Construction method and system of special engineering construction site image data set

By using CVAT and OpenCV combined with YOLOv5 target detection platform at the construction site, the image data set construction method was constructed, which solved the data acquisition and labeling quality problems, and realized high-quality image data set construction, which was suitable for real-time performance at the construction site. and accuracy requirements.

CN119027692APending Publication Date: 2024-11-26POWERCHINA HUADONG ENG CORP LTD

Patent Information

Application Number
CN202411049109.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-01
Publication Date
2024-11-26

AI Technical Summary

Technical Problem

Image data acquisition at the construction site is difficult to cover the entire area, the imaging quality is unstable, and the quality of image data labeling is affected by weather, equipment and technical level, making it difficult to ensure the quality of training data.

Method used

Using interactive image annotation tool CVAT and computer vision and machine learning software library OpenCV, we establish a standardized data set construction method for image acquisition, preprocessing, labeling, quality inspection and data enhancement, and use high-definition cameras, binocular cameras and mobile phones to obtain image data. , and the labeling and quality audits are carried out through the YOLOv5 target detection platform and CVAT tool.

Benefits of technology

It realizes the acquisition of high-quality and diverse image data sets at the construction site, reduces the technical needs of labeling personnel, reduces the influence of human factors, ensures the accuracy and completeness of the training data, and improves the robustness and generalization of the model. ability.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119027692A_ABST
    Figure CN119027692A_ABST
Patent Text Reader

Abstract

The invention provides a construction site-oriented image data set construction method, and belongs to the technical field of image processing. Comprising the following steps: S1, data acquisition: acquiring construction site image data of different time periods, areas, illumination and view angle factors through shooting; s2, data preprocessing: carrying out normalization and type conversion processing on the collected field image data; s3, data labeling: using YOLO and CVAT to detect and label types, positions and track information of field personnel, materials and mechanical equipment; s4, data quality inspection: using an OpenCV function to assist quality assurance personnel to evaluate the accuracy, integrity, consistency and credibility of the annotated data; and S5, data enhancement: amplifying the data with poor integrity based on a computer vision library OpenCV. According to the method, image data annotation is completed with the assistance of the computer vision technology, the diversity and integrity of a data set are ensured, meanwhile, the technical requirement of annotation for personnel is lowered, the influence of human factors is reduced, and the quality of training data is ensured.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the technical field of image processing, and in particular relates to a method for constructing an image data set for a construction site. Background Art

[0002] Thanks to the rapid development of computer technology, target detection methods based on computer vision technology have been applied to the field of engineering construction, and high-quality training data has provided convenience for construction site management. However, due to the long duration of construction and the complex on-site environment, the images collected at the construction site are often difficult to cover the entire construction area and construction period, and are affected by weather and equipment, making it difficult to guarantee the image quality and image diversity. In addition, during the application process, the technical level of image data annotators is also insufficient, making it even more difficult to guarantee the quality of training data.

[0003] To address these challenges, the present invention adopts the interactive image annotation tool CVAT and the computer vision and machine learning software library OpenCV to establish an easy-to-operate standardized dataset construction method that integrates image acquisition, image preprocessing, image data annotation, image data quality inspection and image data enhancement, so as to make full use of image data that is easy to acquire, reduce the technical requirements of the annotation method itself on personnel, reduce the impact of human factors, and ensure the quality of training data.

[0004] Among many target detection algorithms, YOLO (You Only Look Once) is widely used in various scenarios due to its efficient real-time and accuracy. YOLO can predict the position and category of multiple objects in a single forward pass, has high detection speed and accuracy, and is particularly suitable for processing real-time video streams and scenes that require quick response. Compared with other detection methods, such as the R-CNN series of algorithms, the YOLO model is lighter and suitable for running in an environment with limited hardware resources. Therefore, the present invention selects YOLO as the main target detection technology to meet the requirements of real-time and accuracy at the construction site. Summary of the invention

[0005] To achieve the above object, the present invention provides a method for constructing an image dataset for a construction site, comprising the following steps:

[0006] Step S1, data collection: according to the special engineering construction organization design plan and the preset monitoring coverage area of ​​the construction site environment, high-definition cameras and binocular cameras are used within the determined monitoring coverage to obtain video streams of personnel, materials, and mechanical equipment during the construction process, and mobile phones are used to take pictures or videos related to the construction process in areas not covered by the monitoring, so as to obtain image data of multiple different time periods, areas, viewing angles, and lighting conditions;

[0007] Step S2, data preprocessing: configure the OpenCV library in the Python environment, establish connections with the high-definition cameras, binocular cameras and mobile devices arranged on site, integrate the images of construction site personnel, materials, and mechanical equipment captured from different devices, and perform unified normalization and type conversion on these images to achieve standardization of data format;

[0008] Step S3, data labeling: On the built YOLOv5 target detection platform, perform preliminary target detection and labeling on the standardized images, identify the target categories of on-site personnel, materials, and mechanical equipment, and then use the CVAT data labeling tool to further improve the labeling of these targets, including accurately labeling the target category, location, and trajectory information;

[0009] Step S4, data quality check: Use OpenCV functions to read and compare the annotated image and the original image to check and identify errors in the annotation or deviations from the sample. The quality assurance personnel will conduct a final review of the format, accuracy and fit of the annotated images to ensure that all image annotations used for model training are consistent, accurate and complete;

[0010] Step S5, data enhancement: For the annotated images that have passed the quality review, data augmentation is performed through OpenCV functions to improve the robustness and generalization ability of the model.

[0011] Furthermore, step S1 includes the following steps:

[0012] Step S101, according to the special engineering construction organization design plan and the construction site environment, clarify the entry of personnel, materials, and mechanical equipment in the construction scene and their possible activities on the site, and preliminarily select the layout positions of binocular cameras and high-definition cameras;

[0013] Step S102: Arrange a binocular camera and a high-definition camera at the initially selected position, debug the equipment and configure the operating environment, obtain the video stream at the position, and if the video is too dark, exposed or blocked, adjust the positions of the binocular camera and the high-definition camera until a clear and complete video stream is obtained;

[0014] Step S103, using the arranged binocular camera and high-definition camera, obtain the video stream of the personnel, materials, and mechanical equipment at the special engineering construction site, and use the mobile phone to capture relevant images or videos of the uncovered area to ensure that the training image data set covers the entire construction area.

[0015] Further, step S2 includes the following steps:

[0016] Step S201, configuring the computer vision library OpenCV in the Python environment, and testing the built environment to ensure that the computer vision library is successfully introduced;

[0017] Step S202, in the Python-OpenCV environment, using the video operation function cv2.VideoCapture() as an interface, by inputting the device serial numbers of the on-site high-definition camera and the binocular camera, obtaining the device index, and implementing the call of these external devices;

[0018] Step S203, through the cv2.VideoCapture().read() function, skip frames from the video stream obtained by the high-definition camera and the binocular camera to read the images of personnel, materials, and mechanical equipment at the construction site, and store them;

[0019] Step S204, normalize the images read from the high-definition camera and the binocular camera, and adjust the pixel values ​​to the interval (0, 1) to ensure smooth convergence of the model in subsequent data set training.

[0020] Further, step S3 includes the following steps:

[0021] Step S301, setting labels for personnel, materials, and mechanical equipment in the special engineering construction site, using professional terms to distinguish the static and dynamic attributes of these objects, for example, setting the static attribute to excavator, and setting the dynamic attributes to digging, loading, unloading, and transferring;

[0022] Step S302, based on the YOLOv5 target detection platform and the set labels, perform preliminary target detection and target pre-labeling on the standardized image data, generate the corresponding txt labeling file and save the labeling results in a format that can be read by CVAT;

[0023] Step S303, installing and configuring the desktop development environment Docker Desktop in the Windows system, enabling the virtual machine Hyper-V, and downloading and configuring CVAT;

[0024] Step S304: Enter the configured CVAT interface, create a labeling task according to the interface's operating requirements, and select YOLO1.1 format to upload the corresponding ZIP file;

[0025] Step S305, return to the task panel and perform quality inspection on the displayed annotation results.

[0026] Further, step S4 includes the following steps:

[0027] Step S401, in the Python-OpenCV environment, export the annotated image from the CVAT server through the cvat_export_annotations function, and use OpenCV to compare the annotated image with the original image to ensure that the content in the image is accurately annotated, and then store the correctly annotated and incorrectly annotated images separately;

[0028] Step S402, the quality assurance personnel checks the correctly annotated images, verifies the shapes and contours of the annotations, ensures that the data annotations in the images used for training are not missing or cut off, and then stores the images with complete annotations and the images with missing annotations separately;

[0029] Step S403, for the images that have been checked and annotated correctly and completely, check the clarity of the annotated boundaries to ensure that all annotated boundaries are clear, and finally store the qualified and unqualified files separately.

[0030] Further, step S5 includes the following steps:

[0031] Step S501, for images that have passed the quality inspection, are accurately and completely labeled, and cover different scenes and conditions, in the Python-OpenCV environment, scale matching is performed on the original images according to the target size to reduce the gap between the target and the image edge;

[0032] Step S502, performing data augmentation on the scale-matched images, including scaling, flipping, rotating, and cropping operations, to increase the quantity and diversity of image data that can be used for training;

[0033] Step S503: Integrate the amplified image data and uniformly store them as a standardized target detection model training data set.

[0034] Furthermore, the present invention adopts the following technical solutions:

[0035] A construction site-oriented image dataset construction system, comprising:

[0036] Data acquisition module: This module uses high-definition cameras and binocular cameras to obtain a wide range of video streams of personnel, materials, and mechanical equipment during the construction process, and uses mobile phones to take relevant pictures or videos in areas not covered by monitoring to obtain image data of special engineering construction sites under various time periods, areas, viewing angles, and lighting conditions;

[0037] Data preprocessing module: This module configures the OpenCV library in the Python environment, establishes connections with high-definition cameras, binocular cameras, and mobile devices deployed on site, integrates images of construction site personnel, materials, and mechanical equipment captured from different devices, and normalizes and converts these images to standardize the data format;

[0038] Data annotation module: This module uses the built YOLOv5 target detection platform to perform preliminary annotations on standardized images, and then uses the data annotation tool CVAT to further improve the detailed annotation of the categories, locations, and trajectory information of on-site personnel, materials, and mechanical equipment targets;

[0039] Data quality check module: This module uses OpenCV functions to read and compare annotated images with original images to promptly detect and correct images with annotation errors or deviations from the sample. Quality assurance personnel will conduct a final review of the format, accuracy, and consistency of the annotated images to ensure that the model dataset used for training is consistent, accurate, and complete;

[0040] Data augmentation module: For images that have passed the data annotation quality inspection, this module uses OpenCV functions to perform data augmentation on the images to improve the robustness and generalization ability of the pre-training model.

[0041] Based on the above scheme, the present invention has the following beneficial effects compared with the prior art:

[0042] 1. Based on computer vision technology and deep learning technology, the present invention collects images of personnel, materials, and mechanical equipment at the construction site in real time, normalizes the images based on the built Python-OpenCV environment, and adjusts the eigenvalues ​​of different dimensions to a similar range so as to achieve smooth convergence during subsequent model training;

[0043] 2. Based on the Python-OpenCV-YOLOv5 environment, the normalized images are preliminarily labeled according to the set labels of personnel, materials, and mechanical equipment, and the labeled results and attached files are imported into the CVAT interface for secondary labeling, which reduces the number of unlabeled images and avoids the waste of field data to a certain extent;

[0044] 3. Use OpenCV's function library to assist quality assurance personnel in evaluating the quality of labeled data, ensuring that the image annotations involved in subsequent model training are accurate, complete, and clear with minimal staff input;

[0045] 4. Finally, the functions of OpenCV are used to amplify the labeled image data, which increases the robustness and generalization ability of the training model and further improves the utilization rate of field data. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 is a flow chart of a method for constructing an image dataset for a construction site according to the present invention;

[0047] Figure 2It is a schematic diagram of a construction site-oriented image data set construction system of the present invention. DETAILED DESCRIPTION

[0048] The technical solution of the present invention will be described in detail below in conjunction with the accompanying drawings. Obviously, the embodiments described herein are only used to explain the present invention, not to limit it. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0049] like Figure 1 As shown, the present invention provides a method for constructing an image dataset for a construction site, and the present invention adopts the following technical solutions to solve the technical problems therein:

[0050] Step S1, data collection: according to the special engineering construction organization design plan and the preset monitoring coverage area of ​​the construction site environment, high-definition cameras and binocular cameras are used within the determined monitoring coverage to obtain video streams of personnel, materials, and mechanical equipment during the construction process, and mobile phones are used to take pictures or videos related to the construction process in areas not covered by the monitoring, so as to obtain image data of multiple different time periods, areas, viewing angles, and lighting conditions;

[0051] ① According to the special engineering construction organization design plan and the construction site environment, clarify the entry of personnel, materials, and mechanical equipment in the construction scene and their possible activities on the site, and preliminarily select the layout locations of binocular cameras and high-definition cameras;

[0052] ② Arrange the binocular camera and high-definition camera at the initially selected position, and configure the debugging and operating environment to obtain the video stream at that position. If the video is too dark, exposed, or blocked, adjust the position of the binocular camera and high-definition camera until a clear and complete video stream can be obtained;

[0053] ③ Through the arranged binocular cameras and high-definition cameras, the video stream of personnel, materials, and mechanical equipment at the special engineering construction site is obtained. At the same time, the mobile phone is used to capture relevant images or videos of the rest of the area, so that the training image data set covers the entire construction area.

[0054] Step S2, data preprocessing: configure the OpenCV library in the Python environment, establish connections with the high-definition cameras, binocular cameras and mobile devices arranged on site, integrate the images of construction site personnel, materials, and mechanical equipment captured from different devices, and perform unified normalization and type conversion on these images to achieve standardization of data format;

[0055] ① Configure the computer vision library - OpenCV in Python, and test the built environment to ensure that the computer vision library is successfully introduced;

[0056] ② In the Python-OpenCV environment, using the video operation function cv2.VideoCapture() as the interface, by inputting the serial numbers of the on-site HD camera and binocular camera, the device index of the HD camera and binocular camera is obtained to realize the call of the external camera and camera;

[0057] ③ Through the cv2.VideoCapture().read() function, skip frames from the video stream obtained from the high-definition camera and the binocular camera to read the images of personnel, materials, and mechanical equipment at the construction site and store them;

[0058] ④ Normalize the images read from the high-definition camera and the binocular camera, and adjust the pixel values ​​to the interval (0, 1) to ensure smooth convergence during subsequent data set training.

[0059] Step S3, data labeling: On the built YOLOv5 target detection platform, perform preliminary target detection and labeling on the standardized images, identify the target categories of on-site personnel, materials, and mechanical equipment, and then use the CVAT data labeling tool to further improve the labeling of these targets, including accurately labeling the target category, location, and trajectory information;

[0060] ① Set labels for personnel, materials, and mechanical equipment at special engineering construction sites, and use corresponding professional terms to distinguish the static and dynamic attributes of these objects. For example, set the static attribute to excavator, and set the dynamic attributes to excavation, loading, unloading, and transfer;

[0061] ② Based on the YOLOv5 target detection platform and the set labels, pre-label the target and generate the corresponding txt labeling file, and save the labeling results in a format that can be read by CVAT;

[0062] ③ Download, install and start the desktop development environment - Docker Desktop in Windows system, and start the system's built-in virtual machine - Hyper-V, download and install the command line tool - Git Bash Here, enter the commands "git clone https: / / github.com / opencv / cvat" and "cd cvat" in its window to obtain the cvat file, and modify the dockerfile file in the cvat file, then add the image source in the desktop development environment Docker engine, return to the Git Bash Here window and enter the command "CVAT_VERSION = dev docker-compose up -d", ensure that the webpage http: / / localhost:8080 / can be opened normally, and then enter the command "winpty docker exec -it cvat_serverbash -ic'python3~ / manage.py createsuperuser'" to create a super administrator, return to the browser to log in to CVAT, and complete the installation of the annotation tool - CVAT in the Windows system;

[0063] ④ Enter the configured CVAT interface, create a labeling task according to the interface's operating requirements, and select YOLO1.1 format to upload the corresponding ZIP file;

[0064] ⑤ Return to the task panel and conduct a preliminary quality check on the displayed annotation results.

[0065] Step S4, data quality check: Use OpenCV functions to read and compare the annotated image and the original image to check and identify errors in the annotation or deviations from the sample. The quality assurance personnel will conduct a final review of the format, accuracy and fit of the annotated images to ensure that all image annotations used for model training are consistent, accurate and complete;

[0066] ① In the Python-OpenCV environment, export the annotated image from the CVAT server through the cvat_export_annotations function, and use OpenCV to identify the annotated image and the original image to ensure that the content in the image is accurately annotated, and then store the correctly annotated and incorrectly annotated images separately;

[0067] ② The quality assurance personnel call up the images that have been checked and annotated correctly, check the shape and contour of the annotations, ensure that the image data used for training is not missing or cut off, and then store the images with complete and missing annotations separately;

[0068] ③ Call the images that have been checked and annotated correctly and completely, check the clarity of the annotation boundaries, ensure that the annotation boundaries are clear, and finally store the qualified and unqualified files separately.

[0069] Step S5, data enhancement: For the annotated images that have passed the quality review, data augmentation is performed through OpenCV functions to improve the robustness and generalization ability of the model.

[0070] ① For the images with accurate and complete annotations, wide coverage and high credibility found in the quality inspection, the original images are scale-matched according to the target size in the Python-OpenCV environment to narrow the gap with the target;

[0071] ② Scale, flip, rotate, and crop the matched images to expand the existing image data and increase the amount of image data that can be used for training;

[0072] ③Integrate the amplified image data and store them uniformly as a data set that can be used for training the target detection model.

[0073] like Figure 2 As shown, the present invention provides a method for constructing an image data set for a construction site, characterized in that it includes: a data acquisition module, a data preprocessing module, a data annotation module, a data quality inspection module, and a data enhancement module;

[0074] The data acquisition module uses high-definition cameras and binocular cameras to obtain video streams of personnel, materials, and mechanical equipment during the entire construction process. In areas not covered by the monitoring system, mobile phones are used to take relevant pictures or videos, so as to obtain a large amount of image data of special engineering construction sites in different time periods, regions, viewing angles, and lighting conditions.

[0075] The data preprocessing module configures OpenCV in Python to establish connections with high-definition cameras, binocular cameras, and mobile devices deployed on site, integrates images of construction site personnel, materials, and mechanical equipment captured from different devices, and normalizes and converts these images to standardize their formats.

[0076] The data annotation module uses the built YOLOv5 target detection platform to perform preliminary annotation on standardized images, and then uses the data annotation tool - CVAT to further improve the annotation of the target category, location and trajectory information of on-site personnel, materials, and mechanical equipment;

[0077] The data quality check module uses OpenCV functions to read and identify the annotated images and original images, and promptly discovers and corrects images with annotation errors or deviations from the sample. The quality assurance personnel conduct a final review of the format and fit of the annotated images to ensure that the pre-trained models are consistent, accurate, and complete.

[0078] The data augmentation module uses OpenCV functions to amplify images with qualified data annotation quality to improve the robustness and generalization ability of the pre-training model.

[0079] In summary, the image dataset construction method for construction sites comprehensively uses computer vision and machine learning technologies to form an image dataset construction process that integrates construction site image acquisition, image preprocessing, image data annotation, data annotation quality inspection and image data enhancement. It makes full use of on-site data, gradually improves the annotation quality of image data, and provides sufficient and reliable data support for subsequent model training.

[0080] The above technical solutions are only used to illustrate the implementation of the present invention, and are not intended to limit the present invention. For those skilled in the art, several modifications, equivalent substitutions, improvements, etc. can be made based on the concept of the present invention, which are all included in the protection scope of the present invention.

Claims

1. A method for constructing an image dataset for a construction site, characterized in that: The following steps are involved: Step S1, data collection: according to the special engineering construction organization design plan and the preset monitoring coverage area of ​​the construction site environment, high-definition cameras and binocular cameras are used within the determined monitoring coverage to obtain video streams of personnel, materials, and mechanical equipment during the construction process, and mobile phones are used to take pictures or videos related to the construction process in areas not covered by the monitoring, so as to obtain image data of multiple different time periods, areas, viewing angles, and lighting conditions; Step S2, data preprocessing: configure the OpenCV library in the Python environment, establish connections with the high-definition cameras, binocular cameras and mobile devices arranged on site, integrate the images of construction site personnel, materials, and mechanical equipment captured from different devices, and perform unified normalization and type conversion on these images to achieve standardization of data format; Step S3, data labeling: On the built YOLOv5 target detection platform, perform preliminary target detection and labeling on the standardized images, identify the target categories of on-site personnel, materials, and mechanical equipment, and then use the CVAT data labeling tool to further improve the labeling of these targets, including accurately labeling the target category, location, and trajectory information; Step S4, data quality check: Use OpenCV functions to read and compare the annotated image and the original image to check and identify errors in the annotation or deviations from the sample. The quality assurance personnel will conduct a final review of the format, accuracy and fit of the annotated images to ensure that all image annotations used for model training are consistent, accurate and complete; Step S5, data enhancement: For the annotated images that have passed the quality review, data augmentation is performed through OpenCV functions to improve the robustness and generalization ability of the model.

2. A method for constructing an image dataset for a construction site as claimed in claim 1, characterized in that: Step S1 includes the following steps: Step S101, according to the special engineering construction organization design plan and the construction site environment, clarify the entry of personnel, materials, and mechanical equipment in the construction scene and their possible activities on the site, and preliminarily select the layout positions of binocular cameras and high-definition cameras; Step S102: Arrange a binocular camera and a high-definition camera at the initially selected position, debug the equipment and configure the operating environment, obtain the video stream at the position, and if the video is too dark, exposed or blocked, adjust the positions of the binocular camera and the high-definition camera until a clear and complete video stream is obtained; Step S103, using the arranged binocular camera and high-definition camera, obtain the video stream of the personnel, materials, and mechanical equipment at the special engineering construction site, and use the mobile phone to capture relevant images or videos of the uncovered area to ensure that the training image data set covers the entire construction area.

3. The method for constructing an image dataset for a construction site according to claim 1, characterized in that: Step S2 includes the following steps: Step S201, configuring the computer vision library OpenCV in the Python environment, and testing the built environment to ensure that the computer vision library is successfully introduced; Step S202, in the Python-OpenCV environment, using the video operation function cv2.VideoCapture() as an interface, by inputting the device serial numbers of the on-site high-definition camera and the binocular camera, obtaining the device index, and implementing the call of these external devices; Step S203, through the cv2.VideoCapture().read() function, skip frames to read the images of personnel, materials, and mechanical equipment at the construction site from the video stream obtained by the high-definition camera and the binocular camera, and store them; Step S204, normalize the images read from the high-definition camera and the binocular camera, and adjust the pixel values ​​to the interval (0, 1) to ensure smooth convergence of the model in subsequent data set training.

4. The method for constructing an image dataset for a construction site according to claim 1, characterized in that: Step S3 includes the following steps: Step S301, setting labels for personnel, materials, and mechanical equipment in the special engineering construction site, and using professional terminology to distinguish the static and dynamic attributes of these objects; Step S302, based on the YOLOv5 target detection platform and the set labels, perform preliminary target detection and target pre-labeling on the standardized image data, generate the corresponding txt labeling file and save the labeling results in a format that can be read by CVAT; Step S303, installing and configuring the desktop development environment Docker Desktop in the Windows system, enabling the virtual machine Hyper-V, and downloading and configuring CVAT; Step S304: Enter the configured CVAT interface, create a labeling task according to the interface's operating requirements, and select YOLO1.1 format to upload the corresponding ZIP file; Step S305, return to the task panel and perform quality inspection on the displayed annotation results.

5. The method for constructing an image dataset for a construction site according to claim 1, characterized in that: Step S4 includes the following steps: Step S401, in the Python-OpenCV environment, export the annotated image from the CVAT server through the cvat_export_annotations function, and use OpenCV to compare the annotated image with the original image to ensure that the content in the image is accurately annotated, and then store the correctly annotated and incorrectly annotated images separately; Step S402, the quality assurance personnel checks the correctly annotated images, verifies the shapes and contours of the annotations, ensures that the data annotations in the images used for training are not missing or cut off, and then stores the images with complete annotations and the images with missing annotations separately; Step S403, for the images that have been checked and annotated correctly and completely, check the clarity of the annotated boundaries to ensure that all annotated boundaries are clear, and finally store the qualified and unqualified files separately.

6. The method for constructing an image dataset for a construction site according to claim 1, characterized in that: Step S5 includes the following steps: Step S501, for images that have passed the quality inspection, are accurately and completely labeled, and cover different scenes and conditions, in the Python-OpenCV environment, scale matching is performed on the original images according to the target size to reduce the gap between the target and the image edge; Step S502, performing data augmentation on the scale-matched images, including scaling, flipping, rotating, and cropping operations, to increase the quantity and diversity of image data that can be used for training; Step S503: Integrate the amplified image data and uniformly store them as a standardized target detection model training data set.

7. A construction site-oriented image dataset construction system, characterized in that: include: Data acquisition module: This module uses high-definition cameras and binocular cameras to obtain a wide range of video streams of personnel, materials, and mechanical equipment during the construction process, and uses mobile phones to take relevant pictures or videos in areas not covered by monitoring to obtain image data of special engineering construction sites under various time periods, areas, viewing angles, and lighting conditions; Data preprocessing module: This module configures the OpenCV library in the Python environment, establishes connections with high-definition cameras, binocular cameras, and mobile devices deployed on site, integrates images of construction site personnel, materials, and mechanical equipment captured from different devices, and normalizes and converts these images to standardize the data format; Data annotation module: This module uses the built YOLOv5 target detection platform to perform preliminary annotations on standardized images, and then uses the data annotation tool CVAT to further improve the detailed annotation of the categories, locations, and trajectory information of on-site personnel, materials, and mechanical equipment targets; Data quality check module: This module uses OpenCV functions to read and compare annotated images with original images to promptly detect and correct images with annotation errors or deviations from the sample. Quality assurance personnel will conduct a final review of the format, accuracy, and consistency of the annotated images to ensure that the model dataset used for training is consistent, accurate, and complete; Data augmentation module: For images that have passed the data annotation quality inspection, this module uses OpenCV functions to perform data augmentation on the images to improve the robustness and generalization ability of the pre-training model.

Citation Information

Patent Citations

  • Air conditioner outdoor unit appearance detection method and system based on federal learning

    CN113379708A

  • Vehicle damage detection method and device, electronic equipment and storage medium

    CN114723945A

  • Road converse driving and traffic accident event detection method based on target detection

    CN114882393A

  • Unmanned aerial vehicle riverway pollutant inspection method and system based on improved YOLOV5 algorithm

    CN115272895A

  • Kitchen abnormal behavior identification system based on video intelligent identification technology

    CN115359408A

Cited By

  • Transformer rack acceptance inspection data set construction method and system

    CN120564076A