Unmanned aerial vehicle safety monitoring system and method for highway closure operation

Abstract

The invention relates to the technical field of traffic transportation, and discloses an unmanned aerial vehicle safety monitoring system and method for highway closure operation, and the system comprises an unmanned aerial vehicle inspection module which is used for determining an inspection target as a highway closure operation scene, planning an air route, and carrying out the periodic inspection of people, objects and equipment in a closure operation region according to the planned air route; the data storage module is used for carrying out quality evaluation and classification arrangement on the data transmitted by the unmanned aerial vehicle inspection module and storing the data to an object storage service; the image data annotation module is used for carrying out annotation processing on the stored image data; and the model training module is used for carrying out model training by utilizing the marked image data to obtain a security identification model. According to the system, an unmanned aerial vehicle inspection system and an AI image recognition technology are integrated, potential safety hazards possibly existing in highway closure operation are specially recognized, the recognition result is immediately notified to field personnel and a background system, and effective safety emergency management measures are formed.

Description

Technical Field

[0001] The present invention relates to the field of transportation technology, and in particular to an unmanned aerial vehicle safety monitoring system and method for highway road closure operations. Background Art

[0002] The existing management and technical measures for highway road closures cover a number of key links, aiming to fully ensure safety during construction. These measures include systematic identification and analysis of safety risks, the establishment of a strict construction safety monitoring and early warning system, scientific safety risk classification management and hidden danger investigation and treatment process, and detailed emergency plan formulation and implementation. These comprehensive measures together constitute a multi-level safety management system to deal with various possible safety hazards and emergencies.

[0003] However, although these management and technical measures have covered many aspects of highway closure safety, some significant problems still exist.

[0004] Among them, the efficiency of identifying safety risks is low, and the entire identification process is too cumbersome and complicated. In addition, the existing management system relies too much on manual inspections, which not only increases labor costs, but may also lead to inaccurate identification or omissions due to human factors. Summary of the invention

[0005] In view of the deficiencies in the prior art, the present invention proposes a drone safety monitoring system and method for highway road closure operations to solve the above-mentioned technical problems.

[0006] In the first aspect, a drone safety monitoring system for highway road closure operations is provided, comprising:

[0007] The drone inspection module is used to determine the inspection target as the highway road closure operation scene, plan the route, and conduct periodic inspections of people, objects, and equipment in the road closure operation area according to the planned route, capture dynamic information in real time, and transmit the collected data to the drone management platform in real time;

[0008] The data storage module is used to evaluate the quality of the data transmitted by the drone inspection module, classify and organize it, and store it in the object storage service;

[0009] An image data annotation module is used to annotate the stored image data;

[0010] The model training module uses the annotated image data to train the model based on the yolov5x algorithm to obtain a security recognition model;

[0011] Wherein, the model training module is configured with preprocessing technology to improve training speed and accuracy;

[0012] The model integration and application module is used to integrate the trained security recognition model into the system, obtain image data from the object storage server, perform image recognition, and calculate the recognition results. The recognition results and model inference results are stored in the database of the application server for application service calls and image annotation.

[0013] Furthermore, the drone inspection module also includes:

[0014] The route planning submodule uses the drone planning software to design the route, determine the take-off and landing points, and consider the path, altitude, direction, turning radius, and waypoint density when planning the route;

[0015] The data acquisition submodule is used to start the UAV under safe conditions, fly according to the planned route, monitor the flight status and collected data of the UAV in real time, and adjust the flight altitude, speed and direction of the UAV according to the real-time monitoring situation.

[0016] Furthermore, the image data annotation module also includes:

[0017] In the labeling tool selection submodule, LabelImg is selected as the image data labeling tool;

[0018] The annotation process submodule is used to perform preliminary annotation on the image and adjust the preliminary annotation results.

[0019] Furthermore, the model training module also includes:

[0020] The network structure configuration submodule includes four parts: input end, backbone network, Neck network and Prediction output layer;

[0021] The parameter setting submodule is used to set the input parameters of the model;

[0022] The data set division submodule is used to divide the data set into a training set and a test set according to a predetermined ratio.

[0023] Furthermore, the model integration and application module also includes:

[0024] The drone platform building and configuration submodule is used to interact and communicate with the drone, instruct the drone to store image data to the specified object storage server, and obtain the structured information of the image data;

[0025] The object storage service builds and configures the submodule, which is used to store image data and configure the corresponding access permissions and port numbers;

[0026] The security identification model deployment and configuration submodule is used to package the model and its operating environment into a container image and deploy it to the AI ​​management platform.

[0027] In a second aspect, a method for drone safety monitoring of highway road closure operations is provided, based on a drone safety monitoring system for highway road closure operations as described in any one of the preceding texts, comprising the following steps:

[0028] Step S01: Use drones to periodically inspect highway road closure operation scenes, capture dynamic information in real time, and transmit the collected data to the drone management platform in real time;

[0029] Step S02: Perform a preliminary quality assessment on the transmitted data, sort and store it in the object storage service;

[0030] Step S03, annotating the stored image data, including determining the annotation task and object, selecting the annotation tool, performing the annotation process, and data cleaning and exporting;

[0031] Step S04: Based on the yolov5x algorithm, the labeled image data is used for model training to obtain a security recognition model;

[0032] Step S05: Integrate the trained security recognition model into the system, obtain image data from the object storage server, perform image recognition, calculate the recognition result, and store the compliance check result and model inference result in the database of the application server.

[0033] Furthermore, step S01 also includes:

[0034] Use drone planning software to design routes and determine take-off and landing points;

[0035] Start the drone under safe conditions, fly it according to the planned route, and monitor the drone's flight status and collected data in real time.

[0036] Furthermore, step S03 also includes:

[0037] Select LabelImg as the image data annotation tool, and select yolo as the mode;

[0038] Perform preliminary annotation on the image and adjust the preliminary annotation results.

[0039] Furthermore, step S04 also includes:

[0040] The yolov5x algorithm is used to configure the network structure, set the input parameters of the model, and divide the data set into a training set and a test set according to a predetermined ratio.

[0041] Furthermore, step S05 also includes:

[0042] Build the drone platform, configure the address and port number of the drone platform, and instruct the drone to store image data to the specified object storage server;

[0043] Build an object storage service and configure the corresponding access permissions and port numbers;

[0044] Package the model and its operating environment into a container image and deploy it to the AI ​​management platform.

[0045] The invention adopting the above technical solution has the following advantages:

[0046] 1. The present invention integrates the drone inspection system and AI image recognition technology to specifically identify potential safety hazards in highway road closure operations, and immediately notifies on-site personnel and back-end systems of the identification results, forming effective safety emergency management measures.

[0047] 2. The present invention is specifically designed for the AI ​​image recognition algorithm model of safety hazards during highway road closure operations. The algorithm model can automatically and efficiently identify various safety hazards at the work site. The model can identify safety cones and their corresponding status, whether personnel are wearing safety helmets, whether they are wearing safety clothing, and whether the distance between cones and other signs and signs meets safety construction standards and specifications. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the specific implementation of the present invention, the following will briefly introduce the drawings required for use in the specific implementation. In all the drawings, each element or part is not necessarily drawn according to the actual scale.

[0049] Figure 1 This is a flow chart of a drone safety monitoring system and method for highway road closure operations according to the present invention;

[0050] Figure 2 A technical roadmap for a drone safety monitoring system and method for highway road closure operations according to the present invention;

[0051] Figure 3 A drone planning route map in a drone safety monitoring system for highway road closure operations according to the present invention;

[0052] Figure 4 This is a demonstration of the labeling tool in the drone safety monitoring system for highway road closure operations of the present invention. Figure 1 ;

[0053] Figure 5 This is a demonstration of the labeling tool in the drone safety monitoring system for highway road closure operations of the present invention. Figure 2 ;

[0054] Figure 6This is a network structure diagram of yolov5x in a drone safety monitoring system and method for highway road closure operations of the present invention;

[0055] Figure 7 A statistical diagram of the label categories and quantities of a training data set in a drone safety monitoring system and method for highway road closure operations according to the present invention;

[0056] Figure 8 A confusion matrix diagram of a model training loss function diagram in a drone safety monitoring system and method for highway road closure operations of the present invention;

[0057] Fig. 9 This is a model test result diagram of a drone safety monitoring system and method for highway road closure operations according to the present invention;

[0058] Fig.10 This is a diagram of detection results in a drone safety monitoring system and method for highway road closure operations according to the present invention;

[0059] Fig.11 A flowchart of the model integration in a drone safety monitoring system and method for highway road closure operations according to the present invention;

[0060] Fig.12 This is an image information diagram of an object storage service in a drone safety monitoring system and method for highway road closure operations of the present invention;

[0061] Fig.13 This is an image information diagram of a drone platform in a drone safety monitoring system and method for highway road closure operations of the present invention;

[0062] Fig.14 This is a display diagram of a database in a drone safety monitoring system and method for highway road closure operations according to the present invention;

[0063] Fig.15 This is a diagram combining images and information in a drone safety monitoring system and method for highway road closure operations according to the present invention. DETAILED DESCRIPTION

[0064] The following embodiments of the technical solution of the present invention are described in detail in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and are therefore only used as examples, and cannot be used to limit the protection scope of the present invention.

[0065] like Figure 1 to Figure 15 As shown, a drone safety monitoring system for highway road closure operations of the present invention comprises:

[0066] The drone inspection module is used to determine the inspection target as the highway road closure operation scene, plan the route, and conduct periodic inspections of people, objects, and equipment in the road closure operation area according to the planned route, capture dynamic information in real time, and transmit the collected data to the drone management platform in real time;

[0067] The data storage module is used to evaluate the quality of the data transmitted by the drone inspection module, classify and organize it, and store it in the object storage service;

[0068] An image data annotation module is used to annotate the stored image data;

[0069] The model training module uses the annotated image data to train the model based on the yolov5x algorithm to obtain a security recognition model;

[0070] Among them, the model training module is equipped with preprocessing technology to improve training speed and accuracy;

[0071] The model integration and application module is used to integrate the trained security recognition model into the system, obtain image data from the object storage server, perform image recognition, and calculate the recognition results. The recognition results and model inference results are stored in the database of the application server for application service calls and image annotation.

[0072] In some embodiments, the drone inspection module further includes:

[0073] The route planning submodule uses the drone planning software to design the route, determine the take-off and landing points, and consider the path, altitude, direction, turning radius, and waypoint density when planning the route;

[0074] The data acquisition submodule is used to start the UAV under safe conditions, fly according to the planned route, monitor the flight status and collected data of the UAV in real time, and adjust the flight altitude, speed and direction of the UAV according to the real-time monitoring situation.

[0075] In some embodiments, the image data annotation module further includes:

[0076] In the labeling tool selection submodule, LabelImg is selected as the image data labeling tool;

[0077] The annotation process submodule is used to perform preliminary annotation on the image and adjust the preliminary annotation results.

[0078] In some embodiments, the model training module further includes:

[0079] The network structure configuration submodule includes four parts: input end, backbone network, Neck network and Prediction output layer;

[0080] The parameter setting submodule is used to set the input parameters of the model;

[0081] The data set division submodule is used to divide the data set into a training set and a test set according to a predetermined ratio.

[0082] In some embodiments, the model integration and application module further includes:

[0083] The drone platform building and configuration submodule is used to interact and communicate with the drone, instruct the drone to store image data to the specified object storage server, and obtain the structured information of the image data;

[0084] The object storage service builds and configures the submodule, which is used to store image data and configure the corresponding access permissions and port numbers;

[0085] The security identification model deployment and configuration submodule is used to package the model and its operating environment into a container image and deploy it to the AI ​​management platform.

[0086] In other embodiments, a method for drone safety monitoring of highway road closure operations is provided, and a drone safety monitoring system for highway road closure operations based on any one of the above items includes the following steps:

[0087] Step S01: Use drones to periodically inspect highway road closure operation scenes, capture dynamic information in real time, and transmit the collected data to the drone management platform in real time;

[0088] Step S02: Perform a preliminary quality assessment on the transmitted data, sort and store it in the object storage service;

[0089] Step S03, annotating the stored image data, including determining the annotation task and object, selecting the annotation tool, performing the annotation process, and data cleaning and exporting;

[0090] Step S04: Based on the yolov5x algorithm, the labeled image data is used for model training to obtain a security recognition model;

[0091] Step S05: Integrate the trained security recognition model into the system, obtain image data from the object storage server, perform image recognition, calculate the recognition result, and store the compliance check result and model inference result in the database of the application server.

[0092] In some embodiments, step S01 further includes:

[0093] Use drone planning software to design routes and determine take-off and landing points;

[0094] Start the drone under safe conditions, fly it according to the planned route, and monitor the drone's flight status and collected data in real time.

[0095] Specifically, drone inspection and data collection and transmission:

[0096] The process includes determining inspection targets, planning routes, data collection, image transmission, and image storage. The drone conducts periodic inspections of the target area through automated route planning, captures dynamic information in real time, and transmits the data back to the control center. Afterwards, professional software is used to conduct preliminary analysis of the collected images and data.

[0097] Determine inspection targets: The inspection targets are determined as highway road closure operation scenarios, and inspections and checks are carried out on people, objects, equipment, etc. within the road closure operation area.

[0098] Plan the route: Use professional drone planning software to design the route and determine the take-off and landing points. When planning the route, you should consider factors such as path, altitude, direction, turning radius, and waypoint density.

[0099] Data collection: Start the drone under safe conditions, fly according to the planned route, monitor the drone's flight status and collected data in real time, and adjust the drone's flight altitude, speed and direction based on real-time monitoring to ensure data quality.

[0100] Data transmission: The collected data is transmitted to the drone management platform in real time for preliminary inspection to confirm whether the data of all scheduled collection points are complete and ensure that the data is safely backed up in the drone's internal storage device.

[0101] Data storage: Perform preliminary quality assessment on the collected data, such as image clarity, coverage, etc., classify and organize the collected data, and store it in the object storage service.

[0102] In some embodiments, step S03 further includes:

[0103] Select LabelImg as the image data annotation tool, and select yolo as the mode;

[0104] Perform preliminary annotation on the image and adjust the preliminary annotation results.

[0105] Specifically, image data annotation:

[0106] The labeling step in image recognition training is crucial because high-quality labeled data can significantly improve the effect of model training. The specific technical steps are as follows:

[0107] Determine the labeling tasks and objects: By studying the "Risk Analysis of Highway Maintenance Operations", the categories and objects that need to be labeled in the images are determined as shown in the following table:

[0108] Table 1

[0109]

[0110] Select the annotation tool: LabelImg is selected as the image data annotation tool, and yolo is selected as the mode.

[0111] Labeling process: Start preliminary labeling of the image, such as drawing bounding boxes, segmentation areas, classification labels, etc., and make detailed adjustments to the preliminary labeling results to ensure that the labeling box accurately covers the target object. For tasks that require fine segmentation, use polygon tools for precise labeling.

[0112] Data cleaning and export: Clear incorrectly labeled data, remove duplicate or low-quality images, and export the labeled data corresponding to the image as XML format data.

[0113] In some embodiments, step S04 further includes:

[0114] The yolov5x algorithm is used to configure the network structure, set the input parameters of the model, and divide the data set into a training set and a test set according to a predetermined ratio.

[0115] Specifically, model training:

[0116] The training environment of this recognition model is a computer server installed on the Windows 10 operating system, with an Intel i5-12600K CPU and an NVIDIA GeForce RTX 3060 GPU. The programming language is Python 3.8, using the PyTorch 1.13.1 deep learning framework and OpencvCV 4.7 and other software libraries.

[0117] This model is trained based on the yolov5x algorithm, which mainly consists of four parts: input, backbone network, Neck network and Prediction output layer.

[0118] After the training image is input into the network through the input end, the data is first preprocessed, that is, the input image uses Mosaic data enhancement, adaptive anchor box calculation and adaptive image scaling technologies to improve the training speed and accuracy of the model. YOLOv5x uses CSPDarknet53 as the backbone network, combines the Focus structure and CSP structure, and uses the FPN+PAN structure in the Neck part to achieve feature extraction and classification of the detection target.

[0119] In the Prediction output layer, YOLOv5x uses binary cross entropy as the classification loss function and GIoULoss as the bounding box regression loss function, and uses the weighted non-maximum suppression (Weighted NMS) algorithm to eliminate redundant boxes. In the process of rectangular box elimination, Weighted NMS does not directly eliminate those bounding boxes whose IoU with the current rectangular box is greater than the threshold and the same category, but weights them according to the confidence of the network prediction to obtain a new bounding box, which is used as the final predicted bounding box, and then those redundant bounding boxes are eliminated. The specific network structure of the yolov5x model.

[0120] The input parameters of the facility model: This model is trained using the yolov5x algorithm, and the initial weight file is yolov5x. When training the model, the batch size is set to 4, the image resolution imgsz during training is changed to 1280, and other parameters are default. The model is iterated 200 times in total.

[0121] Use the data set to train the model: Before model training, the data set is divided into a training set and a test set in a ratio of 9:1. The model training time is about 52 hours. The model training process is shown in the figure below, which includes the label category and quantity statistics of the training set input by the initial model output during training; the loss function loss curve image of the model in the training set and the test set during the model training process, as well as the model's recognition accuracy Precision and category average precision mAP0.5 change curve; and some test results of the test set after the model training is completed.

[0122] Training result image output: Input the image related to the model, call the model parameter interface, and obtain the image output result.

[0123] In some embodiments, step S05 further includes:

[0124] Build the drone platform, configure the address and port number of the drone platform, and instruct the drone to store image data to the specified object storage server;

[0125] Build an object storage service and configure the corresponding access permissions and port numbers;

[0126] Package the model and its operating environment into a container image and deploy it to the AI ​​management platform.

[0127] Specifically, the model ensemble:

[0128] After the model is trained, the images collected by the drone are stored in the object storage server through the network. The trained model obtains image data from the object storage server. After successful image recognition, the recognition results of the image are stored. For example, after successful recognition of safety cones, it is necessary to calculate the distance between each cone to determine whether the placement of the cones is compliant. The compliance check results and model inference results are stored in the database of the application server. The relevant application services retrieve the image data from the object storage service based on the recognition results and annotate the images.

[0129] UAV platform construction and configuration:

[0130] The main tasks of the drone platform include data exchange and communication with drones, ensuring that the background can monitor the movements of drones in real time, instructing drones to store image data in designated object storage servers, obtaining structured information of image data, communicating with image recognition models to initiate image recognition requests, etc.

[0131] The drone platform is built with a front-end and back-end separation design. The front-end uses the TS+Vue3 framework, the back-end uses the JAVA language and Spring Boot framework, the database is equipped with Mysql, and the server is the Ubuntu16.04 system.

[0132] First, you need to apply for the appId, appKey and appLicense of the drone platform, and apply for the amapKey used for the map; then, configure the applied appId, appKey, appLicense, and amapKey in the configuration file of the front-end program, and start the front-end program; then, configure the relevant mysql, mqtt, redis and object storage service parameters in the configuration file of the back-end (src / main / resources / application.yml), and start the service; finally, perform relevant operations on the drone, configure the address and port number of the drone platform, and when the following figure appears on the drone operation panel, it means that the drone platform has been successfully built and can be successfully connected to the drone.

[0133] The main parameter configuration of the UAV platform is shown in the following table:

[0134] Table 2

[0135]

[0136] Object storage service construction and configuration:

[0137] The main tasks of the object storage service include storing image data, allowing secure image recognition models and specific application services to obtain and call corresponding image data, etc.

[0138] Object storage is built using MinIO, which is an open source, high-performance object storage service that can be used to store large amounts of unstructured data, such as pictures, videos, log files, etc. The object storage server uses the Ubuntu 16.04 system and is deployed through binary files. After deployment, port 9000 is configured as the image data transmission port, and port 9099 is configured as the service background management port. At the same time, in order for the drone platform, secure image recognition model, and application service to access the image, it is necessary to generate the corresponding bucket and the ak and sk for bucket access to ensure the security of image transmission and acquisition. When the access address + port 9099 is logged in and the following interface appears, it means that the object storage service has been successfully built.

[0139] Deployment and configuration of security identification model:

[0140] The model needs to rely on the corresponding environment and platform to run, and be able to access the network for the drone management platform to access and call. Here we use the AI ​​management platform of China Merchants Cloud to deploy the model.

[0141] The first step is to package the model and the running environment into a container image.

[0142] Upload container image files to the AI ​​management platform.

[0143] After uploading the model image successfully, you need to run the model. By setting parameters such as the port number, the model image displays the ready status, indicating that the model has been successfully run.

[0144] Joint debugging and testing:

[0145] Bring the drone to the actual site of the highway, operate the drone to take off, and take pictures.

[0146] After the drone flight is completed, check the object storage server and drone management platform and find that the captured pictures have been uploaded to the platform.

[0147] The security recognition model extracts and calculates image information, and the calculated image structured information can be stored in the corresponding database.

[0148] The application obtains the image from the object storage server and the recognition information from the database, and combines them to form the recognition result shown in the figure.

[0149] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein by equivalents. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present invention, and they should all be included in the scope of the claims and specification of the present invention.

Claims

1. A drone safety monitoring system for highway road closure operations, characterized in that: include: The drone inspection module is used to determine the inspection target as the highway road closure operation scene, plan the route, and conduct periodic inspections of people, objects, and equipment in the road closure operation area according to the planned route, capture dynamic information in real time, and transmit the collected data to the drone management platform in real time; The data storage module is used to evaluate the quality of the data transmitted by the drone inspection module, classify and organize it, and store it in the object storage service; An image data annotation module is used to annotate the stored image data; The model training module uses the annotated image data to train the model based on the yolov5x algorithm to obtain a security recognition model; Wherein, the model training module is configured with preprocessing technology to improve training speed and accuracy; The model integration and application module is used to integrate the trained security recognition model into the system, obtain image data from the object storage server, perform image recognition, and calculate the recognition results. The recognition results and model inference results are stored in the database of the application server for application service calls and image annotation.

2. The unmanned aerial vehicle safety monitoring system for highway road closure operations according to claim 1 is characterized in that: The drone inspection module also includes: The route planning submodule uses the drone planning software to design the route, determine the take-off and landing points, and consider the path, altitude, direction, turning radius, and waypoint density when planning the route; The data acquisition submodule is used to start the UAV under safe conditions, fly according to the planned route, monitor the flight status and collected data of the UAV in real time, and adjust the flight altitude, speed and direction of the UAV according to the real-time monitoring situation.

3. The unmanned aerial vehicle safety monitoring system for highway road closure operations according to claim 2 is characterized in that: The image data annotation module also includes: In the labeling tool selection submodule, LabelImg is selected as the image data labeling tool; The annotation process submodule is used to perform preliminary annotation on the image and adjust the preliminary annotation results.

4. The unmanned aerial vehicle safety monitoring system for highway road closure operations according to claim 3 is characterized in that: The model training module also includes: The network structure configuration submodule includes four parts: input end, backbone network, Neck network and Prediction output layer; The parameter setting submodule is used to set the input parameters of the model; The data set division submodule is used to divide the data set into a training set and a test set according to a predetermined ratio.

5. The drone safety monitoring system for highway road closure operations according to claim 1 is characterized in that: The model integration and application module also includes: The drone platform building and configuration submodule is used to interact and communicate with the drone, instruct the drone to store image data to the specified object storage server, and obtain the structured information of the image data; The object storage service builds and configures the submodule, which is used to store image data and configure the corresponding access permissions and port numbers; The security identification model deployment and configuration submodule is used to package the model and its operating environment into a container image and deploy it to the AI ​​management platform.

6. A method for drone safety monitoring of highway road closure operations, characterized in that: A drone safety monitoring system for highway road closure operations based on any one of claims 1 to 5 comprises the following steps: Step S01: Use drones to periodically inspect highway road closure operation scenes, capture dynamic information in real time, and transmit the collected data to the drone management platform in real time; Step S02: Perform a preliminary quality assessment on the transmitted data, sort and store it in the object storage service; Step S03, annotating the stored image data, including determining the annotation task and object, selecting the annotation tool, performing the annotation process, and data cleaning and exporting; Step S04: Based on the yolov5x algorithm, the labeled image data is used for model training to obtain a security recognition model; Step S05: Integrate the trained security recognition model into the system, obtain image data from the object storage server, perform image recognition, calculate the recognition result, and store the compliance check result and model inference result in the database of the application server.

7. The method for safety monitoring of highway road closure by drone according to claim 6, characterized in that: Step S01 also includes: Use drone planning software to design routes and determine take-off and landing points; Start the drone under safe conditions, fly it according to the planned route, and monitor the drone's flight status and collected data in real time.

8. The method for safety monitoring of highway road closure by drone according to claim 7, characterized in that: Step S03 also includes: Select LabelImg as the image data annotation tool, and select yolo as the mode; Perform preliminary annotation on the image and adjust the preliminary annotation results.

9. The method for safety monitoring of highway road closure by unmanned aerial vehicle according to claim 8, characterized in that: Step S04 also includes: The yolov5x algorithm is used to configure the network structure, set the input parameters of the model, and divide the data set into a training set and a test set according to a predetermined ratio.

10. The method for safety monitoring of highway road closure by drone according to claim 9, characterized in that: Step S05 also includes: Build the drone platform, configure the address and port number of the drone platform, and instruct the drone to store image data to the specified object storage server; Build an object storage service and configure the corresponding access permissions and port numbers; Package the model and its operating environment into a container image and deploy it to the AI ​​management platform.

Images