Reinforcement cage process identification and edge deployment method and device and storage medium

Through machine vision and edge computing technology, automatic identification and time recording of steel cage construction process is realized, solving the problem of steel cage placement time control and data packet loss in drilling pile construction, and improving the digital and intelligent level of construction , reducing the risk of hole collapse accidents.

CN119027867AActive Publication Date: 2024-11-26CCCC SECOND HARBOR ENGINEERING CO LTD

Patent Information

Application Number
CN202410876023.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-02
Publication Date
2024-11-26
Estimated Expiration
2044-07-02

AI Technical Summary

Technical Problem

The lack of scientific and effective process control methods in the construction of drilling piles, resulting in occasional hole collapse accidents, especially the time of the steel cage is not recorded and controlled in time, and the unstable wireless signal quality at the construction site leads to data packets in video streaming data transmission. Lost.

Method used

Using machine vision technology combined with edge computing, a reinforced cage process identification and edge deployment method is designed, data is collected through high-definition surveillance cameras, object detection is performed using the yolov7-tiny-iRMB model, and the model is deployed on the NVIDIA embedded development board to realize Automatic identification and recording of steel cage construction process to solve the problem of data packet loss.

Benefits of technology

Automatic identification and time recording of the steel cage process are realized, the digitalization and intelligence level of construction is improved, the probability of collapse accidents is reduced, and data transmission pressure and packet loss are reduced through local calculations.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119027867A_ABST
    Figure CN119027867A_ABST
Patent Text Reader

Abstract

The invention discloses a reinforcement cage process identification and edge deployment method and device and a storage medium, and the method comprises the following steps: 1, installing a high-definition monitoring camera, collecting reinforcement cage construction process data, and making a data set; step 2, designing a yolov7-tin-iRMB (Initial Register Model Broadcasting) model; step 3, the data set is used for training a yolov7-tin-iRMB (Rolov7-tin-iRMB) model; 4, detecting and outputting coordinates of the detection frame of the reinforcement cage, and obtaining track coordinates in the construction process of the reinforcement cage; 5, analyzing the track coordinates of the reinforcement cage, identifying the hoisting start, sleeve connection start, sleeve connection end and hoisting end procedures of the reinforcement cage, and recording the time of the corresponding procedures; and 6, carrying out model deployment by using an NVIDIA embedded development board. According to the method, automatic identification of the edge side reinforcement cage lowering procedure is achieved, so that technicians are helped to record the lowering time of each section of reinforcement cage, the construction progress is effectively controlled, hole collapse accidents are prevented, and the digitization and intelligence level of cast-in-situ bored pile construction is also improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of bored pile construction, and more specifically, to a method, device and storage medium for identifying a steel cage process and deploying edges. Background Art

[0002] The current construction technology of bored piles is relatively mature and has a high degree of standardization. However, due to inadequate on-site management and the lack of scientific and effective process control methods, hole collapse accidents occur occasionally. Among them, failure to pour concrete in time after cleaning the hole and placing it for too long are one of the causes of hole collapse accidents. How to automatically record the placement time of each section of the steel cage on site and allow on-site technicians to effectively control the steel cage that has been placed for too long has become an important research direction. The main technical problems to be solved in this application include: 1. How to fill the technical gap in the identification of steel cage construction processes based on machine vision in bored pile construction; 2. The quality of wireless signals at the construction site is uneven, and remote transmission of video stream data is prone to data packet loss. The use of edge computing technology realizes the local calculation of the video stream to solve the problem of data packet loss; 3. Implement a vehicle-mounted steel cage construction process identification device, use a 12V vehicle power supply, and solve the power supply problem of the device.

[0003] To solve the above problems, this application uses technologies such as machine vision, data processing and edge computing to propose a machine vision-based steel cage process identification and edge deployment method, equipment and storage medium. It is not only a technological innovation, but also a transformation and upgrade of the bored pile construction method. Summary of the invention

[0004] One purpose of the present invention is to provide a method, device and storage medium for identifying the process of steel cage and edge deployment, which realizes the automatic identification of the lowering process of the steel cage on the edge side, thereby helping technicians to record the lowering time of each section of the steel cage, effectively control the construction progress, prevent the occurrence of hole collapse accidents, and improve the digitization and intelligence level of bored pile construction.

[0005] In order to solve the above technical problems, the present invention provides a reinforcement cage process identification and edge deployment method, comprising the following steps:

[0006] Step 1: Install high-definition surveillance cameras to collect data on the steel cage construction process and create a data set, then divide the data set into a training set, a validation set, and a test set in proportion;

[0007] Step 2: Design the yolov7-tiny-iRMB model for steel cage target detection;

[0008] Step 3: Use the dataset prepared in step 1 to train the yolov7-tiny-iRMB model designed in step 2, that is, through image enhancement and model training, obtain a model suitable for steel cage detection, and convert the model into ONNX format for subsequent edge deployment;

[0009] Step 4: Detect and output the detection frame coordinates of the steel cage to obtain the trajectory coordinates of the steel cage during construction;

[0010] Step 5: By analyzing the acquired steel cage trajectory coordinates, identify the steel cage hoisting start, sleeve connection start, sleeve connection end and hoisting end processes and record the time of the corresponding processes;

[0011] Step 6: Use NVIDIA's embedded development board to deploy the model.

[0012] Preferably, in the step one, a total of two high-definition surveillance cameras are installed above the crawler crane cab and next to the bored piles to record the video of the steel cage construction process, and the hoisting images of the steel cage at different stages of construction are captured therefrom, and the steel cage hoisting images from two perspectives are obtained as a data set; in addition, Labelimg is used to mark the steel cage training set and verification set, and YOLO labels are generated.

[0013] Preferably, in the step 2, an iRMB module is added to the yolov7-tiny lightweight target detection network to design and form a yolov7-tiny-iRMB model.

[0014] Preferably, in step three, in step three, the model training is specifically as follows: first, the steel cage image is flipped symmetrically along its original axis, and the original data set is expanded several times to enhance the original data set; secondly, the data set is trained; finally, after the model is obtained, the model is converted into ONNX format.

[0015] Preferably, in step 5, the specific method for identifying and recording the steel cage construction process is:

[0016] S51, create a dictionary dict to record the number, process and time of the steel cage;

[0017] S52, when it is determined that the acquired steel cage trajectory coordinate data is no longer zero, a steel cage width threshold is set. When the steel cage detection frame width is greater than the set width threshold and exceeds the time t, it is considered that the detection target is correct, the current process is the start of steel cage hoisting, the steel cage sequence number is increased by 1, and the start and time of the steel cage hoisting of the corresponding sequence number are recorded in the dictionary dict;

[0018] S53, after identifying the beginning process of steel cage hoisting, the next process, i.e., the beginning of steel cage sleeve connection, is identified; a steel cage swing amplitude threshold is set, and when the steel cage swing amplitude is less than the set swing amplitude threshold and lasts for 3 seconds, the current process is identified as the beginning of steel cage sleeve connection, and the corresponding sequence number of steel cage sleeve connection start and time are recorded in the dictionary dict;

[0019] S54, after identifying the beginning process of the steel cage sleeve connection, the next process, i.e., the steel cage sleeve connection and the end of the wire coiling welding, is identified; first, whether the steel cage is rising or falling in real time is determined, and then a steel cage height threshold is set. When it is determined that the steel cage is falling in real time and the height is lower than the set height threshold for more than 3 seconds, the current process is identified as the steel cage sleeve connection and the end of the wire coiling welding, and the steel cage sleeve connection and the end of the wire coiling welding and the time of the corresponding sequence number are recorded in the dictionary dict;

[0020] S55. After identifying the end of the steel cage sleeve connection and wire coil welding process, the next process, namely, the end of the steel cage hoisting, will be identified; the steel cage descent speed is calculated and the height of the steel cage above the ground is obtained, and then the time of the end of the steel cage hoisting is calculated, and the end and time of the steel cage hoisting of the corresponding sequence number are recorded in the dictionary dict.

[0021] Preferably, in the process of step S53, a sliding window is used to update the width information of the steel cage, and the size of the sliding window is set to k, and the data in the current window are x1, x2, x3, ..., x k , the average value of the sliding window is Calculated by the following formula:

[0022] Among them, i is the starting position of the current sliding window, j is the index in the sliding window; the window sliding frequency is 2s / time.

[0023] Preferably, in step S54, the rising and falling states of the steel cage are judged in real time by a sliding window combined with a slope change method, assuming that the size of the sliding window is k, and the y coordinates in the current window are y1, y2, y3, ..., y k , then the slope of the sliding window changes:

[0024] Where i is the starting position of the current sliding window, k is the size of the sliding window; if Δslope i When it is greater than 0, it indicates real-time rise; if Δslope i If it is less than 0, it means real-time decrease.

[0025] Preferably, in step six, the specific process is: first, the system is burned and the environment is set up; secondly, the designed file is run in the development board to verify whether the model is inferred normally; again, the reasoning process is executed on the development board GPU, that is, the input data is forward calculated through the TensorRT engine to obtain the output of the model, so as to achieve model optimization and acceleration; then, the reasoning result is written into the steel cage construction process identification algorithm in real time using the publish-subscribe mode for identifying the steel cage process; finally, a module is installed in the development board to publish the recognition result of the steel cage construction process discrimination algorithm.

[0026] The present invention provides a computer device, which includes a memory, a processor, and a computer program for steel cage process identification and edge deployment stored in the memory and executable by the processor. When the processor executes the computer program, the steps of the method described above are implemented.

[0027] The present invention provides a storage medium, wherein the storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method described above are implemented.

[0028] The present invention has at least the following beneficial effects:

[0029] 1. This application, by utilizing computer vision algorithms and image recognition technology, can achieve real-time monitoring and intelligent identification of the construction site, improve the level of digitization and intelligence in engineering construction, and fill the gap in automatic identification technology of steel cage construction processes in bored pile construction.

[0030] 2. This application realizes the automatic identification of the lowering process of the edge side steel cage, thereby helping technicians record the lowering time of each section of the steel cage, effectively controlling the construction progress, preventing the occurrence of hole collapse accidents, and also improving the digitalization and intelligence level of bored pile construction.

[0031] 3. The vehicle-mounted edge computing device developed in this application can calculate the video stream locally, reduce the transmission pressure and packet loss when uploading data to the cloud, and improve the real-time and applicability of the method.

[0032] 4. This application realizes a vehicle-mounted steel cage construction process identification device, which uses a 12V vehicle-mounted power supply to solve the power supply problem of the device.

[0033] Other advantages, objectives and features of the present invention will be embodied in part through the following description, and in part will be understood by those skilled in the art through study and practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 It is a structural diagram of the reinforcement cage data set of the present invention;

[0035] Figure 2 It is the framework diagram of iRMB of the present invention;

[0036] Figure 3 It is the framework diagram of yolov7-tiny-iRMB of the present invention;

[0037] Figure 4 It is a steel cage process identification flow chart of the present invention;

[0038] Figure 5 This is an implementation example of the machine vision-based steel cage process identification device of the present invention. DETAILED DESCRIPTION

[0039] In order to better understand the purpose, structure and function of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings so that those skilled in the art can implement it according to the description.

[0040] It should be noted that the experimental methods described in the following embodiments are conventional methods unless otherwise specified, and the reagents and materials are commercially available unless otherwise specified; in the description of the present invention, the terms "lateral", "longitudinal", "upper", "lower", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention.

[0041] This application uses machine vision, data processing, and edge computing technologies to implement a vehicle-mounted device that records the process of steel cage placement on the edge side, automatically records the current process and placement time of each steel cage, and transmits it to the visualization cloud platform via the http protocol. From the visualization interface of the cloud platform, technicians can understand whether there are steel cages that have been placed for too long on site, and then effectively control the placement time of the steel cage, reduce the probability of collapse accidents, and improve the digitalization and intelligence level of bored pile construction. Therefore, a method, device, and storage medium for identifying steel cage processes and deploying them on the edge based on machine vision is not only a technological innovation, but also a transformation and upgrade of bored pile construction methods.

[0042] This application uses a target detection algorithm to track and identify the steel cage in the camera image to obtain its motion trajectory and coordinates during the lifting process. The process recognition algorithm is used to analyze the changing trend of the steel cage trajectory coordinates in real time to determine the four process nodes, including the start of lifting of each section of the steel cage, the start of sleeve connection, the end of sleeve connection, and the end of lifting, and automatically record the time of each node. This algorithm is deployed on an embedded development board with artificial intelligence computing power and connected to a 12V vehicle power supply to realize the vehicle-mounted device for edge computing. TensorRT is used to improve the model inference speed, thereby improving the computing performance of the embedded development board. The above method realizes the automatic identification of the lowering process of the edge side steel cage, thereby helping technicians record the lowering time of each section of the steel cage, effectively control the construction progress, and prevent collapse accidents.

[0043] The present invention provides a method for identifying the construction process of a steel cage and deploying its edges based on machine vision, and the specific implementation steps are as follows:

[0044] S1. Data collection and dataset creation.

[0045] A total of two high-definition surveillance cameras were installed above the crawler crane cab and next to the bored piles to record the video of the steel cage construction process and capture the hoisting images of the steel cage at different stages of construction. At the same time, the steel cage hoisting images from two perspectives were obtained to increase sample diversity.

[0046] A total of 600 images were captured and divided into training set, validation set and test set in a ratio of 7:2:1. Labelimg was used to label the training set and validation set of steel cages and generate YOLO labels. The structure of the steel cage dataset is as follows: Figure 1 shown.

[0047] S2. Design the yolov7-tiny-iRMB model.

[0048] The construction site environment is complex and there are many interference factors. The camera may be backlit, dark, overexposed, underexposed, or blocked during shooting, which will affect the detection effect of the target detection network on the steel cage. The steel cage process recognition algorithm is mainly based on the coordinate changes of the steel cage detection frame as input. The unstable steel cage detection effect may lead to the possibility of misjudgment of the process recognition algorithm. To avoid this situation, this application adds an iRMB module to the yolov7-tiny lightweight target detection network to enhance the robustness and anti-interference ability of the detection network and improve the model detection accuracy. After adding the iRMB module, the parameters of the entire model only increase slightly, which will not increase the computing burden of the equipment.

[0049] iRMB module framework Figure 2As shown in the figure, Attn Mat is the attention matrix. iRMB contains standard convolution, multi-head self-attention and DW-Conv modules. The DW-Conv module only performs convolution on the input channel dimension, but not on the output channel dimension, which greatly reduces the number of model parameters and the complexity of the model. In the convolution operation, DW-Conv only performs convolution on the channel dimension, but not on the spatial dimension, which can keep the spatial dimension of the input feature map unchanged and help preserve the spatial features of the image. The feature map of each input channel is only convolved with the corresponding convolution kernel, which maintains the independence between the input channels, can better extract features, and reduce the risk of overfitting to a certain extent.

[0050] like Figure 3 As shown in the figure, the iRMB module is added after the 57th, 65th, and 73rd layers of the yolov7-tiny network. Inside each branch, the attention mechanism is used to realize the long-range interaction of features and the global context fusion. The attention mechanism adaptively adjusts the feature weights of each position to enhance the representation of important features and suppress unimportant features. The features of each branch are spliced ​​in the channel direction to form an overall feature representation. The iRMB operation can make full use of multi-scale and multi-semantic feature information by processing multiple branches in parallel and introducing the attention mechanism, thereby enhancing the network's expression and feature representation capabilities.

[0051] S3. Train the yolov7-tiny-iRMB model.

[0052] First, the image is flipped to flip the steel cage image symmetrically, i.e. horizontal flip, vertical flip, and horizontal and vertical flip. After flipping, the original data set can be expanded three times. After the above enhancement of the original data set, the data set images are expanded from 600 to 1800.

[0053] Secondly, write the yolov7-tiny-iRMB.yaml file path and the Rebar Cage dataset path to the train.py file, and start the network model training on the Ubuntu 20.04 system and the GeForce GTX 2080Ti GPU. The SGD stochastic gradient descent algorithm is used in the training process. The number of training epochs is set to 400 in the experiment, the initial learning rate of the network is set to 0.001, and the training batch_size is set to 24.

[0054] Finally, obtain the model best.pt, start the onnx model conversion code gen_wts.py to convert the model architecture, and obtain the best.onnx model to match the TensorRT accelerated inference model format in the subsequent edge deployment process.

[0055] S4. Obtain the trajectory coordinates of the steel cage.

[0056] Start the detect.py file, select best.pt as the weights, select the source path to realize the recorded steel cage construction video or use Opencv to call the camera to obtain the steel cage construction video on site, and detect the detection frame coordinates xyxy of the steel cage for output, so as to obtain the trajectory coordinates of the steel cage during the construction process.

[0057] S5, steel cage construction process identification algorithm. Figure 4 shown.

[0058] 1. Create a dictionary dict_cage to record the number, process and time of the steel cage.

[0059] 2. When the steel cage target is detected, the acquired trajectory coordinate data is no longer zero. In order to eliminate the possibility of false detection, this application determines whether there is a false detection by the size of the steel cage detection frame, and sets the steel cage width threshold to false_cage_w. When the detection frame width is greater than false_cage_w and exceeds the time t, it is considered that the detected target is correct, the current process is the start of the steel cage hoisting, the steel cage sequence number cagecount is increased by 1, and {the start of the hoisting of the cagecount-th steel cage: time} is recorded in dict_cage.

[0060] 3. After identifying the start process of steel cage hoisting, the next process - steel cage sleeve connection will be identified (if the current steel cage sequence number cagecount is 1, there is no need to identify this process, and the first steel cage does not need sleeve connection); at the same time, we need to filter valid data in real time to eliminate the interference of false detection data on the algorithm and improve the stability of the algorithm. We use a sliding window to update the width information of the steel cage. Let the size of the sliding window be k, and the data in the current window is x1, x2, x3, ..., x k The average value of the sliding window is It can be calculated by the following formula:

[0061] Where i is the starting position of the current sliding window and j is the index in the sliding window. This formula means that each time the window is moved, the data in the current window is summed and divided by the window size k to get the average value.

[0062] Each time we move the window, we only need to add the new element and subtract the oldest element in the window, then calculate the sum of the elements in the window again and divide it by the window size to get the new average value. The window sliding frequency is 2s / times.

[0063] 4. This application sets the steel cage swing amplitude threshold to x_dif. When the steel cage swing amplitude is less than x_dif and lasts for 3 seconds, it is recognized that the current process is the start of steel cage sleeve connection, and records {the start of cagecount-th steel cage sleeve connection: time} in dict_cage.

[0064] 5. After identifying the beginning process of the steel cage sleeve connection, the next process - steel cage sleeve connection and wire coiling welding will be identified. During the steel cage sleeve connection and wire coiling welding process, the crane will constantly change the lifting height of the steel cage, which makes it extremely difficult for us to identify this process. This application will use a sliding window combined with a slope change method to judge the rising and falling state of the steel cage in real time. The specific formula is as follows:

[0065] Assume that the size of the sliding window is k, and the y coordinates in the current window are y1, y2, y3, ..., y k . Then the slope of the sliding window changes: Where i is the starting position of the current sliding window, and k is the size of the sliding window. This formula shows that each time the window is moved, the rate of change of the y coordinate in the current window (i.e., the slope change) is calculated.

[0066] Then, the trend of slope change can be determined by setting a threshold. i When it is greater than 0, it indicates real-time rise; if Δslope i If it is less than 0, it means real-time decrease.

[0067] Set the height threshold y_dif of the steel cage. When it is determined that the steel cage is descending in real time and its height is lower than y_dif for more than 3 seconds, recognize that the current process is the connection of the steel cage sleeve and the end of wire welding, and record {the connection of the cagecount-th steel cage sleeve and the end of wire welding: time} in dict_cage.

[0068] 6. After identifying the completion of the steel cage sleeve connection and wire coiling welding process, the next process - the completion of the steel cage hoisting process - will be identified.

[0069] First, calculate the descending speed of the steel cage. The specific formula is as follows:

[0070] Among them, y i is the height of the steel cage process when it is identified as the steel cage sleeve connection and wire welding end process, Δt represents the time from the steel cage sleeve connection and wire welding end to the steel cage target disappearing from the screen, this formula represents the average speed v of the steel cage lowering ↓i .

[0071] Although the steel cage disappears from the camera image, a portion of the steel cage is still exposed on the ground. This application uses the height y of the steel cage exposed on the ground. j and the average speed v of the steel cage lowering ↓i Calculate the time when the steel cage hoisting is completed. The formula is as follows:

[0072] Record in dict_cage {the lifting of cagecount-th steel cage ends: time t j}.

[0073] S7. Edge deployment.

[0074] The present invention uses NVIDIA's embedded development board for environment construction and algorithm deployment. The deployment process is as follows:

[0075] 1. First, burn Ubuntu 18.04 to the development board storage medium - TF card, configure the required environment, and install various support packages, such as CUDA, pip3, jtop, opencv, pytorch, torchvision and other dependencies and libraries. The above storage medium can also be a magnetic disk drive (Magnetic Disk Drive), a traditional hard disk drive (Traditional Hard Disk Drive, HDD), a rotating hard disk (Rotational Hard Disk), a magnetic recording storage (Magnetic Recording Storage), a magnetic storage medium (Magnetic Storage Medium), a solid state drive (Solid State Drive, SSD), a semiconductor storage device (Semiconductor Storage Device), a flash memory hard disk (Flash Memory HardDisk), a non-volatile storage medium (Non-Volatile Storage Medium), a portable flash drive (PortableFlash Drive), a USB flash memory (USB Flash Memory), a portable storage device (Portable StorageDevice), etc.

[0076] 2. After the system is burned and the environment is built, the TF card is currently used as the main storage medium in the embedded system to store the operating system, application programs and other software. Run detect.py on the development board to verify whether the model is inferring normally.

[0077] 3. Use the API provided by TensorRT to add the onnx format model to the TensorRT engine, and select the appropriate inference algorithm and calculation accuracy (FP32, FP16, or INT8, etc.).

[0078] 4. Execute the inference process on the development board GPU, that is, use the TensorRT engine to perform forward calculations on the input data to obtain the output of the model. TensorRT will optimize and parallelize according to the hardware characteristics of the GPU and the structure of the model to improve the inference speed and efficiency.

[0079] 5. Use the publish-subscribe mode to write the inference results into the steel cage construction process discrimination algorithm in real time to identify the steel cage process.

[0080] 6. The SIM7600G-H 4G module is installed in the development board to publish the recognition results of the steel cage construction process identification algorithm.

[0081] The present invention also provides an embodiment of a steel cage process identification device based on machine vision, such as Figure 5 As shown, including:

[0082] Video acquisition module 11, used for collecting steel cage lowering video stream data in real time;

[0083] Reasoning module 22, used for steel cage detection model reasoning, obtaining the steel cage motion trajectory coordinates; and steel cage process identification algorithm calculation, obtaining the current process and time of the steel cage in real time;

[0084] Data transmission module 33, used to transmit text data of current process and time of steel cage;

[0085] The heat dissipation module 44 is used to reduce the temperature of the device so that the device can perform computing performance stably;

[0086] The positioning module 55 is used to determine the position of the device.

[0087] The present invention also provides a computer device, which includes a memory, a processor, and a computer program for steel cage process identification and edge deployment stored in the memory and executable by the processor. When the processor executes the computer program, the steps of the method described in the present application are implemented.

[0088] The present invention also provides a storage medium, wherein the storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method described in the present application are implemented.

[0089] It is understood that the present invention is described by some embodiments, and those skilled in the art are aware that various changes or equivalent substitutions may be made to these features and embodiments without departing from the spirit and scope of the present invention. Although the embodiments of the present invention have been disclosed as above, they are not limited to the applications listed in the specification and the implementation mode, and they can be fully applied to various fields suitable for the present invention. For those familiar with the art, other modifications can be easily implemented, so without departing from the general concept defined by the claims and the equivalent scope, the present invention is not limited to the specific details and the illustrations shown and described here.

Claims

1. A method for identifying and deploying steel cage processes, characterized in that: The steps include: Step 1: Install high-definition surveillance cameras to collect data on the steel cage construction process and create a data set, then divide the data set into a training set, a validation set, and a test set in proportion; Step 2: Design the yolov7-tiny-iRMB model for steel cage target detection; Step 3: Use the dataset prepared in step 1 to train the yolov7-tiny-iRMB model designed in step 2, that is, through image enhancement and model training, obtain a model suitable for steel cage detection, and convert the model into ONNX format for subsequent edge deployment; Step 4: Detect and output the detection frame coordinates of the steel cage to obtain the trajectory coordinates of the steel cage during construction; Step 5: By analyzing the acquired steel cage trajectory coordinates, identify the steel cage hoisting start, sleeve connection start, sleeve connection end and hoisting end processes and record the time of the corresponding processes; Step 6: Use NVIDIA's embedded development board to deploy the model.

2. The steel cage process identification and edge deployment method according to claim 1, characterized in that: In the step one, a total of two high-definition surveillance cameras are installed above the crawler crane cab and next to the bored piles to record the video of the steel cage construction process, and the hoisting images of the steel cage at different stages of construction are captured therefrom, and the steel cage hoisting images from two perspectives are obtained as a data set; in addition, Labelimg is used to mark the steel cage training set and verification set, and YOLO labels are generated.

3. The steel cage process identification and edge deployment method according to claim 1, characterized in that: In the step 2, the iRMB module is added to the yolov7-tiny lightweight target detection network to design and form a yolov7-tiny-iRMB model.

4. The steel cage process identification and edge deployment method according to claim 2, characterized in that: In the step three, the model training is specifically as follows: first, the steel cage image is flipped symmetrically along its original axis, and the original data set is expanded several times to enhance the original data set; secondly, the data set is trained; finally, after the model is obtained, the model is converted into ONNX format.

5. The steel cage process identification and edge deployment method according to claim 1, characterized in that: In step 5, the specific method for identifying and recording the steel cage construction process is: S51, create a dictionary dict to record the number, process and time of the steel cage; S52, when it is determined that the acquired steel cage trajectory coordinate data is no longer zero, a steel cage width threshold is set. When the steel cage detection frame width is greater than the set width threshold and exceeds the time t, it is considered that the detection target is correct, the current process is the start of steel cage hoisting, the steel cage sequence number is increased by 1, and the start and time of the steel cage hoisting of the corresponding sequence number are recorded in the dictionary dict; S53, after identifying the beginning process of steel cage hoisting, the next process, i.e., the beginning of steel cage sleeve connection, is identified; a steel cage swing amplitude threshold is set, and when the steel cage swing amplitude is less than the set swing amplitude threshold and lasts for 3 seconds, the current process is identified as the beginning of steel cage sleeve connection, and the corresponding sequence number of steel cage sleeve connection start and time are recorded in the dictionary dict; S54, after identifying the beginning process of the steel cage sleeve connection, the next process, i.e., the steel cage sleeve connection and the end of the wire coiling welding, is identified; first, whether the steel cage is rising or falling in real time is determined, and then a steel cage height threshold is set. When it is determined that the steel cage is falling in real time and the height is lower than the set height threshold for more than 3 seconds, the current process is identified as the steel cage sleeve connection and the end of the wire coiling welding, and the steel cage sleeve connection and the end of the wire coiling welding and the time of the corresponding sequence number are recorded in the dictionary dict; S55. After identifying the end of the steel cage sleeve connection and wire coil welding process, the next process, namely, the end of the steel cage hoisting, will be identified; the steel cage descent speed is calculated and the height of the steel cage above the ground is obtained, and then the time of the end of the steel cage hoisting is calculated, and the end and time of the steel cage hoisting of the corresponding sequence number are recorded in the dictionary dict.

6. The steel cage process identification and edge deployment method according to claim 5, characterized in that: In the process of step S53, the width information of the steel cage is updated using a sliding window. Assume that the size of the sliding window is k, and the data in the current window are x1, x2, x3, ..., x k , the average value of the sliding window is Calculated by the following formula: Among them, i is the starting position of the current sliding window, j is the index in the sliding window; the window sliding frequency is 2s / time.

7. The steel cage process identification and edge deployment method according to claim 5, characterized in that: In step S54, the rising and falling states of the steel cage are judged in real time by combining the sliding window with the slope change method. Assume that the size of the sliding window is k, and the y coordinates in the current window are y1, y2, y3, ..., y k , then the slope of the sliding window changes: Where i is the starting position of the current sliding window, k is the size of the sliding window; if Δslope i When it is greater than 0, it indicates real-time rise; if Δslope i If it is less than 0, it means real-time decrease.

8. The steel cage process identification and edge deployment method according to claim 1, characterized in that: In step six, the specific process is: first, the system is burned and the environment is set up; secondly, the designed file is run in the development board to verify whether the model is inferred normally; again, the inference process is executed on the development board GPU, that is, the input data is forward calculated through the TensorRT engine to obtain the output of the model, so as to achieve model optimization and acceleration; then, the inference result is written into the steel cage construction process identification algorithm in real time using the publish-subscribe mode, which is used to identify the steel cage process; finally, a module is installed in the development board to publish the identification result of the steel cage construction process discrimination algorithm.

9. A computer device, characterized in that: The computer device includes a memory, a processor, and a computer program for steel cage process identification and edge deployment stored in the memory and executable by the processor. When the processor executes the computer program, the steps of the method according to any one of claims 1 to 8 are implemented.

10. A storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

Citation Information

Patent Citations

  • Working procedure detection device based on deep learning, and working procedure detection method thereof

    CN108491759A

  • Video stream-based reinforcing steel bar binding process monitoring method and monitoring device

    CN116092012A

  • Intelligent forklift system

    CN116246297A

  • Cable accessory installation process management and control system and method based on deep neural network

    CN116958864A

  • Method and system for detecting key process of tunneling roadway

    CN116977941A

Cited By

  • Reinforcement cage lowering process distinguishing method based on machine vision

    CN118781070A

  • Rolov7-based inquiry identification auxiliary method and system

    CN119540836A

  • Precast beam manufacturing process automatic identification method and system based on multi-source data

    CN120146432A

  • Precast beam process identification method, device and equipment based on edge calculation and storage medium

    CN121305436A