid image, and adjust the high-position camera according to the position information to obtain a high-definition close-up image corresponding to the main grid image; Step B2: Identify the regions of interest corresponding to various feature factors from the high-definition close-up image; wherein the feature factors include at least one of the following: human body, lifting machinery, construction equipment; Step B3: Obtain various hazard identification models corresponding to the various feature factors contained in the high-definition close-up image, and use the obtained hazard identification models to identify the corresponding areas of interest in order to obtain the corresponding risk information.

[0032] In this embodiment, the spatial location information of the construction site corresponding to the main grid image is first obtained. Based on this location information, a pan-tilt and focus control command is sent to the high-position camera to control the high-position camera to adjust the azimuth and pitch angles of the pan-tilt and adjust the focus to focus on the site area corresponding to the main grid image and obtain a high-definition close-up image of the area to improve the image resolution of the target area and provide a clear image foundation for subsequent refined identification. Target detection and region labeling are performed on the acquired high-definition close-up image to accurately identify three types of feature factors from the image: human body, lifting machinery, and construction equipment. The pixel areas corresponding to each feature factor in the high-definition close-up image are then delineated, and these pixels are classified... The region serves as the area of ​​interest for the corresponding feature factor, enabling the separation of regions with different feature factors and avoiding interference from different feature factors in subsequent identification. Based on the pre-stored correlation mapping relationship between feature factors and hazard identification models, and according to the areas of interest for each feature factor defined in the high-definition close-up image, a dedicated hazard identification model adapted to each feature factor is retrieved. Then, the image data of each area of ​​interest is input into the corresponding hazard identification model, and each dedicated hazard identification model conducts targeted and refined risk detection and identification for its respective area of ​​interest. Combining the business requirements of construction site safety inspection, the specific safety risks corresponding to the feature factors in each area of ​​interest are analyzed, and finally, accurate risk information matching each area of ​​interest is output.

[0033] Furthermore, step B3 involves acquiring various hazard identification models corresponding to different feature factors contained in the high-definition close-up image, and using these models to identify the corresponding areas of interest to obtain the corresponding risk information. This includes: Step B311: Identify the human body attention area in the high-definition close-up image, calculate the image proportion of each human body attention area in the high-definition close-up image, and retain the crane machinery attention area whose image proportion is greater than the first preset proportion value; Step B312: Use the safety hazard identification model to identify the retained human body areas of interest and obtain the corresponding human image information; Step B313: Perform posture detection, target object detection, and behavior detection on the facial image information, and determine whether the detection results meet the preset business rules. If not, form personnel risk information based on the facial image information and the detection results.

[0034] In this embodiment, after accurately identifying and calibrating the human body area of ​​interest in the high-definition close-up image, the proportion of the number of pixels in each calibrated human body area of ​​interest to the total number of pixels in the high-definition close-up image is calculated. The human body areas of interest are then filtered according to a preset first preset proportion value, removing distant or extremely small human targets whose image proportion is lower than the threshold, and retaining only valid human body areas of interest whose image proportion is greater than the first preset proportion value. This avoids low-resolution or invalid human targets interfering with subsequent identification results and ensures the effectiveness of personnel risk identification. In particular, the first preset proportion value is set according to the actual identification needs of construction site inspections, and in this embodiment, it is preferably 1%.

[0035] The selected and retained effective human body areas of interest are input into a dedicated safety hazard identification model for feature recognition and information extraction, yielding the corresponding human image information within each area of ​​interest. This human image information includes at least core feature information such as human contours and key points. Based on the extracted human image information, a comprehensive detection and analysis is conducted, sequentially performing human posture detection, target object detection, and human behavior detection. Posture detection identifies dangerous human postures such as climbing and falling; target object detection identifies the wearing of safety protective equipment such as helmets and safety belts for working at heights; and behavior detection identifies dangerous work behaviors such as smoking, working near edges without protection, and trespassing into dangerous areas. After the detection is completed, the detection results for posture, target object, and behavior are reviewed and verified according to the preset business rules for construction site safety inspections. For example, the overlap between the safety equipment detection frame and the human image information detection frame is determined by calculating the intersection-union ratio (IUU) to eliminate false detections caused by image occlusion or changes in lighting.

[0036] If the detection result does not conform to the preset business rules, it is determined that there is a personnel safety hazard in the area of ​​interest for that person. The system then integrates the location of the area of ​​interest corresponding to the person's image information with the specific abnormal detection results to form standardized personnel risk information, clearly marking core elements such as the hazard type and the specific location of the hazard. Through multi-dimensional detection of complex scenarios, the false alarm rate is significantly reduced, transforming safety management regulations into calculable judgment logic, making the identification results business-interpretable, and directly supporting compliance checks.

[0037] Furthermore, step B3 involves acquiring various hazard identification models corresponding to different feature factors contained in the high-definition close-up image, and using these models to identify the corresponding areas of interest to obtain the corresponding risk information. This includes: Step B321: Identify the crane machinery interest area in the high-definition close-up image, calculate the image proportion of each crane machinery interest area in the high-definition close-up image, and retain crane machinery interest areas with an image proportion greater than a second preset proportion value; Step B322: Use the crane machinery hazard identification model to identify the retained crane machinery areas of interest, and obtain the type of crane machinery and the surrounding scene within the preset range of the crane machinery; Step B323: Obtain the security devices located in the surrounding scene, and determine whether the placement location and / or device type of the security devices conform to the preset business rules; Step B324: If not, generate crane risk information based on the type of crane machinery, the placement location of the safety equipment, and the equipment type.

[0038] In this embodiment, after accurately identifying and contouring the areas of interest for the lifting machinery within the high-definition close-up image, the proportion of pixels in each area of ​​interest to the total number of pixels in the high-definition close-up image is calculated. The areas of interest are then filtered according to a preset second ratio value, eliminating low-resolution and distant lifting machinery targets with image ratios below the threshold. Only valid areas of interest with image ratios greater than the second preset ratio value are retained, ensuring that the target area has sufficient resolution to support subsequent accurate identification and guaranteeing the effectiveness of lifting machinery risk identification. Specifically, the second preset ratio value can be set according to the actual identification needs of construction site inspections; in this embodiment, it is preferably 8%.

[0039] The selected areas of interest for the retained lifting machinery are input into a dedicated lifting machinery hazard identification model for in-depth feature analysis and identification. This model can accurately identify the specific type of lifting machinery, including common construction site lifting machinery types such as excavators, hoists (construction elevators), tower cranes, truck cranes, crawler cranes, and jib cranes. Simultaneously, it performs a full-domain detection of the surrounding environment within the preset working radius of the lifting machinery, completely extracting relevant information about the environment, facilities, and personnel, forming corresponding lifting machinery type information and surrounding environment information. From the extracted surrounding environment information, various safety devices are accurately identified, specifically including warning lines, safety warning signs, and other dedicated safety protection equipment for lifting machinery operations. Then, based on the preset business rules for the safety management of construction site lifting machinery operations, the identified safety equipment undergoes a dual compliance judgment: firstly, it determines whether the placement of the safety equipment is within the reasonable protection area for lifting machinery operations; secondly, it determines whether the type of safety equipment matches the corresponding lifting machinery operation type and protection requirements. Simultaneously, it verifies the completeness of the safety equipment settings based on the rules, ensuring that the safety protection measures around the lifting machinery meet the regulatory requirements.

[0040] If it is determined that the placement of safety equipment does not comply with preset business rules, the equipment type does not match the protection requirements of the lifting machinery, or there are compliance issues such as missing safety equipment, then a safety hazard is identified in the area of ​​concern for that lifting machinery. At this point, the specific type of lifting machinery in the area, the actual placement of the safety equipment, and the non-compliant equipment types are integrated, along with other anomalies found in the surrounding environment, to generate standardized crane risk information. This risk information clearly marks key elements such as the type of lifting machinery to which the hazard belongs, the specific location of the hazard, and details of the non-compliance of the safety equipment, providing a clear and specific basis for subsequent risk management. Through the joint analysis of equipment type, surrounding environment, and safety facilities, the risk assessment capability is improved from simply identifying machinery to understanding the operational scenario.

[0041] Furthermore, step B3 involves acquiring various hazard identification models corresponding to different feature factors contained in the high-definition close-up image, and using these models to identify the corresponding areas of interest to obtain the corresponding risk information. This includes: Step B331: Identify the areas of interest for construction machinery in the high-definition close-up image, calculate the image proportion of each area of ​​interest for construction machinery in the high-definition close-up image, and retain the areas of interest for construction machinery whose image proportion is greater than the third preset proportion value; Step B332: Use the construction machinery hazard identification model to identify the areas of concern for the retained construction machinery, and obtain the type of construction machinery and the personnel characteristics information within the preset range of the construction machinery; Step B333: If the personnel characteristic information conforms to the preset business rules, it is determined that the construction equipment is in use, and construction equipment risk information is generated based on the type of construction equipment, the personnel characteristic information and the use status.

[0042] In this embodiment, after accurately identifying and contouring the areas of interest for construction machinery within the high-definition close-up image, the proportion of pixels in each area of ​​interest for construction machinery to the total number of pixels in the high-definition close-up image is calculated. The areas of interest for construction machinery are then filtered according to a preset third ratio value. Low-resolution and distant construction machinery targets with an image ratio below this threshold are removed, and only valid areas of interest for construction machinery with an image ratio greater than the third preset ratio value are retained. This ensures that the target area has sufficient resolution to support subsequent accurate identification and guarantees the effectiveness of construction machinery risk identification. Specifically, the third preset ratio value can be set according to the actual identification needs of construction site inspections; in this embodiment, it is preferably 2%.

[0043] The selected and retained areas of interest for construction machinery are input into a dedicated construction machinery hazard identification model for in-depth feature analysis and identification. This model can accurately identify the specific types of construction machinery, including common construction machinery types on construction sites such as circular saws, electric drills, angle grinders, rebar cutters, rammers, welding machines, and winches. At the same time, it performs full-area detection on the area within the preset working range of the construction machinery and extracts personnel feature information within that range. The personnel feature information includes at least the presence of personnel within the working range, the relative position of personnel and construction machinery, and whether personnel have taken basic safety protection measures.

[0044] Based on pre-defined business rules for construction machinery operations at construction sites, the extracted personnel characteristic information is assessed for compliance. These pre-defined business rules, combined with safety management requirements for construction machinery operations, include situations such as personnel remaining within the operating area of ​​the machinery, personnel not maintaining a safe operating distance from operating machinery, and machinery operators not taking corresponding protective measures. If the personnel characteristic information matches the aforementioned pre-defined business rules, it is determined that the construction machinery is actually in use and poses a safety hazard during operation. At this point, core information such as the specific type of construction machinery in the area, details of personnel characteristics within the operating area, and the actual usage status of the construction machinery are integrated to generate standardized construction machinery risk information. This risk information clearly indicates the type of construction machinery to which the hazard belongs, details of personnel violations at the work site, and the machinery's usage status, providing specific and clear evidence for subsequent risk management and on-site verification. Through human-machine correlation analysis and usage status inference, the technical blind spots in dynamic risk assessment of construction machinery are resolved.

[0045] Specifically, the generation of the risk inspection list based on the risk information of each main grid image includes: Step C1: For a main grid image, generate a personnel risk report based on the personnel risk information of the human body attention area; wherein, the personnel risk report includes: human image information, risk type, high-resolution close-up image of the risk type, location of the risk type, and time when the risk type was detected; Step C2: Determine the current project construction stage based on the type of lifting machinery, the type of construction equipment, and the surrounding environment, and generate a machinery risk report based on the lifting machinery risk information and the construction equipment risk information; wherein, the machinery risk report includes: the project construction stage, the lifting machinery and construction equipment present in the surrounding environment, and the existing risk information; Step C3: Generate comprehensive risk information for the main grid image based on the personnel risk report and the machinery risk report, and integrate the comprehensive risk information of all main grid images to form the risk inspection list.

[0046] In this embodiment, for a single main grid image, based on the personnel risk information obtained from the identification of the human body area of ​​interest, multi-dimensional information is collected to generate a standardized personnel risk report. This personnel risk report integrates risk-related elements of the human body area of ​​interest, specifically including the facial image information extracted from the human body area of ​​interest, the specific personnel risk type detected and determined, and a high-definition close-up image corresponding to the risk type as visual evidence. It also combines the actual spatial location of the risk type determined by the pan-tilt angle, tilt angle, and preset point information of the high-position camera, as well as the precise time the system detected the risk type. All elements correspond to each other, forming a traceable and verifiable personnel risk report, providing complete personnel risk evidence for subsequent hazard mitigation. First, by combining the characteristics of machinery and equipment used in each construction phase of the construction site, and based on the specific types of lifting machinery and construction equipment identified in the surrounding scene of the main grid image, as well as the characteristics of the surrounding working environment, the project construction phase of the area is matched and determined, such as the foundation construction phase, the main structure construction phase, and the decoration and finishing phase. Then, the lifting machinery risk information and construction equipment risk information identified under the main grid image are integrated to generate a machinery risk report. This report clearly marks the project construction phase of the area, lists the specific types of all lifting machinery and construction equipment present in the surrounding scene, and corresponds to the specific risk information found in the detection of each type of machinery and equipment, realizing the correlation and matching between machinery risks and construction phases, and meeting the actual safety management needs of project construction.

[0047] For the main grid image, the corresponding personnel risk report and machinery risk report are integrated across the entire area, combining risk elements from both personnel and machinery dimensions to form comprehensive risk information for the main grid image. This ensures that all safety risks in the construction area corresponding to this grid are fully collected without omission. Subsequently, the comprehensive risk information of all main grid images containing characteristic factors identified during this automatic inspection is uniformly summarized and categorized and arranged according to dimensions such as construction area location, risk type, and project construction stage. Finally, a standardized risk inspection checklist is formed. This checklist clearly presents the comprehensive risk distribution, specific risk types, and corresponding risk details of each area of ​​the construction site. It can be directly pushed to the project safety officer, providing clear and systematic guidance for them to conduct targeted on-site risk verification and hazard rectification.

[0048] In this embodiment, the above overall solution achieves the following technical effects: (1) From passive to active: Through automatic inspection by high-point cameras, the transformation from "manual monitoring" to "intelligent patrol" is realized, achieving 24 / 7 uninterrupted and full-coverage active safety monitoring. (2) Accurate and efficient identification: The two-level identification architecture of "large model coarse screening + special model fine identification" is adopted, which takes into account the wide range of detection and the professionalism of identification types, greatly improving the detection rate and identification accuracy of safety hazards in complex construction site environments. (3) Timely closed-loop handling: The online closed-loop management from automatic identification of hazards, real-time early warning, responsibility push to rectification review is realized, which greatly shortens the response time of hazards and improves the handling efficiency. (4) Data-driven management decision-making: Through automatic analysis of massive identification data, trend reports and risk predictions are generated, upgrading safety management from experience-driven to data-driven, helping project managers to grasp the safety situation and formulate accurate preventive measures. (5) Reduce labor costs: Significantly reduce the mechanical inspection and screen monitoring work of safety officers, so that they can focus more on more valuable work such as risk verification and education briefing, and optimize the allocation of human resources.

[0049] Example 2 This invention provides a construction site inspection device based on a high-position camera, such as... Figure 2 As shown, the device specifically includes the following components: The acquisition module 201 is used to acquire target images captured by a high-position camera in a controlled inspection state; wherein the high-position camera is located at a preset high point of the construction site; The detection module 202 is used to divide the target image into multiple grid images, and use a multimodal large model to sequentially detect whether there is at least one preset feature factor in each grid image, so as to obtain a main grid image containing the feature factor; The identification module 203 is used to acquire a hazard identification model corresponding to the feature factors contained in the main grid image, and to use the acquired hazard identification model to identify the main grid image to obtain the risk information present in the main grid image; The generation module 204 is used to generate a risk inspection list based on the risk information of each master grid image.

[0050] Specifically, the acquisition module 201 is used for: According to the preset inspection task, the parameters of the high-position camera are adjusted to the target position and target focal length; Multiple initial images are captured continuously using the adjusted high-position camera, and the current position and current focal length of the high-position camera after capturing the images are obtained. The perceptual hash value of each initial image is calculated using an image hashing algorithm, and the distance between each initial image is calculated based on the perceptual hash value of each initial image. The system determines whether the current position and the target position, as well as the current focal length and the target focal length, are consistent. If all distance values ​​are less than a preset distance threshold and the current position and the target position, as well as the current focal length and the target focal length, are consistent, the system determines that the high-position camera is in a controlled inspection state and uses any one of the multiple initial images as the target image.

[0051] Specifically, the identification module 203 is used for: The position information of the main grid image is obtained, and the high-position camera is adjusted according to the position information to obtain a high-definition close-up image corresponding to the main grid image; Identify regions of interest corresponding to various feature factors from the high-definition close-up images; wherein, the feature factors include at least one of the following: human body, lifting machinery, construction equipment; Various hazard identification models corresponding to the various feature factors contained in the high-definition close-up image are obtained, and the obtained hazard identification models are used to identify the corresponding areas of interest in order to obtain the corresponding risk information.

[0052] Specifically, when the identification module 203 performs the function of acquiring various hazard identification models corresponding to various feature factors contained in the high-definition close-up image, and using the acquired hazard identification models to identify the corresponding areas of interest in order to obtain the corresponding risk information, it is used for: Identify the human body attention area in the high-definition close-up image, calculate the image proportion of each human body attention area in the high-definition close-up image, and retain the crane machinery attention area whose image proportion is greater than a first preset proportion value; The safety hazard identification model is used to identify the retained human body areas of interest, and the corresponding human image information is obtained. The system performs posture detection, target object detection, and behavior detection on the facial image information, and determines whether the detection results conform to preset business rules. If not, it forms personnel risk information based on the facial image information and the detection results.

[0053] Specifically, when executing the function of the identification module 203 to acquire various hazard identification models corresponding to various feature factors contained in the high-definition close-up image, and to use the acquired hazard identification models to identify the corresponding areas of interest in order to obtain the corresponding risk information, it is also used for: Identify the areas of interest for lifting machinery in the high-definition close-up image, calculate the image proportion of each area of ​​interest for lifting machinery in the high-definition close-up image, and retain the areas of interest for lifting machinery whose image proportion is greater than a second preset proportion value; The crane machinery hazard identification model is used to identify the retained crane machinery areas of interest, thereby obtaining the type of crane machinery and the surrounding scene within the preset range of the crane machinery. The system acquires security devices located within the surrounding environment and determines whether the placement and / or type of the security devices conform to preset business rules. If not, crane risk information is generated based on the type of crane machinery, the placement location of the safety equipment, and the equipment type.

[0054] Specifically, when executing the function of the identification module 203 to acquire various hazard identification models corresponding to various feature factors contained in the high-definition close-up image, and to use the acquired hazard identification models to identify the corresponding areas of interest in order to obtain the corresponding risk information, it is also used for: Identify the areas of interest for construction machinery in the high-definition close-up image, calculate the image proportion of each area of ​​interest for construction machinery in the high-definition close-up image, and retain the areas of interest for construction machinery whose image proportion is greater than a third preset proportion value; The construction machinery hazard identification model is used to identify the areas of concern for the retained construction machinery, thereby obtaining the type of construction machinery and personnel characteristics within the preset range of the construction machinery; If the personnel characteristic information conforms to the preset business rules, it is determined that the construction equipment is in use, and construction equipment risk information is generated based on the type of construction equipment, the personnel characteristic information, and the usage status.

[0055] Specifically, the generation module 204 is used for: For a main grid image, a personnel risk report is generated based on the personnel risk information of the human body attention area; wherein, the personnel risk report includes: human image information, risk type, high-definition close-up image of the risk type, location of the risk type, and time when the risk type was detected; The project construction stage is determined based on the type of lifting machinery, the type of construction equipment, and the surrounding environment. A machinery risk report is then generated based on the risk information of the lifting machinery and the construction equipment. The machinery risk report includes: the project construction stage, the lifting machinery and construction equipment present in the surrounding environment, and the existing risk information. Based on the personnel risk report and the machinery risk report, comprehensive risk information of the main grid image is generated, and the comprehensive risk information of all main grid images is integrated to form the risk inspection list.

[0056] Example 3 This embodiment also provides a computer device, such as a smartphone, tablet computer, laptop computer, desktop computer, rack server, blade server, tower server, or cabinet server (including a standalone server or a server cluster composed of multiple servers), etc., capable of executing programs. Figure 3 As shown, the computer device 30 in this embodiment includes, but is not limited to, a memory 301 and a processor 302 that are communicatively connected to each other via a system bus. It should be noted that... Figure 3 Only a computer device 30 with components 301-302 is shown; however, it should be understood that it is not required to implement all of the components shown, and more or fewer components may be implemented instead.

[0057] In this embodiment, the memory 301 (i.e., the readable storage medium) includes flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 301 may be an internal storage unit of the computer device 30, such as the hard disk or memory of the computer device 30. In other embodiments, the memory 301 may also be an external storage device of the computer device 30, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the computer device 30. Of course, the memory 301 may include both the internal storage unit and the external storage device of the computer device 30. In this embodiment, the memory 301 is typically used to store the operating system and various application software installed on the computer device 30. In addition, the memory 301 may also be used to temporarily store various types of data that have been output or will be output.

[0058] In some embodiments, processor 302 may be a central processing unit (CPU), controller, microcontroller, microprocessor, or other data processing chip. This processor 302 is typically used to control the overall operation of the computer device 30.

[0059] Specifically, in this embodiment, the processor 302 is used to execute the program of the construction site inspection method based on a high-position camera stored in the memory 301. When the program of the construction site inspection method based on a high-position camera is executed, it performs the following steps: Acquire target images captured by a high-position camera in a controlled inspection state; wherein the high-position camera is located at a predetermined high point on the construction site; The target image is divided into multiple grid images, and a multimodal large model is used to sequentially detect whether there is at least one preset feature factor in each grid image, so as to obtain a main grid image containing the feature factor; Obtain a hazard identification model corresponding to the feature factors contained in the main grid image, and use the obtained hazard identification model to identify the main grid image to obtain the risk information existing in the main grid image; A risk inspection checklist is generated based on the risk information from each main grid image.

[0060] For a detailed description of the above method steps, please refer to Example 1. This example will not be repeated here.

[0061] Example 4 This embodiment also provides a computer-readable storage medium, such as flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, disk, optical disk, server, app store, etc., which stores a computer program. When the computer program is executed by a processor, it implements the following method steps: Acquire target images captured by a high-position camera in a controlled inspection state; wherein the high-position camera is located at a predetermined high point on the construction site; The target image is divided into multiple grid images, and a multimodal large model is used to sequentially detect whether there is at least one preset feature factor in each grid image, so as to obtain a main grid image containing the feature factor; Obtain a hazard identification model corresponding to the feature factors contained in the main grid image, and use the obtained hazard identification model to identify the main grid image to obtain the risk information existing in the main grid image; A risk inspection checklist is generated based on the risk information from each main grid image.

[0062] For a detailed description of the above method steps, please refer to the first embodiment. This embodiment will not repeat the details here.

[0063] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0064] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0065] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method.

[0066] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A construction site inspection method based on a high-position camera, characterized in that, The method includes: Acquire target images captured by a high-position camera in a controlled inspection state; wherein the high-position camera is located at a predetermined high point on the construction site; The target image is divided into multiple grid images, and a multimodal large model is used to sequentially detect whether there is at least one preset feature factor in each grid image, so as to obtain a main grid image containing the feature factor; Obtain a hazard identification model corresponding to the feature factors contained in the main grid image, and use the obtained hazard identification model to identify the main grid image to obtain the risk information existing in the main grid image; A risk inspection checklist is generated based on the risk information from each main grid image.

2. The construction site inspection method based on a high-position camera according to claim 1, characterized in that, The acquisition of the target image captured by the high-position camera in a controlled inspection state includes: According to the preset inspection task, the parameters of the high-position camera are adjusted to the target position and target focal length; Multiple initial images are captured continuously using the adjusted high-position camera, and the current position and current focal length of the high-position camera after capturing the images are obtained. The perceptual hash value of each initial image is calculated using an image hashing algorithm, and the distance between each initial image is calculated based on the perceptual hash value of each initial image. The system determines whether the current position and the target position, as well as the current focal length and the target focal length, are consistent. If all distance values ​​are less than a preset distance threshold and the current position and the target position, as well as the current focal length and the target focal length, are consistent, the system determines that the high-position camera is in a controlled inspection state and uses any one of the multiple initial images as the target image.

3. The construction site inspection method based on a high-position camera according to claim 2, characterized in that, The step of acquiring a hazard identification model corresponding to the feature factors contained in the main grid image, and using the acquired hazard identification model to identify the main grid image to obtain the risk information existing in the main grid image, includes: The position information of the main grid image is obtained, and the high-position camera is adjusted according to the position information to obtain a high-definition close-up image corresponding to the main grid image; Identify regions of interest corresponding to various feature factors from the high-definition close-up images; wherein, the feature factors include at least one of the following: human body, lifting machinery, construction equipment; Various hazard identification models corresponding to the various feature factors contained in the high-definition close-up image are obtained, and the obtained hazard identification models are used to identify the corresponding areas of interest in order to obtain the corresponding risk information.

4. The construction site inspection method based on a high-position camera according to claim 3, characterized in that, The process of acquiring various hazard identification models corresponding to various feature factors contained in the high-definition close-up image, and using the acquired hazard identification models to identify the corresponding areas of interest in order to obtain the corresponding risk information, includes: Identify the human body attention area in the high-definition close-up image, calculate the image proportion of each human body attention area in the high-definition close-up image, and retain the crane machinery attention area whose image proportion is greater than a first preset proportion value; The safety hazard identification model is used to identify the retained human body areas of interest, and the corresponding human image information is obtained. The system performs posture detection, target object detection, and behavior detection on the facial image information, and determines whether the detection results conform to preset business rules. If not, it forms personnel risk information based on the facial image information and the detection results.

5. The construction site inspection method based on a high-position camera according to claim 3, characterized in that, The process of acquiring various hazard identification models corresponding to various feature factors contained in the high-definition close-up image, and using the acquired hazard identification models to identify the corresponding areas of interest in order to obtain the corresponding risk information, includes: Identify the areas of interest for lifting machinery in the high-definition close-up image, calculate the image proportion of each area of ​​interest for lifting machinery in the high-definition close-up image, and retain the areas of interest for lifting machinery whose image proportion is greater than a second preset proportion value; The crane machinery hazard identification model is used to identify the retained crane machinery areas of interest, thereby obtaining the type of crane machinery and the surrounding scene within the preset range of the crane machinery. The system acquires security devices located within the surrounding environment and determines whether the placement and / or type of the security devices conform to preset business rules. If not, crane risk information is generated based on the type of crane machinery, the placement location of the safety equipment, and the equipment type.

6. The construction site inspection method based on a high-position camera according to claim 3, characterized in that, The process of acquiring various hazard identification models corresponding to various feature factors contained in the high-definition close-up image, and using the acquired hazard identification models to identify the corresponding areas of interest in order to obtain the corresponding risk information, includes: Identify the areas of interest for construction machinery in the high-definition close-up image, calculate the image proportion of each area of ​​interest for construction machinery in the high-definition close-up image, and retain the areas of interest for construction machinery whose image proportion is greater than a third preset proportion value; The construction machinery hazard identification model is used to identify the areas of concern for the retained construction machinery, thereby obtaining the type of construction machinery and personnel characteristics within the preset range of the construction machinery; If the personnel characteristic information conforms to the preset business rules, it is determined that the construction equipment is in use, and construction equipment risk information is generated based on the type of construction equipment, the personnel characteristic information, and the usage status.

7. The construction site inspection method based on a high-position camera according to any one of claims 4 to 6, characterized in that, The risk inspection list generated based on the risk information of each main grid image includes: For a main grid image, a personnel risk report is generated based on the personnel risk information of the human body attention area; wherein, the personnel risk report includes: human image information, risk type, high-definition close-up image of the risk type, location of the risk type, and time when the risk type was detected; The project construction stage is determined based on the type of lifting machinery, the type of construction equipment, and the surrounding environment. A machinery risk report is then generated based on the risk information of the lifting machinery and the construction equipment. The machinery risk report includes: the project construction stage, the lifting machinery and construction equipment present in the surrounding environment, and the existing risk information. Based on the personnel risk report and the machinery risk report, comprehensive risk information of the main grid image is generated, and the comprehensive risk information of all main grid images is integrated to form the risk inspection list.

8. A construction site inspection device based on a high-position camera, characterized in that, The device includes: The acquisition module is used to acquire target images captured by a high-position camera in a controlled inspection state; wherein the high-position camera is located at a preset high point of the construction site; The detection module is used to divide the target image into multiple grid images, and use a multimodal large model to sequentially detect whether there is at least one preset feature factor in each grid image, so as to obtain a main grid image containing the feature factor; The identification module is used to acquire a hazard identification model corresponding to the feature factors contained in the main grid image, and to use the acquired hazard identification model to identify the main grid image to obtain the risk information present in the main grid image; The generation module is used to generate a risk inspection list based on the risk information of each master grid image.

9. A computer device, the computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor, when executing the computer program, implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images