Construction site safety hazard identification method, device and equipment based on artificial intelligence

Abstract

The application discloses a construction site safety hidden danger identification method, device and equipment based on artificial intelligence, which is characterized in that a current construction site image set is first acquired; a first type of site image corresponding to a no-entry area is acquired, a first safety hidden danger identification result is acquired based on a first target detection model; a second type of site image corresponding to a construction material accumulation area is acquired, a second safety hidden danger identification result is acquired based on a second target detection model and an accumulation dumping detection strategy; and a current construction site safety hidden danger identification result is determined according to the above two identification results. The embodiment of the application can use different detection models to detect whether there is a person or equipment entering the no-entry area in the construction site and whether there is a dumping risk in the construction material accumulation area, thereby not only reducing the false detection of the no-entry area, but also improving the static feature extraction capability of the construction material to more accurately determine the dumping risk, and meanwhile, the detection accuracy of both is taken into account to output more accurate results.

Description

Technical Field

[0001] This invention relates to the field of visual recognition technology in artificial intelligence, and in particular to a method, apparatus and equipment for identifying safety hazards at construction sites based on artificial intelligence. Background Technology

[0002] In the construction industry, smart construction site safety monitoring technology has been widely applied. For example, by deploying multiple image acquisition devices (such as cameras) on the construction site, and utilizing object detection algorithms (such as the YOLO series models; YOLO stands for You Only Look Once, a deep learning-based object detection model) to continuously monitor the construction site. Currently, smart construction site safety monitoring technology is widely used for personnel compliance checks (such as whether safety helmets, reflective vests, and safety belts are worn) and area intrusion hazard alarms (such as personnel crossing the tower crane's slewing radius or restricted areas).

[0003] However, existing smart construction site visual recognition systems typically employ a single, general-purpose object detection network (such as a general-purpose YOLO network) to comprehensively identify all elements on the construction site (people, vehicles, materials, boundaries, etc.). However, the differences between crossing restricted areas and material dumping on a construction site are significant in feature space. A single, general-purpose object detection network is highly susceptible to feature confusion during training and inference, resulting in a high false alarm rate for people in restricted areas, while its feature extraction capabilities for static materials are severely insufficient, failing to balance detection accuracy for both. Summary of the Invention

[0004] This invention provides a method, apparatus, and equipment for identifying safety hazards at construction sites based on artificial intelligence. It aims to solve the problem that the existing technology uses a single general target detection network to detect all elements at the construction site, resulting in a high false alarm rate for personnel in restricted areas, while the feature extraction capability for static materials is seriously insufficient, making it impossible to balance the detection accuracy of both.

[0005] In a first aspect, embodiments of the present invention provide a method for identifying safety hazards at construction sites based on artificial intelligence, comprising: In response to the on-site safety hazard identification command sent by the user terminal, the system acquires the current on-site image set of the construction site; The system acquires a first type of site image corresponding to a restricted area from the current construction site image set, obtains a first target detection result corresponding to the first type of site image based on a pre-trained first target detection model, and determines a first safety hazard identification result based on the first target detection result; wherein, the first target detection model is used to determine whether there are personnel or equipment in the first type of site image; The system acquires a second type of site image corresponding to the area where construction materials are piled up, based on a pre-trained second target detection model and a preset dumping detection strategy. It then obtains a second target detection result corresponding to the second type of site image and determines a second safety hazard identification result based on the second target detection result. The dumping detection strategy is used to determine the dumping risk level based on whether the maximum stacking angle relative to the ground exceeds a safe angle of repose threshold, or to determine the dumping risk level based on whether the maximum aspect ratio of the second target detection result exceeds an aspect ratio threshold. The current site safety hazard identification result is determined based on the first safety hazard identification result and the second safety hazard identification result, and the current site safety hazard identification result is sent to the user terminal.

[0006] Secondly, embodiments of the present invention also provide an artificial intelligence-based construction site safety hazard identification device, which includes: The on-site image set acquisition unit is used to acquire the current on-site image set in response to the on-site safety hazard identification command sent by the user terminal. The first type of hazard identification unit is used to acquire first type site images corresponding to the restricted area in the current construction site image set, acquire first target detection results corresponding to the first type of site images based on a pre-trained first target detection model, and determine first safety hazard identification results based on the first target detection results; wherein, the first target detection model is used to determine whether there are personnel or equipment in the first type of site images; The second type of hazard identification unit is used to acquire second type site images corresponding to the construction material accumulation area in the current construction site image set, acquire second target detection results corresponding to the second type of site images based on a pre-trained second target detection model and a preset material dumping detection strategy, and determine the second safety hazard identification result based on the second target detection results; wherein, the material dumping detection strategy is used to determine the material dumping risk level based on whether the maximum stacking angle relative to the ground of the second target detection result corresponding to the second type of site image exceeds the safe angle of repose threshold, or to determine the material dumping risk level based on whether the maximum aspect ratio of the second target detection result corresponding to the second type of site image exceeds the aspect ratio threshold; The safety hazard identification result determination unit is used to determine the current construction site safety hazard identification result based on the first safety hazard identification result and the second safety hazard identification result, and send the current construction site safety hazard identification result to the user terminal.

[0007] Thirdly, embodiments of the present invention also provide a computer device, which includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method described in the first aspect above.

[0008] Fourthly, embodiments of the present invention also provide a computer-readable storage medium storing a computer program, the computer program including program instructions that, when executed by a processor, can implement the method described in the first aspect above.

[0009] This invention provides a method, apparatus, and equipment for identifying safety hazards at construction sites based on artificial intelligence. The method includes: responding to a site safety hazard identification command sent by a user terminal, acquiring a current set of construction site images; acquiring a first type of site image corresponding to a restricted area in the current set of construction site images, acquiring a first target detection result corresponding to the first type of site image based on a pre-trained first target detection model, and determining a first safety hazard identification result based on the first target detection result; acquiring a second type of site image corresponding to a construction material accumulation area in the current set of construction site images, acquiring a second target detection result corresponding to the second type of site image based on a pre-trained second target detection model and a preset accumulation dumping detection strategy, and determining a second safety hazard identification result based on the second target detection result; determining the current site safety hazard identification result based on the first and second safety hazard identification results, and sending the current site safety hazard identification result to the user terminal. The embodiments of the present invention can use different target detection models to detect whether people or equipment have entered the restricted areas of the construction site and whether there is a risk of dumping in the areas where building materials are piled up. This not only reduces the false detection of people or equipment in the restricted areas, but also improves the ability to extract static features of statically piled building materials to more accurately determine the risk of dumping. At the same time, it takes into account the detection accuracy of both to output more accurate results for identifying construction site safety hazards. Attached Figure Description

[0010] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 A schematic diagram illustrating an application scenario of the artificial intelligence-based construction site safety hazard identification method provided in this embodiment of the invention; Figure 2A flowchart illustrating the artificial intelligence-based method for identifying safety hazards at construction sites provided in this embodiment of the invention; Figure 3 A schematic diagram of the first sub-process of the construction site safety hazard identification method based on artificial intelligence provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the second sub-process of the construction site safety hazard identification method based on artificial intelligence provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of the third sub-process of the artificial intelligence-based construction site safety hazard identification method provided in an embodiment of the present invention; Figure 6 This is a schematic diagram of the fourth sub-process of the artificial intelligence-based construction site safety hazard identification method provided in an embodiment of the present invention; Figure 7 This is a schematic diagram of the fifth sub-process of the artificial intelligence-based construction site safety hazard identification method provided in an embodiment of the present invention; Figure 8 A schematic diagram of the sixth sub-process of the artificial intelligence-based construction site safety hazard identification method provided in this embodiment of the invention; Figure 9 A schematic block diagram of an artificial intelligence-based construction site safety hazard identification device provided in an embodiment of the present invention; Figure 10 A schematic block diagram of a computer device provided for an embodiment of the present invention. Detailed Implementation

[0012] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0013] It should be understood that, when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0014] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0015] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0016] Please also refer to Figure 1 and Figure 2 ,in Figure 1 This is a schematic diagram illustrating a scenario of the construction site safety hazard identification method based on artificial intelligence, according to an embodiment of the present invention. Figure 2 This is a flowchart illustrating the artificial intelligence-based method for identifying safety hazards at construction sites provided in an embodiment of the present invention. Figure 1 As shown, the artificial intelligence-based construction site safety hazard identification method provided in this embodiment of the invention is applied to server 10, and user terminal 20 is communicatively connected to server 10, wherein user terminal 20 is a smart terminal used by the user (such as a smartphone, tablet computer, etc.). Figure 2 As shown, the method includes the following steps S110 to S140.

[0017] S110: In response to the on-site safety hazard identification command sent by the user terminal, obtain the current on-site image set of the construction site.

[0018] In this embodiment, the technical solution is described with the server as the executing entity. Image acquisition devices are set up at multiple important monitoring locations on the construction site to be monitored. Each image acquisition device has a unique device number and a location number within the construction site, and all of these image acquisition devices are communicatively connected to the server. For example, the construction site to be monitored includes the main construction area (specifically including the main building construction area, foundation pit excavation area, earthwork operation area, etc.), material storage and processing area (specifically including building material storage area, component processing area, etc.), earthwork and slag-related areas (specifically including temporary slag storage area, vehicle washing and entrance / exit area, etc.), and auxiliary living and supporting areas (specifically including construction living area, machinery parking area, etc.). In the main construction area, the operating radius of large machinery, such as the slewing radius of tower cranes, the restricted area for excavator operations, and below the lifting and hoisting operation area, is high-risk for mechanical collisions. Personnel and equipment are prohibited from being stationed or entering these areas. Image acquisition devices aimed at these areas are required for on-site image acquisition. In the material storage and processing area, materials required for construction are piled up. Improper stacking may also pose a risk of tipping over. Image acquisition devices aimed at these areas are also required for on-site image acquisition.

[0019] In a more specific example, image acquisition device 1 (item1 with a unique device number and location number pos1) is set up targeting the slewing radius of the tower crane; image acquisition device 2 (item2 with a unique device number and location number pos2) is set up targeting the steel bar stacking area of ​​the building material stacking area; and image acquisition device 3 (item3 with a unique device number and location number pos3) is set up targeting the sand and gravel stacking area of ​​the building material stacking area. All of these image acquisition devices acquire on-site video or images of their respective targeted areas at equal and preset image acquisition cycles and upload them to the server. Alternatively, the user terminal can send an on-site image acquisition request to the server at any time, and the server can forward the request to each image acquisition device for immediate image acquisition before uploading to the server. For example, at a certain time T1, the user terminal sends a site safety hazard identification command to the server. At this time, the server forwards the on-site image acquisition request corresponding to the site safety hazard identification command to each image acquisition device for immediate image acquisition before uploading to the server, resulting in the current construction site image set corresponding to time T1. Then, image analysis can be performed on the current construction site image set on the server.

[0020] S120. Obtain a first type of site image corresponding to the restricted area in the current construction site image set, obtain a first target detection result corresponding to the first type of site image based on the pre-trained first target detection model, and determine a first safety hazard identification result based on the first target detection result.

[0021] In this embodiment, the first target detection model is used to determine whether there are personnel or equipment in the first type of on-site image. Since the location and unique device number of each image acquisition device connected to the server are initially configured and known, the server can quickly determine whether the on-site image type corresponds to a restricted area or an area with accumulated construction materials based on the location of each image acquisition device.

[0022] After obtaining the first type of site image corresponding to the restricted area in the current construction site image set in the server according to the above method, taking the processing process of one of the current construction site images corresponding to the restricted area as an example, the server obtains the first target detection result corresponding to the first type of site image based on the pre-trained first target detection model (which focuses on detecting whether there are people or non-tower crane equipment in the restricted area), and determines the first safety hazard identification result based on the first target detection result.

[0023] In one embodiment, such as Figure 3 As shown, step S120 includes: S121. Obtain the first image feature corresponding to the first type of on-site image; S122. Input the first image features into the first target detection model to obtain the first target detection result, which includes at least the first target object distribution bounding box range, the first detection object type, and the first confidence level.

[0024] In this embodiment, the first target detection model can specifically adopt the YOLOv8 model, which mainly includes a backbone layer (mainly using the C2f-CSP Darknet backbone, and including 4 levels of downsampling C2f modules, wherein the shallow layer of each downsampling C2f module is used to extract the details of personnel limbs and equipment contours, and the deep layer is used to extract the global semantics of vehicles and human bodies, and the embedded SPPF pooling layer is used to expand the receptive field), a neck feature fusion layer (mainly using the PAN-FPN bidirectional feature fusion layer), and a head detection head (mainly using the Anchor-Free decoupled detection head). The backbone layer of the first target detection model can extract the first image features corresponding to the first type of on-site image, and then input the first image features into the neck feature fusion layer and the head detection head in the first target detection model, so as to output the first target detection result, which includes at least the first target object distribution bounding box range, the first detection object type, and the first confidence level.

[0025] For example, if no one and no equipment are detected in the first type of scene image, then both the first target object distribution border range and the first detection object type are null values. If someone and equipment are detected in the first type of scene image, then both the first target object distribution border range and the first detection object type are not null values, and the specific values ​​of the first target object distribution border range and the first detection object type are set according to the actual detected target object distribution border range and object type. If the first target detection result is obtained, and the first detection object type is determined to be either a construction worker type or a construction equipment type, that is, there are personnel or other equipment besides the tower crane entering, then the corresponding first safety hazard identification result is that there is a first type of safety hazard and the corresponding first detection object type is either a construction worker type or a construction equipment type. If the first target detection result is obtained, and the first detection object type is determined to be either a construction worker type or a construction equipment type, that is, there are no personnel or other equipment besides the tower crane entering the restricted area, then the corresponding first safety hazard identification result is that there is no first type of safety hazard and the corresponding first detection object type is null. As can be seen, the first target detection model described above can focus more on detecting whether there are people or non-tower crane equipment entering the restricted area, thus obtaining more accurate detection results.

[0026] S130. Obtain a second type of site image corresponding to the construction material accumulation area in the current construction site image set. Based on the pre-trained second target detection model and the preset accumulation dumping detection strategy, obtain the second target detection result corresponding to the second type of site image, and determine the second safety hazard identification result based on the second target detection result.

[0027] In this embodiment, the material dumping detection strategy is used to determine the material dumping risk level based on whether the maximum stacking angle relative to the ground of the second target detection result corresponding to the second type of site image exceeds the safe angle of repose threshold, or it is used to determine the material dumping risk level based on whether the maximum aspect ratio of the second target detection result corresponding to the second type of site image exceeds the aspect ratio threshold. After obtaining the second type of site image corresponding to the construction material accumulation area in the current construction site image set in the server according to the above method, taking the processing process of one of the current construction site images corresponding to the construction material accumulation area as an example, the server obtains the second target detection result corresponding to the second type of site image based on the pre-trained second target detection model (which focuses on detecting whether there is a risk of dumping in the construction material accumulation state in the construction material accumulation area), and determines the second safety hazard identification result based on the second target detection result.

[0028] In one embodiment, such as Figure 4 As shown, step S130 includes: S131. Obtain the second image feature corresponding to the second type of on-site image; S132. Input the second image features into the second target detection model to obtain a second initial target detection result that includes at least the second target object distribution bounding range, the second detection object type, and the second confidence level. S133. Based on the distribution border range of the second target object in the second initial target detection result and the dumping detection strategy of the accumulated material, determine the corresponding dumping risk level of the accumulated material as the second target detection result.

[0029] In this embodiment, the second target detection model can specifically adopt the Mask R-CNN model, which mainly includes a backbone network (specifically including a ResNet residual network and a Feature Pyramid Network (FPN), used to extract multi-scale and high-level semantic features from the second type of scene image and use them as the second image features. Specifically, the convolutional layers and downsampling layers of the ResNet first extract the low-level edge, texture features and high-level semantic features of the second type of scene image, and then input these output features into the Feature Pyramid Network (FPN) to fuse the high-level strong semantic features with the low-level strong spatial location features, generating feature layers with different scales such as P2, P3, P4, P5 and P6), a region generation network (i.e., the RPN network, used to filter out the regions of interest containing construction materials from the second image features output by the backbone network), a region feature alignment layer (used to achieve pixel-level absolute alignment between the corresponding filtered regions of interest in the second image features and the actual spatial location of the construction materials in the second type of scene image), and a multi-task prediction head (i.e., Multi-task). The head is used to determine the second detection object type of construction materials and its corresponding second confidence level by the classification branch in the multi-task prediction head after the alignment of the region of interest. The regression branch in the multi-task prediction head determines the distribution bounding area of ​​the second target object of the construction materials, and the mask branch in the multi-task prediction head determines the precise geometric mask corresponding to the region of interest. The backbone network of the second target detection model can extract the second image features corresponding to the second type of on-site image. These second image features are then input into the region generation network, region feature alignment layer, and multi-task prediction head in the second target detection model for sequential processing. This outputs the second target detection result, which includes at least the distribution bounding area of ​​the second target object, the second detection object type, and the second confidence level.

[0030] In one embodiment, as a first embodiment of step S133, such as Figure 5 As shown, step S133 includes: S13311. Obtain multiple side contour lines in the distribution border range of the second target object, calculate and obtain the stacking angle between the fitted line corresponding to each side contour line and the ground, and select the maximum stacking angle between the line and the ground as the current target stacking angle. S13312. Determine the corresponding dynamic safe angle of repose threshold according to the second detection object type and the preset material angle of repose dynamic mapping table; S13313. If it is determined that the current target stacking angle is greater than the dynamic safety angle of repose threshold, then the high dumping risk level is used as the dumping risk level of the stack, and is used as the detection result of the second target; S13314. If it is determined that the current target stacking angle is less than or equal to the dynamic safety angle of repose threshold, then the low dumping risk level is used as the dumping risk level of the stack, and is used as the detection result of the second target.

[0031] In this embodiment, as the first embodiment of S133, when determining the risk level of pile dumping corresponding to the distribution border range of the second target object based on the pile dumping detection strategy, multiple side contour lines in the distribution border range of the second target object (which may initially consist of multiple discrete pixels) can be screened first. Then, after obtaining the corresponding fitted lines through polynomial fitting, the angle between these fitted lines and the horizontal plane corresponding to the ground is calculated, and the maximum stacking angle with the ground is selected from the above multiple angles as the current target stacking angle.

[0032] Next, it is necessary to determine the corresponding dynamic safe angle of repose threshold by combining the specific construction material type (such as steel bars, sand, etc.) corresponding to the second detection object type and the preset dynamic mapping table of material angle of repose, to avoid using the same fixed safe angle of repose threshold for all types of construction materials. After obtaining the corresponding dynamic safe angle of repose threshold by combining the material type, the relationship between the current target stacking angle and the dynamic safe angle of repose threshold is determined. Specifically, if it is determined that the current target stacking angle is greater than the dynamic safe angle of repose threshold, then the high tipping risk level is used as the tipping risk level of the pile, and it is used as the second target detection result; if it is determined that the current target stacking angle is less than or equal to the dynamic safe angle of repose threshold, then the low tipping risk level is used as the tipping risk level of the pile, and it is used as the second target detection result. It can be seen that by analyzing the stacking angle corresponding to the construction materials within the distribution border of the second target object, it is possible to determine whether there is a tipping risk.

[0033] Of course, once the server acquires multiple consecutive second-type on-site images of the same location according to the image acquisition cycle, it can obtain a time sequence of the current target stacking angle at the same construction material stacking location. By continuously monitoring the change rate of the angle value in this time sequence, if the change rate suddenly changes to more than three times the previous change rate, even if the current target stacking angle is not greater than the dynamic safety angle of repose threshold, a prompt message including the unique device number of the corresponding image acquisition device, the location number, the tipping risk warning information, and the second-type on-site image should be sent to the user terminal in a timely manner to prompt the user to come to the site for inspection and confirmation and to rearrange the construction materials to avoid the risk of tipping over.

[0034] In one embodiment, such as Figure 6 As shown, step S13312 includes: S133121. Obtain the reference angle of repose corresponding to the second detection object type from the material angle of repose dynamic mapping table; S133122. Obtain the current site environmental humidity parameters corresponding to the site location of the second type of site image, and input them into the preset humidity attenuation function to obtain the angle of repose attenuation compensation value for the second detection object type. S133123. Subtract the angle of repose attenuation compensation value from the reference angle of repose to obtain the dynamic safe angle of repose threshold.

[0035] In this embodiment, a dynamic mapping table of material angles of repose can be pre-set in the server. This dynamic mapping table sets different reference angles of repose for different construction materials. The reference angle of repose corresponding to the second detection object type (which reflects the specific material type of the construction material, such as steel bars, sand, and gravel) can be obtained by looking up the table. Then, if each image acquisition device is also equipped with a temperature and humidity sensor, the current site environmental humidity parameter of the site location corresponding to the second type of site image can also be obtained and input into the humidity attenuation function to obtain the angle of repose attenuation compensation value for the second detection object type. For example, the formula corresponding to the humidity attenuation function is δθ=α×TH1+β×T1, where δθ represents the angle of repose attenuation compensation value, α represents the humidity sensitivity factor for the second detection object type, β represents the temperature sensitivity factor for the second detection object type, TH1 represents the relative humidity parameter (its value is 0~100%) in the current site environmental humidity parameters, and T1 represents the current temperature parameter in the current site environmental humidity parameters. After determining the angle of repose attenuation compensation value corresponding to the second detection object type, the dynamic safe angle of repose threshold is obtained by subtracting the angle of repose attenuation compensation value from the reference angle of repose. This results in a more accurate dynamic safe angle of repose threshold that fully considers multiple factors such as the specific material type at the construction material storage site and the influence of the current site humidity parameters on the angle of repose.

[0036] In one embodiment, as a second embodiment of step S133, such as Figure 7 As shown, step S133 includes: S13321. Obtain the maximum height and maximum width within the distribution border range of the second target object; S13322: Obtain the current ratio between the maximum height and the maximum width; S13323. If it is determined that the current ratio is greater than the preset aspect ratio threshold, then the high dumping risk level is used as the dumping risk level of the accumulation, and is used as the detection result of the second target. S13324. If it is determined that the current ratio is less than or equal to the aspect ratio threshold, then the low dumping risk level is used as the dumping risk level of the accumulation, and is used as the detection result of the second target.

[0037] In this embodiment, as a second embodiment of S133, when determining the dumping risk level of the accumulated material corresponding to the distribution border range of the second target object based on the dumping detection strategy, the maximum height and maximum width within the distribution border range of the second target object can be obtained first. Then, the current ratio between the maximum height and the maximum width can be calculated, and finally, the relationship between the current ratio and a preset aspect ratio threshold can be compared. Specifically, if the current ratio is determined to be greater than the preset aspect ratio threshold, then a high dumping risk level is used as the dumping risk level of the accumulated material and as the detection result of the second target; if the current ratio is determined to be less than or equal to the aspect ratio threshold, then a low dumping risk level is used as the dumping risk level of the accumulated material and as the detection result of the second target. It can be seen that the dumping risk level of the accumulated material can be quickly determined by analyzing the aspect ratio within the distribution border range of the second target object.

[0038] S140. Determine the current site safety hazard identification result based on the first safety hazard identification result and the second safety hazard identification result, and send the current site safety hazard identification result to the user terminal.

[0039] In this embodiment, after obtaining the first safety hazard identification result and the second safety hazard identification result from the server, the current construction site safety hazard identification result is determined by comprehensively considering the two results, thereby realizing the detection of multi-dimensional safety hazards at the construction site.

[0040] In one embodiment, such as Figure 8 As shown, step S140 includes: S141. If it is determined that the first detection object type corresponding to the first safety hazard identification result is the construction worker type or the construction equipment type, or if it is determined that the second safety hazard identification result corresponds to the high dumping risk level, then the existence of safety hazards at the construction site shall be taken as the current site safety hazard identification result. S142. If it is determined that the first detection object type corresponding to the first safety hazard identification result is a null value, and it is determined that the second safety hazard identification result corresponds to a low dumping risk level, then the absence of safety hazards at the construction site shall be taken as the current construction site safety hazard identification result.

[0041] In this embodiment, if it is determined that the first detection object type corresponding to the first safety hazard identification result is a construction worker type or a construction equipment type, or if it is determined that the second safety hazard identification result corresponds to a high tipping risk level, it indicates that a situation has occurred in which a person or other equipment other than tower crane equipment has entered the restricted area, or that a situation has occurred in which the construction materials piled up in the construction material accumulation area have a high tipping risk. In this case, the existence of safety hazards at the construction site is taken as the current site safety hazard identification result.

[0042] If it is determined that the first detection object type corresponding to the first safety hazard identification result is a null value, and it is determined that the second safety hazard identification result corresponds to a low tipping risk level, it means that neither people nor other equipment other than tower crane equipment have entered the restricted area, nor have the construction materials piled up in the construction material accumulation area had a high tipping risk. In this case, the absence of safety hazards at the construction site is taken as the current site safety hazard identification result.

[0043] As can be seen, the implementation of this method uses different target detection models to detect whether people or equipment have entered the restricted areas of the construction site and whether there is a risk of dumping in the areas where building materials are piled up. This not only reduces the false detection of people or equipment in the restricted areas, but also improves the ability to extract static features of statically piled building materials to more accurately determine the risk of dumping. At the same time, it takes into account the detection accuracy of both and outputs more accurate results for identifying construction site safety hazards.

[0044] Figure 9 This is a schematic block diagram of a construction site safety hazard identification device based on artificial intelligence, provided in an embodiment of the present invention. Figure 9 As shown, corresponding to the above-described AI-based construction site safety hazard identification method, the present invention also provides an AI-based construction site safety hazard identification device 100. This AI-based construction site safety hazard identification device 100 includes a unit for executing the above-described AI-based construction site safety hazard identification method. Please refer to... Figure 9 The construction site safety hazard identification device 100 based on artificial intelligence includes: a site image set acquisition unit 110, a first type of hazard identification unit 120, a second type of hazard identification unit 130, and a safety hazard identification result judgment unit 140.

[0045] The on-site image set acquisition unit 110 is used to acquire the current construction site image set in response to the on-site safety hazard identification command sent by the user terminal.

[0046] In this embodiment, the technical solution is described with the server as the executing entity. Image acquisition devices are set up at multiple important monitoring locations on the construction site to be monitored. Each image acquisition device has a unique device number and a location number within the construction site, and all of these image acquisition devices are communicatively connected to the server. For example, the construction site to be monitored includes the main construction area (specifically including the main building construction area, foundation pit excavation area, earthwork operation area, etc.), material storage and processing area (specifically including building material storage area, component processing area, etc.), earthwork and slag-related areas (specifically including temporary slag storage area, vehicle washing and entrance / exit area, etc.), and auxiliary living and supporting areas (specifically including construction living area, machinery parking area, etc.). In the main construction area, the operating radius of large machinery, such as the slewing radius of tower cranes, the restricted area for excavator operations, and below the lifting and hoisting operation area, is high-risk for mechanical collisions. Personnel and equipment are prohibited from being stationed or entering these areas. Image acquisition devices aimed at these areas are required for on-site image acquisition. In the material storage and processing area, materials required for construction are piled up. Improper stacking may also pose a risk of tipping over. Image acquisition devices aimed at these areas are also required for on-site image acquisition.

[0047] In a more specific example, image acquisition device 1 (item1 with a unique device number and location number pos1) is set up targeting the slewing radius of the tower crane; image acquisition device 2 (item2 with a unique device number and location number pos2) is set up targeting the steel bar stacking area of ​​the building material stacking area; and image acquisition device 3 (item3 with a unique device number and location number pos3) is set up targeting the sand and gravel stacking area of ​​the building material stacking area. All of these image acquisition devices acquire on-site video or images of their respective targeted areas at equal and preset image acquisition cycles and upload them to the server. Alternatively, the user terminal can send an on-site image acquisition request to the server at any time, and the server can forward the request to each image acquisition device for immediate image acquisition before uploading to the server. For example, at a certain time T1, the user terminal sends a site safety hazard identification command to the server. At this time, the server forwards the on-site image acquisition request corresponding to the site safety hazard identification command to each image acquisition device for immediate image acquisition before uploading to the server, resulting in the current construction site image set corresponding to time T1. Then, image analysis can be performed on the current construction site image set on the server.

[0048] The first type of hazard identification unit 120 is used to acquire a first type of site image corresponding to a restricted area in the current construction site image set, acquire a first target detection result corresponding to the first type of site image based on a pre-trained first target detection model, and determine a first safety hazard identification result based on the first target detection result.

[0049] In this embodiment, since the location and unique device number of each image acquisition device connected to the server are initially configured and known, the server can quickly determine whether the on-site image type corresponds to a restricted area or a construction material accumulation area based on the location of each image acquisition device.

[0050] After obtaining the first type of site image corresponding to the restricted area in the current construction site image set in the server according to the above method, taking the processing process of one of the current construction site images corresponding to the restricted area as an example, the server obtains the first target detection result corresponding to the first type of site image based on the pre-trained first target detection model (which focuses on detecting whether there are people or non-tower crane equipment in the restricted area), and determines the first safety hazard identification result based on the first target detection result.

[0051] In one embodiment, the first type of hazard identification unit 120 is specifically used for: Obtain the first image feature corresponding to the first type of on-site image; The first image features are input into the first target detection model to obtain the first target detection result, which includes at least the first target object distribution bounding box range, the first detection object type, and the first confidence level.

[0052] In this embodiment, the first target detection model can specifically adopt the YOLOv8 model, which mainly includes a backbone layer (mainly using the C2f-CSP Darknet backbone, and including 4 levels of downsampling C2f modules, wherein the shallow layer of each downsampling C2f module is used to extract the details of personnel limbs and equipment contours, and the deep layer is used to extract the global semantics of vehicles and human bodies, and the embedded SPPF pooling layer is used to expand the receptive field), a neck feature fusion layer (mainly using the PAN-FPN bidirectional feature fusion layer), and a head detection head (mainly using the Anchor-Free decoupled detection head). The backbone layer of the first target detection model can extract the first image features corresponding to the first type of on-site image, and then input the first image features into the neck feature fusion layer and the head detection head in the first target detection model, so as to output the first target detection result, which includes at least the first target object distribution bounding box range, the first detection object type, and the first confidence level.

[0053] For example, if no one and no equipment are detected in the first type of scene image, then both the first target object distribution border range and the first detection object type are null values. If someone and equipment are detected in the first type of scene image, then both the first target object distribution border range and the first detection object type are not null values, and the specific values ​​of the first target object distribution border range and the first detection object type are set according to the actual detected target object distribution border range and object type. If the first target detection result is obtained, and the first detection object type is determined to be either a construction worker type or a construction equipment type, that is, there are personnel or other equipment besides the tower crane entering, then the corresponding first safety hazard identification result is that there is a first type of safety hazard and the corresponding first detection object type is either a construction worker type or a construction equipment type. If the first target detection result is obtained, and the first detection object type is determined to be either a construction worker type or a construction equipment type, that is, there are no personnel or other equipment besides the tower crane entering the restricted area, then the corresponding first safety hazard identification result is that there is no first type of safety hazard and the corresponding first detection object type is null. As can be seen, the first target detection model described above can focus more on detecting whether there are people or non-tower crane equipment entering the restricted area, thus obtaining more accurate detection results.

[0054] The second type of hazard identification unit 130 is used to acquire a second type of site image corresponding to the area where construction materials are piled up in the current construction site image set, acquire a second target detection result corresponding to the second type of site image based on a pre-trained second target detection model and a preset pile dumping detection strategy, and determine a second safety hazard identification result based on the second target detection result.

[0055] In this embodiment, after the server obtains the second type of site image corresponding to the construction material accumulation area in the current construction site image set according to the above method, taking the processing process of one of the current construction site images corresponding to the construction material accumulation area as an example, the server obtains the second target detection result corresponding to the second type of site image based on the pre-trained second target detection model (which focuses on detecting whether there is a risk of dumping in the construction material accumulation area under the state of construction material accumulation), and determines the second safety hazard identification result based on the second target detection result.

[0056] In one embodiment, the second type of hazard identification unit 130 is specifically used for: Obtain the second image features corresponding to the second type of on-site image; The second image features are input into the second target detection model to obtain a second initial target detection result that includes at least the second target object distribution bounding range, the second detection object type, and the second confidence level. Based on the distribution border range of the second target object in the second initial target detection result and the dumping detection strategy, the corresponding dumping risk level of the dumped material is determined as the second target detection result.

[0057] In this embodiment, the second target detection model can specifically adopt the Mask R-CNN model, which mainly includes a backbone network (specifically including a ResNet residual network and a Feature Pyramid Network (FPN), used to extract multi-scale and high-level semantic features from the second type of scene image and use them as the second image features. Specifically, the convolutional layers and downsampling layers of the ResNet first extract the low-level edge, texture features and high-level semantic features of the second type of scene image, and then input these output features into the Feature Pyramid Network (FPN) to fuse the high-level strong semantic features with the low-level strong spatial location features, generating feature layers with different scales such as P2, P3, P4, P5 and P6), a region generation network (i.e., the RPN network, used to filter out the regions of interest containing construction materials from the second image features output by the backbone network), a region feature alignment layer (used to achieve pixel-level absolute alignment between the corresponding filtered regions of interest in the second image features and the actual spatial location of the construction materials in the second type of scene image), and a multi-task prediction head (i.e., Multi-task). The head is used to determine the second detection object type of construction materials and its corresponding second confidence level by the classification branch in the multi-task prediction head after the alignment of the region of interest. The regression branch in the multi-task prediction head determines the distribution bounding area of ​​the second target object of the construction materials, and the mask branch in the multi-task prediction head determines the precise geometric mask corresponding to the region of interest. The backbone network of the second target detection model can extract the second image features corresponding to the second type of on-site image. These second image features are then input into the region generation network, region feature alignment layer, and multi-task prediction head in the second target detection model for sequential processing. This outputs the second target detection result, which includes at least the distribution bounding area of ​​the second target object, the second detection object type, and the second confidence level.

[0058] In one embodiment, as a first embodiment of determining the corresponding dumping risk level of the debris based on the distribution border range of the second target object in the second initial target detection result and the debris dumping detection strategy, the method includes: Obtain multiple side contour lines within the distribution bounding box of the second target object, calculate the stacking angle between the fitted line corresponding to each side contour line and the ground, and select the maximum stacking angle with the ground as the current target stacking angle. The corresponding dynamic safe angle of repose threshold is determined based on the second detection object type and the preset material angle of repose dynamic mapping table; If it is determined that the current target stacking angle is greater than the dynamic safe angle of repose threshold, then the high dumping risk level is used as the dumping risk level of the stack, and is used as the detection result of the second target; If it is determined that the current target stacking angle is less than or equal to the dynamic safety angle of repose threshold, then the low dumping risk level is used as the dumping risk level of the stack, and is used as the detection result of the second target.

[0059] In this embodiment, when determining the risk level of pile dumping corresponding to the distribution border range of the second target object based on the pile dumping detection strategy, multiple side contour lines in the distribution border range of the second target object (which may initially consist of multiple discrete pixels) can be screened first. Then, after obtaining the corresponding fitted lines through polynomial fitting, the angle between these fitted lines and the horizontal plane corresponding to the ground is calculated, and the maximum stacking angle with the ground is selected from the above multiple angles as the current target stacking angle.

[0060] Next, it is necessary to determine the corresponding dynamic safe angle of repose threshold by combining the specific construction material type (such as steel bars, sand, etc.) corresponding to the second detection object type and the preset dynamic mapping table of material angle of repose, to avoid using the same fixed safe angle of repose threshold for all types of construction materials. After obtaining the corresponding dynamic safe angle of repose threshold by combining the material type, the relationship between the current target stacking angle and the dynamic safe angle of repose threshold is determined. Specifically, if it is determined that the current target stacking angle is greater than the dynamic safe angle of repose threshold, then the high tipping risk level is used as the tipping risk level of the pile, and it is used as the second target detection result; if it is determined that the current target stacking angle is less than or equal to the dynamic safe angle of repose threshold, then the low tipping risk level is used as the tipping risk level of the pile, and it is used as the second target detection result. It can be seen that by analyzing the stacking angle corresponding to the construction materials within the distribution border of the second target object, it is possible to determine whether there is a tipping risk.

[0061] Of course, once the server acquires multiple consecutive second-type on-site images of the same location according to the image acquisition cycle, it can obtain a time sequence of the current target stacking angle at the same construction material stacking location. By continuously monitoring the change rate of the angle value in this time sequence, if the change rate suddenly changes to more than three times the previous change rate, even if the current target stacking angle is not greater than the dynamic safety angle of repose threshold, a prompt message including the unique device number of the corresponding image acquisition device, the location number, the tipping risk warning information, and the second-type on-site image should be sent to the user terminal in a timely manner to prompt the user to come to the site for inspection and confirmation and to rearrange the construction materials to avoid the risk of tipping over.

[0062] In one embodiment, determining the corresponding dynamic safe angle of repose threshold based on the second detection object type and a preset dynamic mapping table of material angles of repose includes: The reference angle of repose corresponding to the second detection object type is obtained by querying the dynamic mapping table of the material angle of repose. Obtain the current site environmental humidity parameters corresponding to the site location of the second type of site image, and input them into the preset humidity attenuation function to obtain the angle of repose attenuation compensation value for the second detection object type; The dynamic safe angle of repose threshold is obtained by subtracting the angle of repose attenuation compensation value from the reference angle of repose.

[0063] In this embodiment, a dynamic mapping table of material angles of repose can be pre-set in the server. This dynamic mapping table sets different reference angles of repose for different construction materials. The reference angle of repose corresponding to the second detection object type (which reflects the specific material type of the construction material, such as steel bars, sand, and gravel) can be obtained by looking up the table. Then, if each image acquisition device is also equipped with a temperature and humidity sensor, the current site environmental humidity parameter of the site location corresponding to the second type of site image can also be obtained and input into the humidity attenuation function to obtain the angle of repose attenuation compensation value for the second detection object type. For example, the formula corresponding to the humidity attenuation function is δθ=α×TH1+β×T1, where δθ represents the angle of repose attenuation compensation value, α represents the humidity sensitivity factor for the second detection object type, β represents the temperature sensitivity factor for the second detection object type, TH1 represents the relative humidity parameter (its value is 0~100%) in the current site environmental humidity parameters, and T1 represents the current temperature parameter in the current site environmental humidity parameters. After determining the angle of repose attenuation compensation value corresponding to the second detection object type, the dynamic safe angle of repose threshold is obtained by subtracting the angle of repose attenuation compensation value from the reference angle of repose. This results in a more accurate dynamic safe angle of repose threshold that fully considers multiple factors such as the specific material type at the construction material storage site and the influence of the current site humidity parameters on the angle of repose.

[0064] In one embodiment, as a second embodiment of determining the corresponding dumping risk level of the debris based on the distribution border range of the second target object in the second initial target detection result and the debris dumping detection strategy, the method includes: Obtain the maximum height and maximum width within the bounding area of ​​the second target object; Obtain the current ratio between the maximum height and the maximum width; If it is determined that the current ratio is greater than the preset aspect ratio threshold, then the high dumping risk level is used as the dumping risk level of the accumulation, and is used as the detection result of the second target; If it is determined that the current ratio is less than or equal to the aspect ratio threshold, then the low dumping risk level is used as the dumping risk level of the accumulation, and is used as the detection result of the second target.

[0065] In this embodiment, when determining the dumping risk level of the accumulated material corresponding to the distribution border range of the second target object based on the dumping detection strategy, the maximum height and maximum width within the distribution border range of the second target object can be obtained first. Then, the current ratio between the maximum height and the maximum width can be calculated. Finally, the current ratio can be compared with a preset aspect ratio threshold. Specifically, if the current ratio is determined to be greater than the preset aspect ratio threshold, a high dumping risk level is used as the dumping risk level of the accumulated material and as the detection result of the second target object. If the current ratio is determined to be less than or equal to the aspect ratio threshold, a low dumping risk level is used as the dumping risk level of the accumulated material and as the detection result of the second target object. It can be seen that the dumping risk level can be quickly determined by analyzing the aspect ratio within the distribution border range of the second target object.

[0066] The safety hazard identification result determination unit 140 is used to determine the current construction site safety hazard identification result based on the first safety hazard identification result and the second safety hazard identification result, and send the current construction site safety hazard identification result to the user terminal.

[0067] In this embodiment, after obtaining the first safety hazard identification result and the second safety hazard identification result from the server, the current construction site safety hazard identification result is determined by comprehensively considering the two results, thereby realizing the detection of multi-dimensional safety hazards at the construction site.

[0068] In one embodiment, the safety hazard identification result determination unit 140 is specifically used for: If it is determined that the first detection object type corresponding to the first safety hazard identification result is a construction worker type or a construction equipment type, or if it is determined that the second safety hazard identification result corresponds to a high dumping risk level, then the existence of safety hazards at the construction site shall be taken as the current site safety hazard identification result. If it is determined that the first detection object type corresponding to the first safety hazard identification result is a null value, and it is determined that the second safety hazard identification result corresponds to a low dumping risk level, then the absence of safety hazards at the construction site is taken as the current site safety hazard identification result.

[0069] In this embodiment, if it is determined that the first detection object type corresponding to the first safety hazard identification result is a construction worker type or a construction equipment type, or if it is determined that the second safety hazard identification result corresponds to a high tipping risk level, it indicates that a situation has occurred in which a person or other equipment other than tower crane equipment has entered the restricted area, or that a situation has occurred in which the construction materials piled up in the construction material accumulation area have a high tipping risk. In this case, the existence of safety hazards at the construction site is taken as the current site safety hazard identification result.

[0070] If it is determined that the first detection object type corresponding to the first safety hazard identification result is a null value, and it is determined that the second safety hazard identification result corresponds to a low tipping risk level, it means that neither people nor other equipment other than tower crane equipment have entered the restricted area, nor have the construction materials piled up in the construction material accumulation area had a high tipping risk. In this case, the absence of safety hazards at the construction site is taken as the current site safety hazard identification result.

[0071] As can be seen, the embodiments implementing this device use different target detection models to detect whether people or equipment have entered the restricted areas of the construction site and whether there is a risk of dumping in the areas where building materials are piled up. This not only reduces false detection of people or equipment in the restricted areas, but also improves the ability to extract static features of statically piled building materials to more accurately determine the risk of dumping. At the same time, it takes into account the detection accuracy of both and outputs more accurate results for identifying construction site safety hazards.

[0072] The aforementioned AI-based construction site safety hazard identification device can be implemented as a computer program, which can, for example... Figure 10 It runs on the computer device shown.

[0073] Please see Figure 10 , Figure 10 This is a schematic block diagram of a computer device provided in an embodiment of the present invention. This computer device integrates any of the artificial intelligence-based construction site safety hazard identification devices provided in this embodiment of the present invention.

[0074] See Figure 10 The computer device 400 includes a processor 402, a memory, and a network interface 405 connected via a system bus 401. The memory may include a storage medium 403 and internal memory 404.

[0075] The storage medium 403 may store an operating system 4031 and a computer program 4032. The computer program 4032 includes program instructions that, when executed, cause the processor 402 to perform an artificial intelligence-based method for identifying safety hazards at construction sites.

[0076] The processor 402 is used to provide computing and control capabilities to support the operation of the entire computer device.

[0077] The internal memory 404 provides an environment for the computer program 4032 in the storage medium 403 to run. When the computer program 4032 is executed by the processor 402, the processor 402 can execute the above-mentioned artificial intelligence-based construction site safety hazard identification method.

[0078] This network interface 405 is used for network communication with other devices. Those skilled in the art will understand that... Figure 10 The structure shown is merely a block diagram of a portion of the structure related to the present invention and does not constitute a limitation on the computer device to which the present invention is applied. A specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0079] The processor 402 is used to run the computer program 4032 stored in the memory to implement the artificial intelligence-based construction site safety hazard identification method as described above.

[0080] It should be understood that, in this embodiment of the invention, the processor 402 may be a Central Processing Unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0081] It will be understood by those skilled in the art that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program includes program instructions and can be stored in a storage medium, which is a computer-readable storage medium. The program instructions are executed by at least one processor in the computer system to implement the process steps of the embodiments of the above methods.

[0082] Therefore, the present invention also provides a computer-readable storage medium. This computer-readable storage medium stores a computer program, wherein the computer program includes program instructions. When executed by a processor, the program instructions cause the processor to perform the artificial intelligence-based construction site safety hazard identification method described above.

[0083] The storage medium can be any computer-readable storage medium that can store program code, such as a USB flash drive, external hard drive, read-only memory (ROM), magnetic disk, or optical disk.

[0084] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0085] In the several embodiments provided by this invention, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For example, the division of each unit is merely a logical functional division, and there may be other division methods in actual implementation. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed.

[0086] The steps in the method of this invention can be adjusted, merged, or reduced in order according to actual needs. The units in the device of this invention can be merged, divided, or reduced according to actual needs. Furthermore, the functional units in the various embodiments of this invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0087] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, a terminal, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention.

[0088] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for identifying safety hazards at construction sites based on artificial intelligence, characterized in that, include: In response to the on-site safety hazard identification command sent by the user terminal, the system acquires the current on-site image set of the construction site; The system acquires a first type of site image corresponding to a restricted area from the current construction site image set, obtains a first target detection result corresponding to the first type of site image based on a pre-trained first target detection model, and determines a first safety hazard identification result based on the first target detection result; wherein, the first target detection model is used to determine whether there are personnel or equipment in the first type of site image; The system acquires a second type of site image corresponding to the area where construction materials are piled up, based on a pre-trained second target detection model and a preset dumping detection strategy. It then obtains a second target detection result corresponding to the second type of site image and determines a second safety hazard identification result based on the second target detection result. The dumping detection strategy is used to determine the dumping risk level based on whether the maximum stacking angle relative to the ground exceeds a safe angle of repose threshold, or to determine the dumping risk level based on whether the maximum aspect ratio of the second target detection result exceeds an aspect ratio threshold. Based on the first safety hazard identification result and the second safety hazard identification result, the current construction site safety hazard identification result is determined and sent to the user terminal; The second target detection model based on pre-training and the preset dumping detection strategy for accumulated materials obtain the second target detection result corresponding to the second type of on-site image, including: Obtain the second image features corresponding to the second type of on-site image; The second image features are input into the second target detection model to obtain a second initial target detection result that includes at least the second target object distribution bounding range, the second detection object type, and the second confidence level. Based on the distribution border range of the second target object in the second initial target detection result and the dumping detection strategy of the accumulated material, the corresponding dumping risk level of the accumulated material is determined as the second target detection result; The step of determining the corresponding dumping risk level of the accumulated material as the second target detection result based on the distribution border range of the second target object in the second initial target detection result and the dumping detection strategy includes: Obtain multiple side contour lines within the distribution bounding box of the second target object, calculate the stacking angle between the fitted line corresponding to each side contour line and the ground, and select the maximum stacking angle with the ground as the current target stacking angle. The corresponding dynamic safe angle of repose threshold is determined based on the second detection object type and the preset material angle of repose dynamic mapping table; If it is determined that the current target stacking angle is greater than the dynamic safe angle of repose threshold, then the high dumping risk level is used as the dumping risk level of the stack, and is used as the detection result of the second target; If it is determined that the current target stacking angle is less than or equal to the dynamic safety angle of repose threshold, then the low dumping risk level is used as the dumping risk level of the stack, and is used as the detection result of the second target; The step of determining the corresponding dynamic safe angle of repose threshold based on the second detection object type and the preset material angle of repose dynamic mapping table includes: The reference angle of repose corresponding to the second detection object type is obtained by querying the dynamic mapping table of the material angle of repose. Obtain the current site environmental humidity parameters corresponding to the site location of the second type of site image, and input them into the preset humidity attenuation function to obtain the angle of repose attenuation compensation value for the second detection object type; The dynamic safe angle of repose threshold is obtained by subtracting the angle of repose attenuation compensation value from the reference angle of repose.

2. The method according to claim 1, characterized in that, The first target detection model based on pre-training acquires the first target detection result corresponding to the first type of scene image, including: Obtain the first image feature corresponding to the first type of on-site image; The first image features are input into the first target detection model to obtain the first target detection result, which includes at least the first target object distribution bounding box range, the first detection object type, and the first confidence level.

3. The method according to claim 1, characterized in that, Based on the distribution boundary range of the second target object in the second initial target detection result and the dumping detection strategy, the corresponding dumping risk level is determined as the second target detection result, including: Obtain the maximum height and maximum width within the bounding area of ​​the second target object; Obtain the current ratio between the maximum height and the maximum width; If it is determined that the current ratio is greater than the preset aspect ratio threshold, then the high dumping risk level is used as the dumping risk level of the accumulation, and is used as the detection result of the second target; If it is determined that the current ratio is less than or equal to the aspect ratio threshold, then the low dumping risk level is used as the dumping risk level of the accumulation, and is used as the detection result of the second target.

4. The method according to claim 1, characterized in that, The step of determining the current site safety hazard identification result based on the first safety hazard identification result and the second safety hazard identification result includes: If it is determined that the first detection object type corresponding to the first safety hazard identification result is a construction worker type or a construction equipment type, or if it is determined that the second safety hazard identification result corresponds to a high dumping risk level, then the existence of safety hazards at the construction site shall be taken as the current site safety hazard identification result. If it is determined that the first detection object type corresponding to the first safety hazard identification result is a null value, and it is determined that the second safety hazard identification result corresponds to a low dumping risk level, then the absence of safety hazards at the construction site is taken as the current site safety hazard identification result.

5. A construction site safety hazard identification device based on artificial intelligence, characterized in that, include: The on-site image set acquisition unit is used to acquire the current on-site image set in response to the on-site safety hazard identification command sent by the user terminal. The first type of hazard identification unit is used to acquire first type site images corresponding to the restricted area in the current construction site image set, acquire first target detection results corresponding to the first type of site images based on a pre-trained first target detection model, and determine first safety hazard identification results based on the first target detection results; wherein, the first target detection model is used to determine whether there are personnel or equipment in the first type of site images; The second type of hazard identification unit is used to acquire second type site images corresponding to the construction material accumulation area in the current construction site image set, acquire second target detection results corresponding to the second type of site images based on a pre-trained second target detection model and a preset material dumping detection strategy, and determine the second safety hazard identification result based on the second target detection results; wherein, the material dumping detection strategy is used to determine the material dumping risk level based on whether the maximum stacking angle relative to the ground of the second target detection result corresponding to the second type of site image exceeds the safe angle of repose threshold, or to determine the material dumping risk level based on whether the maximum aspect ratio of the second target detection result corresponding to the second type of site image exceeds the aspect ratio threshold; The safety hazard identification result determination unit is used to determine the current construction site safety hazard identification result based on the first safety hazard identification result and the second safety hazard identification result, and send the current construction site safety hazard identification result to the user terminal; The second type of hazard identification unit is specifically used for: Obtain the second image features corresponding to the second type of on-site image; The second image features are input into the second target detection model to obtain a second initial target detection result that includes at least the second target object distribution bounding range, the second detection object type, and the second confidence level. Based on the distribution border range of the second target object in the second initial target detection result and the dumping detection strategy of the accumulated material, the corresponding dumping risk level of the accumulated material is determined as the second target detection result; The step of determining the corresponding dumping risk level of the accumulated material as the second target detection result based on the distribution border range of the second target object in the second initial target detection result and the dumping detection strategy includes: Obtain multiple side contour lines within the distribution bounding box of the second target object, calculate the stacking angle between the fitted line corresponding to each side contour line and the ground, and select the maximum stacking angle with the ground as the current target stacking angle. The corresponding dynamic safe angle of repose threshold is determined based on the second detection object type and the preset material angle of repose dynamic mapping table; If it is determined that the current target stacking angle is greater than the dynamic safe angle of repose threshold, then the high dumping risk level is used as the dumping risk level of the stack, and is used as the detection result of the second target; If it is determined that the current target stacking angle is less than or equal to the dynamic safety angle of repose threshold, then the low dumping risk level is used as the dumping risk level of the stack, and is used as the detection result of the second target; The step of determining the corresponding dynamic safe angle of repose threshold based on the second detection object type and the preset material angle of repose dynamic mapping table includes: The reference angle of repose corresponding to the second detection object type is obtained by querying the dynamic mapping table of the material angle of repose. Obtain the current site environmental humidity parameters corresponding to the site location of the second type of site image, and input them into the preset humidity attenuation function to obtain the angle of repose attenuation compensation value for the second detection object type; The dynamic safe angle of repose threshold is obtained by subtracting the angle of repose attenuation compensation value from the reference angle of repose.

6. A computer device, characterized in that, The computer device includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements the artificial intelligence-based construction site safety hazard identification method as described in any one of claims 1-4.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which includes program instructions that, when executed by a processor, can implement the artificial intelligence-based construction site safety hazard identification method as described in any one of claims 1-4.

Images