A method and device for detecting delayed and illegal installation of steel support

Abstract

The present application provides a method and device for detecting steel support erection delay violations, which relate to the field of image processing technology. The installation tray is detected from the monitoring image of the construction site through a target detection algorithm, and the possible existence range of the steel support structure is expanded. The edge detection algorithm is then used to perform edge detection, which can limit the area of ​​interest to a smaller range, avoid interference from useless information, and improve recognition speed and recognition accuracy. Furthermore, by comparing the installation tray detection results and the steel support structure detection results of the current frame and the historical frame, it is possible to automatically determine whether the steel support erection meets the construction requirements from the time dimension and / or space dimension, and the recognition is timely and accurate, which helps to improve construction safety.

Description

Technical Field

[0001] The present application relates to the field of image processing technology, and in particular to a method and device for detecting delayed and illegal installation of steel supports. Background Art

[0002] The excavation of subway engineering structures can be mainly divided into two types: open excavation and dark excavation. The open excavation method is a method of directly excavating a foundation pit from the ground to build facilities such as platforms. The open excavation method has the advantages of simple construction, fast, economical and safe. It was used as the preferred excavation technology in the early stages of the development of urban underground tunnel projects. After the open excavation method is used to excavate the foundation pit, the upper soil is excavated, which is equivalent to unloading the base and side soil, breaking the original load balance, causing stress release on the base and side soil, and causing the foundation and side soil to deform and bulge. Therefore, in the process of open excavation method foundation pit excavation, the earthwork needs to be excavated in layers, steps and symmetry, and the steel support structure should be erected in time to form a frame support system to avoid cracking, breaking, shearing or buckling of the enclosure body itself in the open excavation pit, causing the structure to lose its bearing capacity and collapse and damage, causing serious accidents.

[0003] At present, the existing technology uses manual inspection to check whether the steel support of the open excavation pit meets the erection requirements. That is, the personnel of the construction site supervision unit or the construction unit introduces a third-party inspector to inspect the on-site foundation pit excavation process. If any violation of the steel support erection is found, the construction unit will be notified in writing to immediately stop earthwork development and install steel support. However, this manual inspection method has a large lag and cannot detect violations of the steel support erection in time, thus posing a hidden danger to construction safety. Summary of the invention

[0004] In view of this, the purpose of this application is to provide a method and device for detecting delayed violations in steel support erection, so as to solve the problem in the prior art that manual inspections have a large lag and cannot promptly detect violations in steel support erection, so as to achieve the purpose of automatically and promptly detecting whether the steel support erection meets the construction requirements and whether there are any delayed violations, thereby ensuring construction safety.

[0005] The present application provides a method for detecting a steel support erection delay violation, the method comprising:

[0006] Continuously collect monitoring images of the construction site according to the predetermined collection frequency;

[0007] For the current frame monitoring image, a pre-trained target detection model is used to determine whether a new installation pallet is detected in the monitoring image, and a current installation pallet detection result is obtained; wherein each steel support structure is fixedly installed on the foundation pit support at the construction site through the installation pallets at both ends to support the foundation pit support;

[0008] If the current installation tray detection result is that a new installation tray is detected, then combining the morphological prior knowledge of the steel support structure, the possible existence range of the steel support structure is expanded from the location of the new installation tray in the monitoring image, and the image of interest is segmented from the monitoring image according to the possible existence range;

[0009] Use a pre-trained edge detection model to detect whether there is an edge of a steel support structure in the image of interest, and obtain a current steel support structure detection result;

[0010] By comparing the current installation pallet detection results and the current steel support structure detection results of the current frame monitoring image with the installation pallet detection results and the steel support structure detection results of the historical frame monitoring image, it is determined whether the steel support erection at the construction site meets the construction requirements; wherein the construction requirements include requirements for the erection time dimension and / or the erection space dimension.

[0011] The present application also provides a device for detecting a steel support erection delay violation, the device comprising:

[0012] A collection module, used to continuously collect monitoring images of the construction site according to a predetermined collection frequency;

[0013] The pallet detection module is used to determine whether a new installation pallet is detected in the monitoring image of the current frame using a pre-trained target detection model, and obtain the current installation pallet detection result; wherein each steel support structure is fixedly installed on the foundation pit support at the construction site through the installation pallets at both ends to support the foundation pit support;

[0014] an expansion module, for, if the current installation tray detection result is that a new installation tray is detected, combining the morphological prior knowledge of the steel support structure, expanding the possible existence range of the steel support structure from the location of the new installation tray in the monitoring image, and segmenting the image of interest from the monitoring image according to the possible existence range;

[0015] A steel support detection module is used to detect whether there is a steel support structure edge in the image of interest using a pre-trained edge detection model to obtain a current steel support structure detection result;

[0016] A comparison module is used to determine whether the steel support erection at the construction site meets the construction requirements by comparing the current installation pallet detection result and the current steel support structure detection result of the current frame monitoring image with the installation pallet detection result and the steel support structure detection result of the historical frame monitoring image; wherein the construction requirements include requirements for the erection time dimension and / or the erection space dimension.

[0017] An embodiment of the present application also provides an electronic device, including: a processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor and the memory communicate through the bus, and when the machine-readable instructions are executed by the processor, the steps of the method for detecting delayed violation of steel support erection are performed as described above.

[0018] An embodiment of the present application also provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the method for detecting delayed violation of steel support erection are executed as described above.

[0019] The embodiment of the present application provides a method and device for detecting delayed violation of steel support erection. The installation pallet is detected by a target detection algorithm and the possible existence range of the steel support structure is expanded. The edge detection is then performed by an edge detection algorithm. The region of interest can be limited to a smaller range, avoiding interference from useless information and improving recognition speed and accuracy. Furthermore, by comparing the installation pallet detection results and the steel support structure detection results of the current frame and the historical frame, it is possible to automatically determine whether the steel support erection meets the construction requirements from the time dimension and / or space dimension, determine whether the steel support erection has a delayed violation, and identify whether it is timely and accurate, which helps to improve construction safety.

[0020] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, preferred embodiments are specifically cited below and described in detail with reference to the attached drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0022] Figure 1 A flow chart showing a method for detecting a steel support erection delay violation provided in an embodiment of the present application is shown;

[0023] Figure 2 A schematic diagram of a construction site provided by an embodiment of the present application is shown;

[0024] Figure 3 A schematic diagram of the structure of a target detection model provided in an embodiment of the present application is shown;

[0025] Figure 4 A schematic diagram of the structure of an edge detection model provided in an embodiment of the present application is shown;

[0026] Figure 5 A schematic diagram of the structure of a detection device for delayed steel support erection violations provided in an embodiment of the present application is shown;

[0027] Figure 6 A schematic structural diagram of an electronic device provided in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0028] To make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all of the embodiments. The components of the embodiments of the present application usually described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the application claimed for protection, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, each other embodiment obtained by those skilled in the art without making creative work belongs to the scope of protection of the present application.

[0029] Steel support erection refers to the erection of support structures made of steel to support formwork, concrete or other building materials during the construction process to ensure construction safety and structural stability. Steel supports are used as horizontal supports after underground excavation to resist deformation of the foundation pit. It overcomes the shortcomings of concrete horizontal supports and has the advantages of being reusable and environmentally friendly. For example, for the excavation of subway engineering structures, earth excavation must be carried out in layers, steps and symmetry during the open-cut foundation pit excavation process, and steel support structures must be erected in a timely manner to form a framework support system to prevent the open-cut foundation pit from cracking, breaking, shearing or buckling due to the enclosure itself, causing the structure to lose its bearing capacity and collapse, resulting in serious accidents.

[0030] After research, it was found that, at present, the existing technology for determining whether the steel support of the open excavation pit meets the erection requirements uses manual inspection, that is, the construction site supervision unit personnel or the construction unit introduces third-party inspectors to inspect the on-site foundation pit excavation process. If any violation of the steel support erection is found, the construction unit will be notified in writing to immediately stop earthwork development and install steel supports. However, this manual inspection method has a large lag and cannot detect violations of the steel support erection in a timely manner, thus posing a hidden danger to construction safety.

[0031] Based on this, an embodiment of the present application provides a method for detecting delayed violations in steel support erection, so as to automatically and timely detect whether the steel support erection meets the construction requirements, whether there are delayed violations, and ensure construction safety.

[0032] See also Figure 1 , Figure 1 A flow chart of a method for detecting a steel support erection delay violation provided in an embodiment of the present application. Figure 1 As shown in, the detection method provided in the embodiment of the present application includes:

[0033] S101. Continuously collect monitoring images of the construction site according to a predetermined collection frequency.

[0034] Here, a fixed camera can be set up at the construction site to collect monitoring images frame by frame at a predetermined collection frequency, such as collecting a monitoring image frame every 10 minutes, 30 minutes, etc. Figure 2 , Figure 2 This is a schematic diagram of a construction site provided in an embodiment of the present application. Figure 2 , and detailed description of the implementation process of the detection method.

[0035] S102: for the current frame monitoring image, use a pre-trained target detection model to determine whether a new installation tray is detected in the monitoring image, and obtain a current installation tray detection result.

[0036] like Figure 2 As shown in , a plurality of steel support structures are set up in the construction site, and each steel support structure is fixedly installed on the foundation pit support of the construction site through mounting trays at both ends to support the foundation pit support, such as supporting it between two foundation pit support walls.

[0037] In this step, for the collected current frame monitoring image, the frame monitoring image can be input into a pre-trained target detection model to determine whether a new installation tray is detected in the monitoring image, and obtain the current installation tray detection result; Figure 2 As shown in , in the current installation tray detection result, the new installation tray can be marked in the monitoring image in the form of a rectangular frame.

[0038] In specific implementation, the target detection model can detect all installation pallets in the current frame monitoring image, and then compare them with all installation pallets with steel support structures that have been erected in the historical frame monitoring image to determine whether new installation pallets are detected; it can also use a mask to cover up the installation pallets with steel support structures that have been detected in the historical frame monitoring image, so as to directly detect whether there are new installation pallets in the current frame monitoring image.

[0039] In a possible implementation, the target detection model is built on the basis of the YoloV5 neural network model. Here, YoloV5 is selected instead of previous versions such as YoloV4 because, although YoloV5 is slightly weaker than YoloV4 in performance, it is much stronger than YoloV4 in flexibility and speed, and has a strong advantage in the rapid deployment of the model; the reason for not selecting other networks such as R-CNN is that Yolo, as a single-stage algorithm, only needs to perform image processing once to obtain the result, and is much faster than two-stage algorithms such as R-CNN. This algorithm will be used for construction site safety detection and safety warning based on monitoring, so the speed of the algorithm is particularly important.

[0040] See also Figure 3 , Figure 3 A schematic diagram of the structure of a target detection model provided in an embodiment of the present application. Figure 3 As shown in , the target detection model includes an input module, a BackBone backbone module, a Neck intermediate module and a Head prediction module.

[0041] Regarding the input module (not shown in the figure), during the model training stage, Mosaic data enhancement is used. Four pictures are randomly selected for random scaling, random cropping and random arrangement operations, and then the four pictures are spliced ​​into a large picture so that multiple targets can be contained in one image at the same time. Adaptive anchor frame calculation is used. During training, yolov5 outputs a predicted frame based on the initial anchor frame, and compares it with the groundtruth (correct answer, abbreviated as gt) and updates it in reverse. Adaptive image scaling is used to improve the algorithm for filling black borders at the edges of the picture, and the original image is adaptively added with the least black border (a small scaling factor is selected) to convert the image into 608*608*3 (size 608*608, 3 is the RGB color channel).

[0042] The BackBone backbone module includes a Focus structure connected in sequence, a combination of at least one DCNBL unit and a CSP unit, a DCNBL unit, an SPP structure and a GAM attention layer.

[0043] Among them, the key to the Focus structure is the slicing operation: the original 608*608*3 image is input into the Focus structure, and the slicing operation is used to first convert it into a 304*304*12 feature map, and then undergoes a convolution operation with 32 convolution kernels, and finally becomes a 304*304*32 feature map.

[0044] The full name of the CSP structure is Cross Stage Partial Network. Its design mainly solves the problem of high computational complexity during reasoning. CSPNet believes that the high amount of inference computation is mainly caused by the duplication of gradient information in network optimization. Therefore, the CSP module first divides the feature map of the base layer into two parts, and then merges them through a cross-stage hierarchy. This can reduce the amount of computation and ensure accuracy. In the backbone network of YOLOv5, the CSP1_X structure is used as part of the Backbone backbone network, and another CSP2_X structure is applied in the Neck intermediate module. The convolution kernel size before each CSP module is 3x3, and the stride is set to 2, so it has the function of downsampling.

[0045] Each DCNBL unit includes a deformable convolution layer, a batch normalization layer, and an activation function layer; the deformable convolution layer is used to extract features, the batch normalization layer BN is used to accelerate network convergence and improve stability, and the activation function Leaky ReLU enhances the network expression ability by introducing nonlinearity. Here, the embodiment of the present application notes that: some irregular or deformed target features are difficult to extract accurately, and traditional models may cause information loss or increase in errors when processing irregularly shaped targets, and cannot perform fine analysis of local details of the target. These problems will lead to a decrease in detection accuracy. In addition, traditional models have problems such as local receptive field, information loss, scale change, and lack of contextual information, making it difficult to capture the global features of the entire image, thereby limiting the global understanding and accurate detection capabilities of the target, that is, lack of effective modeling of global information. Therefore, in order to extract the desired information more accurately and comprehensively, the embodiment of the present application introduces variable convolution to replace the traditional CNN as the core operator, replaces the Conv2d convolution layer in the original CBL module with the DCNv3 variable convolution to form a DCNBL module, and replaces the CBL in the CSP1_X module with the DCNBL unit to form a new CSP1_X_DCN structure.

[0046] SPP structure, SPP is called Spatial Pyramid Pooling. SPP can generate fixed-size output by using maximum pooling of different sizes, which is not affected by the size of the input image. In YOLOv5, the SPP module usually contains multiple maximum pooling layers of different kernel sizes (such as 5x5, 9x9, 13x13), and then splices the results. SPP can significantly increase the receptive field and capture more contextual information while hardly increasing the amount of computation.

[0047] Since small target detection is easily affected by image background and noise, which hinders subsequent detection, the embodiment of the present application reduces this problem by adding an attention mechanism. Because the BackBone end is in the feature extraction stage, the model should focus on more important key features. In addition, since the deep semantic information that has undergone feature fusion is close to the head detection head. Therefore, a GAM attention layer is added after the SPPF layer of the backbone network, so that relatively important information can be filtered out from a large amount of information, the degree of attention to the underlying features is enhanced, and the degree of attention to small targets is strengthened, thereby improving the detection accuracy of the model.

[0048] For the Neck intermediate module, it includes FPN structure, PAN structure and CSP unit; FPN (Feature Pyramid Network) module, from top to bottom, transmits and fuses high-level feature information through upsampling to obtain feature maps for prediction, and upsamples 19*19 images to 76*76. PAN (Path Aggregation Network) module, a bottom-up feature pyramid, processes the strong semantic features transmitted from top to bottom by the FPN layer and transmits strong positioning features from bottom to top, aggregates parameters of different detection layers from different backbone layers, and downsamples 76*76 images to 19*19; Different from yolov4, the target detection model does not only have ordinary convolution operations, but borrows the CSP2_X structure designed by CSPnet to enhance the ability of network feature fusion.

[0049] For the Head prediction module, CIOU_Loss is used as the loss function, and the detection frame of the installation tray is post-processed through non-maximum suppression to obtain the current installation tray detection result. Among them, CIOU_Loss is used as the loss function of the Boundingbox, and the intersection scale measurement method is added; IOU defines the ratio of the intersection area of ​​the predicted box and the true box to the union area, that is:

[0050]

[0051] The formula for CIOU is:

[0052]

[0053] Among them, d represents the distance between the center point of the predicted box and the real box, and c represents the diagonal distance of the minimum enclosing rectangle. Represents a correction factor used to further adjust the loss function to take into account the shape and orientation of the target box.

[0054] The final CIoU Loss formula is:

[0055]

[0056] NMS non-maximum suppression is a post-processing technique used to eliminate redundant detection frames so that the originally occluded target can also be detected. The final output is equal to the original image size and the detection frame is marked. Its main steps are as follows: 1. Sort all detection frames by confidence; 2. Select the frame with the highest confidence; 3. Calculate the IoU between this frame and all other frames; 4. Remove frames with IoU higher than a certain threshold; 5. Repeat steps 2-4 until all frames are processed and finally get the current installation tray detection result.

[0057] S103. If the current installation pallet detection result is that a new installation pallet is detected, then combined with the morphological prior knowledge of the steel support structure, the possible existence range of the steel support structure is expanded from the location of the new installation pallet in the monitoring image, and the image of interest is cut out from the monitoring image according to the possible existence range.

[0058] Here, the present application has found through research that each steel support structure in the monitoring image has a different shape, and the method of directly calibrating the entire steel support structure and detecting it has a low accuracy rate. Therefore, the embodiment of the present application first detects the new installation tray in the monitoring image, and then expands the possible existence range of the steel support structure based on the location of the new installation tray in the current installation tray detection result, combined with the prior knowledge of the steel support structure's shape, to obtain a rough steel support outline.

[0059] In this step, the morphological prior knowledge includes the installation angle of the steel support structure in the monitoring image. Therefore, step S103 may include: in the monitoring image, starting from the location of the new installation tray, extending according to the installation angle, and expanding the possible existence range of the steel support structure. Figure 2 As shown in the figure, a straight line extending outward from the end mounting tray at a certain angle can be used to detect one side of the entire steel support structure, that is, Figure 2 The red line part in the middle. Afterwards, the possible existence range of the entire steel support structure can be expanded based on one edge of the entire steel support structure; or, another point can be selected from the location of the end mounting tray and another straight line can be extended outward at a certain angle, and then combined with the image boundary to enclose the possible existence range of the entire steel support structure; in addition, the length of the extended straight line can be determined based on the prior knowledge of the pixel length of the steel support structure.

[0060] The possible existence range of the steel support structure can be represented as a quadrilateral object frame in the monitoring image. The monitoring image is sliced ​​according to the object frame to cut out an image of interest from the monitoring image and send it to a subsequent edge detection model.

[0061] In this way, a rough outline of the steel support can be obtained first, thereby limiting the area of ​​interest to a smaller range, avoiding the interference of useless information, and improving the recognition speed and accuracy.

[0062] S104: Use a pre-trained edge detection model to detect whether there is an edge of a steel support structure in the image of interest, and obtain a current steel support structure detection result.

[0063] In this step, the image of interest can be input into the pre-trained edge detection model to detect whether there is a steel support structure edge in the image of interest, thereby determining whether there is a steel support structure in the image of interest and obtaining the current steel support structure detection result. Here, if the steel support structure is detected, the ROI frame will be drawn with bbox to obtain the current steel support structure detection result (i.e., edge detection map).

[0064] In this way, using the combination of target detection and edge detection, the target detection algorithm first detects the steel support mounting tray and limits the ROI area to a smaller range to avoid capturing a lot of useless information. Then, the ROI area is expanded through prior knowledge so that the entire steel support is located within the expanded ROI selection area, and edge detection is performed on the steel support in this area. This can greatly accelerate the running speed of the model and improve the accuracy of recognition.

[0065] In a possible implementation, the edge detection model in the embodiment of the present application is improved based on the RCF model (Richer Convolutional Features for Edge Detection). VGG16 is used as a pre-trained model. The edge detection model is divided into multiple stages of edge detection modules connected in sequence, which are used to extract edge details of different granularities. The edge extracted by the latter edge detection module will be rougher and rougher than the edge extracted by the previous edge detection module. For example, the edge extracted by the edge detection module of the first stage has details, and the edge extracted by the edge detection module of the last stage is the entire contour; the two adjacent stages are downsampled by the pooling layer. Accordingly, step S104 may include:

[0066] First, the image of interest is input into the first edge detection module in the edge detection model to obtain a feature map output by the first edge detection module.

[0067] Afterwards, the input of the next edge detection module is determined according to the features of the intermediate convolution layer output of the previous edge detection module and the input of the previous edge detection module, so as to obtain the feature map output by the next edge detection module. Specifically, each edge detection module may include multiple convolution layers. The input of the next edge detection module can be obtained by adding the features of the intermediate convolution layer output of the previous edge detection module and the input of the previous edge detection module through residual connection, and then input into the next edge detection module to obtain the feature map output by the next edge detection module.

[0068] Finally, the feature maps output by each edge detection module are fused to obtain the current steel support structure detection result.

[0069] See also Figure 4 , Figure 4 This is a schematic diagram of the structure of an edge detection model provided in an embodiment of the present application. Figure 4 As shown in , the edge detection model includes five edge detection modules connected in sequence.

[0070] Then step S104 may include:

[0071] Stage 1: Input the image of interest into the first edge detection module to obtain the first feature of the first convolution layer output in the first edge detection module and the first feature map output by the first edge detection module. Specifically, the image of interest is sent to the first convolution layer of 3*3-64 (the convolution kernel size is 3*3, there are 64 convolution kernels, and the following description is based on this) in the first edge detection module for two convolutions to extract the first feature of the image, and then the two features are respectively subjected to 1*1-21 convolution to reduce the latitude of the feature channel, and then the two results are concat-operated and sent to 1*1-1 convolution to map the feature map to a single-channel output, and then the first feature map R1 of the normalized result is obtained through conv2d and batchnorm.

[0072] Stage2: Input the first feature into the second edge detection module to obtain the second feature output by the first convolution layer in the second edge detection module and the second feature map output by the second edge detection module. Specifically, input the feature map (i.e., the first feature) after two 3*3-64 first convolution layers in stage1 into the second edge detection module, perform two 3*3-128 first convolution layers and output the second feature, then concatenate with 1*1-21 convolutions respectively, and then send to 1*1-1 convolution, and then use deconvolution to make the image size the same as the result in stage1, and obtain the second feature map R2 of stage2.

[0073] Stage 3: First, the second feature is input into the CBAM attention layer, and the output feature of the CBAM attention layer is residually connected with the first feature. Specifically, the input (first feature) in stage 2 and the feature map after two 3*3*128 convolutions and one CBAM attention in stage 2 are residually connected as input.

[0074] After that, the result of the residual connection is used as the input of the third edge detection module to obtain the third feature output by the first convolution layer in the third edge detection module and the third feature map output by the third edge detection module. Specifically, the input is convolved with the first convolution layer of 3*3-256 to obtain the third feature; then concat with the convolution of 1*1-21, and then sent to the 1*1-1 convolution, and then the image size is made the same as the result in stage1 through deconvolution to obtain the third feature map R3 of stage3.

[0075] Stage4: Input the third feature into the fourth edge detection module to obtain the fourth feature output by the first convolution layer in the fourth edge detection module and the fourth feature map output by the fourth edge detection module. Specifically, the three 3*3*256 convolutions and 2*2 pooling downsampling in stage3 are used as input, and the first convolution layer of 3*3-512 is convolved to obtain the fourth feature, and then concat with 1*1-21 convolution, and then sent to 1*1-1 convolution, and then through deconvolution, the image size is made the same as the result in stage1, and the fourth feature map R4 of stage4 is obtained.

[0076] Stage 5: First, the fourth feature is input into the CBAM attention layer, and the output feature of the CBAM attention layer is residually connected with the input of the third edge detection module. Specifically, the input in stage 3 and the feature map after three 3*3*512 convolutions and CBAM attention in stage 4 are residually connected.

[0077] After that, the result of the residual connection is used as the input of the fifth edge detection module to obtain the fifth feature map output by the fifth edge detection module. Specifically, 3*3-512 convolution and 1*1-21 convolution are performed respectively, and then concat is performed, and then sent to 1*1-1 convolution, and then the image size is made the same as the result in stage1 through deconvolution, and the fifth feature map R5 of stage5 is obtained.

[0078] Finally, the first feature map, the second feature map, the third feature map, the fourth feature map and the fifth feature map are fused to obtain the current steel support structure detection result. Specifically, the images generated by the five stages are concat to obtain R_all, which is then fed into a 1*1-1 convolution. In the detection stage, the probability of each pixel belonging to an edge is obtained through the sigmoid function, thereby obtaining the current steel support structure detection result.

[0079] During the training phase, the loss function needs to be calculated. The loss function measures the difference between the edge detection map output by the model and the true edge map. Then, the model parameters are adjusted by optimizing the loss function so that the edge detection map output by the model is as close to the true edge map as possible.

[0080] In the above manner, the target detection model in the embodiment of the present application adds residuals to some stages, and the residual connection provides a direct path for information to bypass one or more network layers and pass directly to subsequent layers. This design makes it easier for the error signal to backpropagate to the early layers, thereby improving the optimization process of the network. On the other hand, the CBAM attention operation is added before the residual operation. CBAM uses channel and spatial attention mechanisms to selectively enhance important edge features in the image while suppressing irrelevant background information. This mechanism is particularly suitable for edge detection tasks because it can highlight contours and boundaries in the image while reducing the impact of noise features. The channel attention mechanism assigns weights on the feature channel dimension, emphasizing feature channels that contribute more to edge detection. The spatial attention mechanism is weighted on the spatial dimension of the feature map, focusing on areas containing significant edges.

[0081] On the other hand, the loss function of the edge detection model is determined according to the fusion feature map output during the training process; the loss function is expressed as:

[0082]

[0083]

[0084]

[0085] In the formula, y i , X i and W represent the true value, input and weight respectively; P(X) represents the sigmoid function; η is the set threshold, by setting η, some weak edges can be ignored; y i =0 indicates non-edge area, 1-P(X i ; W); indicates that the probability of the model predicting non-edge pixels is close to 0, log(1-P(X i ; W)) is used to map this probability to a loss value; α and β represent weight parameters; Y + represents the number of edge pixels, and Y- represents the number of non-edge pixels. This loss function is designed so that the model minimizes its predicted value in non-edge areas in order to better distinguish between edges and non-edges. Therefore, when the actual value is 0, this part of the loss function helps constrain the behavior of the model to ensure appropriate predictions in non-edge areas.

[0086] In this way, compared with the original RCF model, the original loss function all_layer_loss is modified to last_layer_loss. All_layer_loss applies the loss function to the output of each stage (R1-R5) and the fusion output (R_all). The final total loss function is the sum of the loss functions of all stages. Last_layer_loss only applies the loss function to the fusion output, and the calculated loss is the total loss. All_layer_loss adds loss to the output of each stage, which may cause the model to over-focus on the output of each stage and ignore the final fusion output. Last_layer_loss only focuses on the final output, avoiding this situation and helping to improve the generalization ability of the model. Last_layer_loss directly optimizes the final fusion output, allowing the model to better learn the final edge map without being disturbed by the output of each stage. This method is more direct and more conducive to improving the edge detection performance of the model.

[0087] In this way, the target detection model in the embodiment of the present application adds residuals to some stages to improve the optimization process of the network; adds CBAM attention operation before the residual operation to highlight the contours and boundaries in the image while reducing the impact of noise features; and modifies the loss function all_layer_loss to last_layer_loss. Therefore, after obtaining the edge detection map, compared with directly using the target detection box or canny operator to select the frame, the target can be detected more accurately, the accuracy of software detection is improved, and a better basis is provided for the subsequent identification of erection violations, making the calculated distance between steel supports more accurate.

[0088] S105. Determine whether the steel support erection at the construction site meets the construction requirements by comparing the current installation pallet detection result and the current steel support structure detection result of the current frame monitoring image with the installation pallet detection result and the steel support structure detection result of the historical frame monitoring image.

[0089] The construction requirements include requirements for erection time dimension and / or erection space dimension.

[0090] Specifically, step S105 may include:

[0091] Case 1: If the current installation pallet detection result is that no new installation pallet is detected, then based on the installation pallet detection result of the historical frame monitoring image, determine the first time duration in which no new installation pallet is detected, and based on the first time duration and the preset pallet time duration threshold, determine whether the steel support erection at the construction site meets the construction requirements of the erection time dimension.

[0092] For time detection, if the current installation pallet detection result is that no new installation pallet is detected, the number of image frames in which no new installation pallet is detected can be determined based on the installation pallet detection result of the historical frame monitoring image, and then the first duration in which no new installation pallet is detected can be determined in combination with the image acquisition time. Then, based on the first duration and the preset pallet duration threshold, it is determined whether the steel support erection at the construction site meets the construction requirements of the erection time dimension; for example, when the first duration exceeds the preset pallet duration threshold, it is considered that the pallet has not been installed within the time limit, and it is determined that the steel support erection at the construction site does not meet the construction requirements of the erection time dimension.

[0093] In this way, by manually setting the time interval threshold of adjacent pallets and providing intervals for alarm logic judgment, it is possible to effectively detect whether an installation pallet has not been installed within the time limit, thereby effectively improving construction safety.

[0094] Case 2: If the current steel support structure detection result is that no new steel support structure is detected, then the second time duration in which no new steel support structure is detected is determined based on the steel support structure detection result of the historical frame monitoring image, and based on the second time duration and the preset steel structure time duration threshold, it is determined whether the steel support erection at the construction site meets the construction requirements of the erection time dimension.

[0095] For time detection, if the current installation pallet detection result is that a new installation pallet is detected, and the current steel support structure detection result is that no new steel support structure is detected, then the number of image frames in which no new steel support structure is detected can be determined based on the steel support structure detection result of the historical frame monitoring image, and then combined with the image acquisition time, the second duration in which no new steel support structure is detected can be determined. Then, based on the second duration and the preset steel structure duration threshold, determine whether the steel support erection at the construction site meets the construction requirements of the erection time dimension; for example, when the second duration exceeds the preset steel structure duration threshold, even if the pallet is installed on time, it cannot actually play a supporting role because the steel support structure is not installed within the time limit, and it can also be determined that the steel support erection at the construction site does not meet the construction requirements of the erection time dimension.

[0096] In this way, by manually setting the threshold of the time interval between the installation of adjacent steel support structures, an interval is provided for alarm logic judgment, which can effectively detect whether the steel support is not installed after the time delay, thereby effectively improving the construction safety. In addition, the combination of the two time detections in case one and case two can comprehensively detect the erection of the steel support from the two aspects of the installation tray and each steel support structure, thereby more comprehensively detecting erection violations.

[0097] Case three: If the current steel support structure detection result is that a new steel support structure is detected, the erection distance between the new steel support structure and the adjacent steel support structure is determined based on the steel support structure detection result of the historical frame monitoring image, and based on the erection distance and the preset distance range, it is determined whether the steel support erection at the construction site meets the construction requirements of the erection space dimension.

[0098] Here, the construction site has corresponding spatial requirements for the distance between steel support structures. For example, a steel support should be set up every 30 meters to fully support the foundation pit support. Therefore, based on the current steel support structure detection results and the steel support structure detection results of the historical frame monitoring image, combined with the camera parameters of the collected monitoring image, the monocular camera ranging principle (such as similar triangles) can be used to calculate the distance of each steel support structure from the camera, and then determine the erection distance between adjacent steel support structures. Afterwards, based on the erection distance and the preset distance range, determine whether the steel support erection at the construction site meets the construction requirements of the erection space dimension. For example, if the steel supports are erected too densely or too sparsely, it can be determined that the steel support erection at the construction site does not meet the construction requirements of the erection space dimension.

[0099] Furthermore, after determining that the steel support erection at the construction site does not meet the construction requirements and there are installation delays and violations, a variety of methods can be used to issue alarms to promptly notify relevant personnel of the steel support erection violations and the specific locations, causes and parameters of the violations, which will help relevant personnel to promptly identify safety hazards.

[0100] The embodiment of the present application provides a method for detecting delayed violations of steel support erection. The installation pallet is detected through a target detection algorithm and the possible existence range of the steel support structure is expanded. The edge detection is then performed through an edge detection algorithm. The region of interest can be limited to a smaller range, avoiding interference from useless information and improving recognition speed and accuracy. Furthermore, by comparing the installation pallet detection results and the steel support structure detection results of the current frame and the historical frame, it is possible to automatically determine whether the steel support erection meets the construction requirements from the time dimension and / or space dimension, determine whether the steel support erection has delayed violations, and identify promptly and accurately, which helps to improve construction safety.

[0101] See also Figure 5 , Figure 5 This is a schematic diagram of the structure of a detection device for delayed steel support erection violations provided in an embodiment of the present application. Figure 5 As shown in , the detection device 200 includes:

[0102] The acquisition module 210 is used to continuously acquire monitoring images of the construction site according to a predetermined acquisition frequency;

[0103] The pallet detection module 220 is used to determine whether a new installation pallet is detected in the monitoring image of the current frame using a pre-trained target detection model, and obtain a current installation pallet detection result; wherein each steel support structure is fixedly installed on the foundation pit support at the construction site through the installation pallets at both ends to support the foundation pit support;

[0104] An expansion module 230 is configured to, if the current installation tray detection result is that a new installation tray is detected, expand the possible existence range of the steel support structure from the location of the new installation tray in the monitoring image in combination with the morphological prior knowledge of the steel support structure, and segment the image of interest from the monitoring image according to the possible existence range;

[0105] A steel support detection module 240 is used to detect whether there is a steel support structure edge in the image of interest using a pre-trained edge detection model to obtain a current steel support structure detection result;

[0106] The comparison module 250 is used to determine whether the steel support erection at the construction site meets the construction requirements by comparing the current installation pallet detection result and the current steel support structure detection result of the current frame monitoring image with the installation pallet detection result and the steel support structure detection result of the historical frame monitoring image; wherein the construction requirements include requirements for the erection time dimension and / or the erection space dimension.

[0107] Further, when the comparison module 250 is used to determine whether the steel support erection at the construction site meets the construction requirements by comparing the current installation tray detection result and the current steel support structure detection result of the current frame monitoring image with the installation tray detection result and the steel support structure detection result of the historical frame monitoring image, the comparison module 250 is used to:

[0108] If the current installation pallet detection result is that no new installation pallet is detected, then according to the installation pallet detection result of the historical frame monitoring image, determine the first time duration during which no new installation pallet is detected, and according to the first time duration and the preset pallet time duration threshold, determine whether the steel support erection at the construction site meets the construction requirements of the erection time dimension;

[0109] If the current steel support structure detection result is that no new steel support structure is detected, then according to the steel support structure detection result of the historical frame monitoring image, determine the second time duration during which no new steel support structure is detected, and according to the second time duration and the preset steel structure time duration threshold, determine whether the steel support erection at the construction site meets the construction requirements of the erection time dimension;

[0110] If the current steel support structure detection result is that a new steel support structure is detected, the erection distance between the new steel support structure and the adjacent steel support structure is determined based on the steel support structure detection result of the historical frame monitoring image, and based on the erection distance and the preset distance range, it is determined whether the steel support erection at the construction site meets the construction requirements of the erection space dimension.

[0111] Further, the morphological prior knowledge includes the erection angle of the steel support structure in the monitoring image; when the expansion module 230 is used to expand the possible existence range of the steel support structure from the position of the new installation pallet in the monitoring image in combination with the morphological prior knowledge of the steel support structure if the current installation pallet detection result is that a new installation pallet is detected, the expansion module 230 is used to:

[0112] In the monitoring image, the location of the new installation tray is taken as the starting point and extended according to the installation angle to obtain the possible existence range of the steel support structure.

[0113] Further, the edge detection model includes a plurality of edge detection modules connected in sequence; each edge detection module is used to extract edge details of different granularities; when the steel support detection module 240 is used to detect whether there is a steel support structure edge in the image of interest using a pre-trained edge detection model and obtain the current steel support structure detection result, the steel support detection module 240 is used to:

[0114] Inputting the image of interest into a first edge detection module in the edge detection model to obtain a feature map output by the first edge detection module;

[0115] Determine the input of the next edge detection module according to the features of the intermediate convolutional layer output of the previous edge detection module and the input of the previous edge detection module to obtain a feature map output by the next edge detection module;

[0116] The feature maps output by each edge detection module are fused to obtain the current steel support structure detection result.

[0117] Further, the edge detection model includes five edge detection modules connected in sequence; when the steel support detection module 240 is used to detect whether there is a steel support structure edge in the image of interest using a pre-trained edge detection model and obtain a current steel support structure detection result, the steel support detection module 240 is used to:

[0118] Inputting the image of interest into a first edge detection module to obtain a first feature output by a first convolutional layer in the first edge detection module and a first feature map output by the first edge detection module;

[0119] Inputting the first feature into a second edge detection module to obtain a second feature output by the first convolutional layer in the second edge detection module and a second feature map output by the second edge detection module;

[0120] Input the second feature into the CBAM attention layer, and perform a residual connection between the output feature of the CBAM attention layer and the first feature;

[0121] The result of the residual connection is used as the input of the third edge detection module to obtain the third feature output by the first convolutional layer in the third edge detection module and the third feature map output by the third edge detection module;

[0122] Inputting the third feature into a fourth edge detection module to obtain a fourth feature output by the first convolutional layer in the fourth edge detection module and a fourth feature map output by the fourth edge detection module;

[0123] Input the fourth feature into the CBAM attention layer, and perform a residual connection between the output feature of the CBAM attention layer and the input of the third edge detection module;

[0124] The result of the residual connection is used as the input of the fifth edge detection module to obtain a fifth feature map output by the fifth edge detection module;

[0125] The first characteristic map, the second characteristic map, the third characteristic map, the fourth characteristic map and the fifth characteristic map are integrated to obtain the current steel support structure detection result.

[0126] Furthermore, the loss function of the edge detection model is determined according to the fusion feature map output during the training process; the loss function is expressed as:

[0127]

[0128]

[0129]

[0130] In the formula, y i , X i and W represent the true value, input and weight respectively; P(X) represents the sigmoid function; η is the set threshold; y i = 0 indicates non-edge area; α and β indicate weight parameters; Y + Indicates the number of edge pixels, Y - Indicates the number of non-edge pixels.

[0131] Furthermore, the target detection model is constructed based on the YoloV5 neural network model; the target detection model includes an input module, a BackBone backbone module, a Neck intermediate module and a Head prediction module;

[0132] The BackBone module includes a Focus structure, a combination of at least one DCNBL unit and a CSP unit, a DCNBL unit, an SPP structure, and a GAM attention layer connected in sequence; wherein each DCNBL unit includes a variable convolution layer, a batch normalization layer, and an activation function layer;

[0133] The Neck middle module includes FPN structure, PAN structure and CSP unit;

[0134] The head prediction module uses CIOU_Loss as the loss function and post-processes the detection frame of the installation tray through non-maximum suppression to obtain the current installation tray detection result.

[0135] See also Figure 6 , Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. Figure 6 As shown in , the electronic device 400 includes a processor 410 , a memory 420 and a bus 430 .

[0136] The memory 420 stores machine-readable instructions executable by the processor 410. When the electronic device 400 is running, the processor 410 communicates with the memory 420 via the bus 430. When the machine-readable instructions are executed by the processor 410, the above-mentioned Figure 1 The steps of the method for detecting the delayed violation of the steel support erection in the method embodiment shown in the embodiment and the specific implementation method thereof can be referred to the method embodiment, which will not be described in detail here.

[0137] The present application also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the computer program can execute the above-mentioned Figure 1 The steps of the method for detecting the delayed violation of the steel support erection in the method embodiment shown in the embodiment and the specific implementation method thereof can be referred to the method embodiment, which will not be described in detail here.

[0138] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0139] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some communication interfaces, and the indirect coupling or communication connection of devices or units can be electrical, mechanical or other forms.

[0140] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0141] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0142] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a non-volatile computer-readable storage medium that is executable by a processor. Based on this understanding, the technical solution of the present application can essentially be embodied in the form of a software product, or in other words, the part that contributes to the prior art or the part of the technical solution. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0143] Finally, it should be noted that the above-described embodiments are only specific implementation methods of the present application, which are used to illustrate the technical solutions of the present application, rather than to limit them. The protection scope of the present application is not limited thereto. Although the present application is described in detail with reference to the above-mentioned embodiments, ordinary technicians in the field should understand that any technician familiar with the technical field can still modify the technical solutions recorded in the above-mentioned embodiments within the technical scope disclosed in the present application, or can easily think of changes, or make equivalent replacements for some of the technical features therein; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be based on the protection scope of the claims.

Claims

1. A method for detecting a steel support erection delay violation, characterized in that: The detection method comprises: Continuously collect monitoring images of the construction site according to the predetermined collection frequency; For the current frame monitoring image, a pre-trained target detection model is used to determine whether a new installation pallet is detected in the monitoring image, and a current installation pallet detection result is obtained; wherein each steel support structure is fixedly installed on the foundation pit support at the construction site through the installation pallets at both ends to support the foundation pit support; If the current installation tray detection result is that a new installation tray is detected, then combining the morphological prior knowledge of the steel support structure, the possible existence range of the steel support structure is expanded from the location of the new installation tray in the monitoring image, and the image of interest is segmented from the monitoring image according to the possible existence range; Use a pre-trained edge detection model to detect whether there is an edge of a steel support structure in the image of interest, and obtain a current steel support structure detection result; By comparing the current installation pallet detection results and the current steel support structure detection results of the current frame monitoring image with the installation pallet detection results and the steel support structure detection results of the historical frame monitoring image, it is determined whether the steel support erection at the construction site meets the construction requirements; wherein the construction requirements include requirements for the erection time dimension and / or the erection space dimension.

2. The detection method according to claim 1, characterized in that: By comparing the current installation pallet detection result and the current steel support structure detection result of the current frame monitoring image with the installation pallet detection result and the steel support structure detection result of the historical frame monitoring image, it is determined whether the steel support erection at the construction site meets the construction requirements, including: If the current installation pallet detection result is that no new installation pallet is detected, then according to the installation pallet detection result of the historical frame monitoring image, determine the first time duration during which no new installation pallet is detected, and according to the first time duration and the preset pallet time duration threshold, determine whether the steel support erection at the construction site meets the construction requirements of the erection time dimension; If the current steel support structure detection result is that no new steel support structure is detected, then according to the steel support structure detection result of the historical frame monitoring image, determine the second time duration during which no new steel support structure is detected, and according to the second time duration and the preset steel structure time duration threshold, determine whether the steel support erection at the construction site meets the construction requirements of the erection time dimension; If the current steel support structure detection result is that a new steel support structure is detected, the erection distance between the new steel support structure and the adjacent steel support structure is determined based on the steel support structure detection result of the historical frame monitoring image, and based on the erection distance and the preset distance range, it is determined whether the steel support erection at the construction site meets the construction requirements of the erection space dimension.

3. The detection method according to claim 1, characterized in that: The morphological prior knowledge includes the installation angle of the steel support structure in the monitoring image; the possible existence range of the steel support structure is expanded from the position of the new installation tray in the monitoring image in combination with the morphological prior knowledge of the steel support structure, including: In the monitoring image, the location of the new installation tray is taken as the starting point and extended according to the installation angle to obtain the possible existence range of the steel support structure.

4. The detection method according to claim 1, characterized in that: The edge detection model includes a plurality of edge detection modules connected in sequence; each edge detection module is used to extract edge details of different granularities; The using of the pre-trained edge detection model to detect whether there is an edge of a steel support structure in the image of interest to obtain a current steel support structure detection result includes: Inputting the image of interest into a first edge detection module in the edge detection model to obtain a feature map output by the first edge detection module; Determine the input of the next edge detection module according to the features of the intermediate convolutional layer output of the previous edge detection module and the input of the previous edge detection module to obtain a feature map output by the next edge detection module; The feature maps output by each edge detection module are fused to obtain the current steel support structure detection result.

5. The detection method according to claim 4, characterized in that: The edge detection model includes five edge detection modules connected in sequence; the use of the pre-trained edge detection model to detect whether there is an edge of a steel support structure in the image of interest to obtain a current steel support structure detection result includes: Inputting the image of interest into a first edge detection module to obtain a first feature output by a first convolutional layer in the first edge detection module and a first feature map output by the first edge detection module; Inputting the first feature into a second edge detection module to obtain a second feature output by the first convolutional layer in the second edge detection module and a second feature map output by the second edge detection module; Input the second feature into the CBAM attention layer, and perform a residual connection between the output feature of the CBAM attention layer and the first feature; The result of the residual connection is used as the input of the third edge detection module to obtain the third feature output by the first convolutional layer in the third edge detection module and the third feature map output by the third edge detection module; Inputting the third feature into a fourth edge detection module to obtain a fourth feature output by the first convolutional layer in the fourth edge detection module and a fourth feature map output by the fourth edge detection module; Input the fourth feature into the CBAM attention layer, and perform a residual connection between the output feature of the CBAM attention layer and the input of the third edge detection module; The result of the residual connection is used as the input of the fifth edge detection module to obtain a fifth feature map output by the fifth edge detection module; The first characteristic map, the second characteristic map, the third characteristic map, the fourth characteristic map and the fifth characteristic map are integrated to obtain the current steel support structure detection result.

6. The detection method according to claim 1, characterized in that: The loss function of the edge detection model is determined according to the fusion feature map output during the training process; the loss function is expressed as: In the formula, y i , X i and W represent the true value, input and weight respectively; P(X) represents the sigmoid function; η is the set threshold; y i = 0 indicates non-edge area; α and β indicate weight parameters; Y + Indicates the number of edge pixels, Y - Indicates the number of non-edge pixels.

7. The detection method according to claim 1, characterized in that: The target detection model is built on the basis of the YoloV5 neural network model; the target detection model includes an input module, a BackBone backbone module, a Neck intermediate module and a Head prediction module; The BackBone module includes a Focus structure, a combination of at least one DCNBL unit and a CSP unit, a DCNBL unit, an SPP structure, and a GAM attention layer connected in sequence; wherein each DCNBL unit includes a variable convolution layer, a batch normalization layer, and an activation function layer; The Neck middle module includes FPN structure, PAN structure and CSP unit; The head prediction module uses CIOU_Loss as the loss function and post-processes the detection frame of the installation tray through non-maximum suppression to obtain the current installation tray detection result.

8. A detection device for steel support erection delay violation, characterized in that: The detection device comprises: A collection module, used to continuously collect monitoring images of the construction site according to a predetermined collection frequency; The pallet detection module is used to determine whether a new installation pallet is detected in the monitoring image of the current frame using a pre-trained target detection model, and obtain the current installation pallet detection result; wherein each steel support structure is fixedly installed on the foundation pit support at the construction site through the installation pallets at both ends to support the foundation pit support; an expansion module, for, if the current installation tray detection result is that a new installation tray is detected, combining the morphological prior knowledge of the steel support structure, expanding the possible existence range of the steel support structure from the location of the new installation tray in the monitoring image, and segmenting the image of interest from the monitoring image according to the possible existence range; A steel support detection module is used to detect whether there is a steel support structure edge in the image of interest using a pre-trained edge detection model to obtain a current steel support structure detection result; A comparison module is used to determine whether the steel support erection at the construction site meets the construction requirements by comparing the current installation pallet detection result and the current steel support structure detection result of the current frame monitoring image with the installation pallet detection result and the steel support structure detection result of the historical frame monitoring image; wherein the construction requirements include requirements for the erection time dimension and / or the erection space dimension.

9. An electronic device, characterized in that: include: A processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor. When the electronic device is running, the processor and the memory communicate via the bus. When the machine-readable instructions are run by the processor, the steps of the method for detecting delayed violation of steel support erection as described in any one of claims 1 to 7 are executed.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method for detecting delayed violation of steel support erection as described in any one of claims 1 to 7 are executed.

Images