A machine vision-based personnel safety identification system and method for a steelmaking converter

Abstract

The application discloses a kind of personnel safety identification system and method based on machine vision of steelmaking converter, belong to industrial safety monitoring and machine vision application technical field, system includes data acquisition unit, processing unit, alarm unit and storage unit;The method is by deep learning algorithm self-identification converter iron, add scrap, wait for three working conditions, in high-risk operation state, automatically trigger dangerous area personnel detection, determine the instant push alarm information after illegal intrusion event, and simultaneously store full-dimension data to enterprise management platform.The application does not need to rely on traditional PLC signal, solves the problems of low efficiency of manual supervision, many false or missed judgments, early warning lag and data untraceable, realizes the full-process automation control of personnel safety in high-risk scene of converter, effectively reduces the risk of scalding, splashing and other safety accidents.

Description

Technical Field

[0001] This invention relates to the field of industrial safety monitoring and machine vision application technology, and in particular to a machine vision-based safety identification system and method for personnel in steelmaking converters. Background Technology

[0002] Steelmaking converters are core equipment in steel production. Key operations such as molten iron charging and scrap steel addition are characterized by high temperatures, high dust levels, and strong radiation, making the work area an extremely high-risk location. If personnel illegally enter the dangerous area around the converter opening, it can easily lead to serious safety accidents such as scalding from molten iron and high-temperature splashes, causing casualties, property damage, and production interruptions, severely impacting the safe production order of steel enterprises.

[0003] Currently, steel companies mainly rely on traditional technical methods for personnel safety management in hazardous areas of converters, which has the following significant drawbacks: 1. The limitations of manual supervision: The method of fixed-point inspection by safety officers or manual video surveillance is severely affected by harsh environments such as strong light and dust, which seriously reduces human vision and judgment; in addition, the operation area is wide and there are many key points, and the limited number of safety officers cannot achieve full-time coverage without blind spots; at the same time, manual monitoring is easily affected by factors such as fatigue and distraction, resulting in delayed identification of violations and inability to issue timely warnings.

[0004] 2. Blind spots exist in relying on PLC signal triggering: Traditional automated control systems rely on PLC signals from converter equipment for triggering detection, but most old converter sites lack available PLC signal interfaces, making it impossible to implement automated control solutions.

[0005] 3. Inadequate data traceability and early warning mechanisms: Traditional methods cannot fully record the process information of violations, and there is a lack of effective data support for cause analysis and accountability after an accident occurs; moreover, early warning information needs to be reported level by level through walkie-talkies and telephones, which involves many transmission links and takes a long time, making it easy to miss the best time for handling.

[0006] Therefore, developing an intelligent safety identification system and method that does not rely on PLC signals, can autonomously identify converter operating conditions, and monitor personnel violations in real time has become a technical problem that the steel industry urgently needs to solve. Summary of the Invention

[0007] To address the technical problems of low efficiency in manual supervision, blind spots in reliance on PLC signals, untimely early warnings, and lack of data traceability in existing steelmaking converter personnel safety management technologies, this invention provides a machine vision-based steelmaking converter personnel safety identification system and method. This system enables fully automated and intelligent management of personnel safety in high-risk converter operation scenarios.

[0008] The technical solution adopted by the present invention, a machine vision-based safety identification system and method for steelmaking converter personnel, is as follows: A machine vision-based safety identification system for personnel in steelmaking converters includes, The data acquisition unit is used to collect real-time image data of the steelmaking converter operating area; The processing unit is communicatively connected to the data acquisition unit and is used to autonomously identify the working status of the converter based on deep learning algorithms. When a high-risk working status is identified, it automatically triggers personnel detection in dangerous areas to determine whether there is a personnel unauthorized intrusion incident. An alarm unit, which is communicatively connected to the processing unit, is used to generate and push standardized alarm information when an unauthorized intrusion event is determined to exist. The evidence storage unit is communicatively connected to the processing unit and is used to store full-dimensional data on unauthorized intrusion events and synchronously upload them to the enterprise-level management platform.

[0009] A further improvement of the technical solution of the present invention is that the data acquisition unit includes at least one industrial camera component that is resistant to high temperature and dust. The industrial camera component is deployed at preset monitoring points around the converter furnace mouth, the iron charging operation area and the scrap steel feeding area, and its image acquisition parameter is a frame rate of 30 frames / second.

[0010] A further improvement of the technical solution of the present invention is that the processing unit has a built-in converter operating condition identification model, which can autonomously distinguish the three working states of the converter: iron charging operation, scrap steel adding operation, and standby waiting. The high-risk working states include the iron charging operation state and the scrap steel adding operation state.

[0011] A further improvement of the technical solution of the present invention is that the processing unit also has a built-in personnel target detection model, which is used to identify the main body of the personnel in the image after triggering the detection of personnel in the dangerous area, and combined with the customizable electronic fence boundary, when the detection of personnel with both feet completely entering the electronic fence area is determined as an illegal intrusion event.

[0012] A further improvement of the technical solution of the present invention is that: the alarm unit includes an enterprise instant messaging interface module, which is used to push alarm information including event time, location, real-time image and type to a preset security management personnel terminal through an application programming interface; the evidence storage unit stores full-dimensional data including basic event attribute data and visual evidence data.

[0013] A machine vision-based method for safety identification of personnel in steelmaking converters includes the following steps: S1. Continuously collect real-time image data of the steelmaking converter operation area; S2. Based on deep learning algorithms, identify the current working status of the converter. When a high-risk working status is identified, automatically trigger the personnel detection process in the dangerous area. S3. Perform personnel target detection on the collected images and determine whether there is a personnel unauthorized intrusion event based on the preset electronic fence boundary; S4. If an unauthorized intrusion is detected, a standardized alarm message will be generated immediately and pushed to the relevant security management personnel. S5 automatically stores full-dimensional data on unauthorized intrusion events and synchronously uploads it to the enterprise-level management platform.

[0014] A further improvement of the technical solution of the present invention is that the method for identifying the working state of the converter in step S2 is as follows: collect real-time images of the converter under different working states to construct a labeled dataset, train a converter working condition identification model through a convolutional neural network algorithm, and analyze the real-time images through the model to distinguish the three working states of the converter: iron charging operation, scrap steel adding operation, and standby waiting.

[0015] A further improvement of the technical solution of the present invention is that the method for determining the unauthorized entry event in step S3 is as follows: the target detection algorithm is used to identify the main body of the person in the image and obtain its coordinate information. The coordinates of the person are compared with the preset electronic fence boundary. When it is detected that both feet of the person have entered the electronic fence area, it is determined to be an unauthorized entry event, and key frame images containing the person's position, dangerous area and work scene are captured as evidence.

[0016] A further improvement of the technical solution of the present invention is that the standardized alarm information in step S4 includes the event occurrence time, specific dangerous area, key frame image of the violator, and event type.

[0017] A further improvement of the technical solution of the present invention is that: the full-dimensional data in step S5 includes basic event attribute data and visual evidence data. The basic event attribute data includes the occurrence time, danger zone number and operation type, and the visual evidence data includes key frame images of the violation event.

[0018] The technological advancements achieved by this invention due to the adoption of the above technical solutions are as follows: This invention uses deep learning algorithms to train a converter operating condition recognition model. It does not rely on the original PLC signal of the equipment and can accurately distinguish the three working states of the converter through machine vision alone. It solves the problem of converter safety management in scenarios without PLC signal and has a wider range of applications.

[0019] This invention achieves targeted triggering of violation detection through working condition identification, avoiding misjudgment and missed judgment under non-hazardous working conditions; at the same time, the use of personnel detection algorithm can effectively eliminate interference from on-site equipment and materials.

[0020] This invention integrates the entire process of "work condition identification - personnel detection - real-time alarm - data storage"; it also automatically saves complete event data, enabling traceability and analysis of safety hazards. This invention replaces the traditional manual monitoring mode, reducing the manpower required for safety personnel and improving the efficiency and targeting of safety management. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments. In the following description, descriptions of well-known structures and technologies are omitted to avoid unnecessarily obscuring the concepts of this invention. Example 1

[0022] This embodiment discloses a machine vision-based safety identification system for steelmaking converter personnel, which mainly consists of a data acquisition unit, a processing unit, an alarm unit, and an evidence storage unit.

[0023] In this embodiment, the data acquisition unit includes four industrial cameras, corresponding high-temperature resistant protective covers, supplementary lights, and stainless steel mounting brackets. The industrial cameras support image acquisition at 30 frames per second and a resolution of 1920×1080. The protective covers are made of double-layer stainless steel and have an internal axial fan, enabling them to operate normally in ambient temperatures up to 80℃. The supplementary lights (13) are infrared supplementary lights. The four industrial cameras are respectively fixed to the top of the converter furnace opening, the north side of the iron-feeding rail, and the east and south sides of the scrap steel feeding port, covering all hazardous operating areas of the converter. The industrial cameras are connected to an industrial switch via Cat5e network cables.

[0024] In this embodiment, the processing unit includes an industrial switch and an industrial computer. The industrial switch is installed in the control cabinet of the converter operation room and is connected to four industrial cameras via network cables, and also to the industrial computer via network cables. The industrial computer is pre-installed with machine vision recognition software, which integrates an image preprocessing module, a converter operating condition recognition module, a personnel detection module, and a violation judgment module.

[0025] The converter operating condition recognition module uses a labeled dataset created from 20,000 on-site photos of the converter under different operating conditions. It is trained using the ResNet50 convolutional neural network algorithm, achieving a recognition accuracy of 98.5% for the three converter operating conditions after training. The personnel detection module uses the YOLOv8 algorithm, trained on a labeled dataset of personnel images including those with different clothing and postures, enabling rapid and accurate identification of personnel targets in complex backgrounds. The violation judgment module supports visual drawing of electronic fence boundaries through a backend management interface, allowing users to adjust the range and shape of hazardous areas according to actual operational needs.

[0026] In this embodiment, the alarm unit is a WeChat Work interface module, integrated into the machine vision recognition software, and connected to the WeChat Work server via the Internet. Users can configure the alarm recipient list in the backend management interface, supporting the addition of multiple recipients, and can set different push permissions according to the event type.

[0027] In this embodiment, the evidence storage unit includes a local storage module and a data upload module. The local storage module uses a solid-state drive built into the industrial control computer, capable of storing at least 6 months of violation event data. The data upload module connects to the company's intelligent brain platform via the enterprise intranet, enabling the real-time uploading of violation event data to the platform. Example 2

[0028] This embodiment provides a machine vision-based method for safety identification of personnel in steelmaking converters. The specific process is as follows: S1. Operating Condition Monitoring and Triggering: After system startup, industrial cameras continuously collect real-time images of the converter operating area and transmit them to the industrial control computer. The converter operating condition recognition module analyzes the images in real time to identify the current operating status of the converter. When the converter is detected to be entering the iron charging or scrap steel adding operation state, the hazardous area personnel detection process is automatically triggered.

[0029] S2. Image Acquisition and Preprocessing: After triggering personnel detection, the image preprocessing module performs noise reduction, contrast enhancement, and other processing on the acquired images to improve image quality and prepare for subsequent personnel detection.

[0030] S3. Personnel Detection and Violation Detection: The personnel detection module analyzes the pre-processed image, identifies personnel targets in the image, and obtains the personnel's coordinates. The violation detection module compares the personnel coordinates with the preset electronic fence boundary. When it detects that both of a person's feet have entered the electronic fence area, it is determined as a violation and a keyframe image containing the personnel's location, the danger zone, and the work scene is captured as evidence.

[0031] S4. Alarm Information Push: Once an unauthorized intrusion event is determined, the alarm unit immediately generates a standardized alarm notification, which includes the time of the event (accurate to the second), the specific danger area (such as "No. 1 converter iron-making area"), key frame images of the unauthorized personnel, and the event type (such as "personnel intrusion during iron-making"), and pushes it to the preset safety management personnel's mobile phone via the enterprise WeChat interface.

[0032] S5. Data Storage and Upload: The evidence storage unit saves basic information and keyframe images of violations to a local solid-state drive, while the data upload module uploads the data to the company's intelligent platform in real time. The safety supervision department can use the intelligent platform to query historical violation records, generate statistical analysis reports, and conduct hazard investigations and safety training.

[0033] In the above embodiments, a machine vision-based safety identification system and method for steelmaking converter personnel is provided. This invention uses deep learning algorithms to train a converter operating condition identification model, eliminating reliance on existing PLC signals and accurately distinguishing the three operating states of the converter solely through machine vision. This solves the problem of converter safety management in scenarios without PLC signals, and has a wider range of applications. This invention achieves targeted triggering of violation detection through operating condition identification, avoiding misjudgments and omissions in non-hazardous operating conditions. Simultaneously, the personnel detection algorithm effectively eliminates interference from on-site equipment and materials. This invention integrates the entire process of "operating condition identification - personnel detection - real-time alarm - data storage" and automatically saves complete event data, enabling traceability and analysis of safety hazards. This invention replaces the traditional manual monitoring mode, reducing the manpower required for safety personnel and improving the efficiency and targeting of safety management.

[0034] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the concept and scope of the present invention. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the inventive concept should fall within the protection scope of the present invention. All technical contents for which protection is sought in this invention are fully described in the claims.

Claims

1. A machine vision-based safety identification system for personnel in steelmaking converters, characterized in that: include, The data acquisition unit is used to collect real-time image data of the steelmaking converter operating area; The processing unit is communicatively connected to the data acquisition unit and is used to autonomously identify the working status of the converter based on deep learning algorithms. When a high-risk working status is identified, it automatically triggers personnel detection in dangerous areas to determine whether there is a personnel unauthorized intrusion incident. An alarm unit, which is communicatively connected to the processing unit, is used to generate and push standardized alarm information when an unauthorized intrusion event is determined to exist. The evidence storage unit is communicatively connected to the processing unit and is used to store full-dimensional data on unauthorized intrusion events and synchronously upload them to the enterprise-level management platform.

2. The machine vision-based safety identification system for steelmaking converter personnel according to claim 1, characterized in that: The data acquisition unit includes at least one high-temperature and dust-resistant industrial camera module. The industrial camera module is deployed at preset monitoring points around the converter furnace mouth, the iron charging operation area, and the scrap steel feeding area. Its image acquisition parameter is a frame rate of 30 frames per second.

3. The machine vision-based safety identification system for steelmaking converter personnel according to claim 1, characterized in that: The processing unit has a built-in converter operating condition identification model, which can autonomously distinguish between three working states of the converter: iron charging operation, scrap steel adding operation, and standby waiting. The high-risk working states include iron charging operation and scrap steel adding operation.

4. The machine vision-based safety identification system for steelmaking converter personnel according to claim 3, characterized in that: The processing unit also has a built-in personnel target detection model, which is used to identify the main body of the person in the image after triggering the personnel detection in the dangerous area. Combined with the customizable electronic fence boundary, when it is detected that both feet of the person have completely entered the electronic fence area, it is determined as an unauthorized intrusion event.

5. A machine vision-based safety identification system for steelmaking converter personnel according to claim 1, characterized in that: The alarm unit includes an enterprise instant messaging interface module, which is used to push alarm information, including event time, location, real-time image and type, to a preset security management personnel terminal via an application programming interface; the evidence storage unit stores full-dimensional data including basic event attribute data and visual evidence data.

6. A machine vision-based method for safety identification of personnel in steelmaking converters, characterized in that, Includes the following steps: S1. Continuously collect real-time image data of the steelmaking converter operation area; S2. Based on deep learning algorithms, identify the current working status of the converter. When a high-risk working status is identified, automatically trigger the personnel detection process in the dangerous area. S3. Perform personnel target detection on the collected images and determine whether there is a personnel unauthorized intrusion event based on the preset electronic fence boundary; S4. If an unauthorized intrusion is detected, a standardized alarm message will be generated immediately and pushed to the relevant security management personnel. S5 automatically stores full-dimensional data on unauthorized intrusion events and synchronously uploads it to the enterprise-level management platform.

7. A machine vision-based safety identification method for steelmaking converter personnel according to claim 6, characterized in that, The method for identifying the working status of the converter in step S2 is as follows: collect real-time images of the converter under different working conditions to construct a labeled dataset, train a converter working condition identification model through a convolutional neural network algorithm, and analyze the real-time images through the model to distinguish the three working states of the converter: iron charging operation, scrap steel adding operation, and standby waiting.

8. A machine vision-based safety identification method for steelmaking converter personnel according to claim 6, characterized in that, The method for determining the unauthorized entry incident in step S3 is as follows: a target detection algorithm is used to identify the main body of the person in the image and obtain its coordinate information. The coordinates of the person are compared with the preset electronic fence boundary. When it is detected that both feet of the person have entered the electronic fence area, it is determined to be an unauthorized entry incident. Key frame images containing the person's location, dangerous area and work scene are captured as evidence.

9. A machine vision-based method for safety identification of personnel in steelmaking converters according to claim 6, characterized in that: The standardized alarm information in step S4 includes the time of the event, the specific dangerous area, the key frame image of the person violating the rules, and the event type.

10. A machine vision-based method for safety identification of personnel in steelmaking converters according to claim 6, characterized in that: The full-dimensional data in step S5 includes basic event attribute data and visual evidence data. The basic event attribute data includes the occurrence time, danger zone number, and operation type. The visual evidence data includes keyframe images of the violation event.