Limited space operation supervision method and system assisted by AI vision

Abstract

The application provides an auxiliary AI vision limited space operation supervision method and system, relates to the technical field of anomaly detection, and comprises the following steps: collecting operation images through a high-definition camera, compressing and transcoding the operation images by a video server, cropping and detecting operation region image data by an industrial computer; detecting the image data by a detection model to obtain three detection results, performing danger assessment according to the results, and generating a control instruction to control an alarm to realize real-time monitoring and safety management of limited space operation. Through the application, real-time safety detection of an operation site can be realized, a specific region can be detected, the detection efficiency is low, only the operation region is used for detection, the detection accuracy is improved, false positives are reduced, intelligent judgment of potential dangers can be realized, automatic alarm and intervention of the operation site can be realized, and the operation safety is improved.

Description

Technical Field

[0001] This application relates to the field of anomaly detection technology, and in particular to a method and system for monitoring confined space operations using AI-assisted vision. Background Technology

[0002] With the continuous advancement of industrialization, workplace safety has become a crucial issue for enterprise development. In modern industrial production, especially in harsh environments with limited space, such as in the chemical industry, if the work area is confined and ventilation and hazardous gas content monitoring are not conducted, hazardous gas levels may exceed safe limits. This leads to further exposure of workers to dangerous conditions, causing various health problems such as respiratory illnesses, skin diseases, and poisoning. Confined space work, due to its unique environmental conditions, presents numerous potential hazards, thus requiring extremely high safety standards. Traditional manual monitoring methods are insufficient for comprehensive and real-time monitoring.

[0003] Currently, existing technologies are insufficient to perform real-time detection of all areas at the work site, and cannot detect specific areas, resulting in low detection efficiency. Summary of the Invention

[0004] The purpose of this application is to provide a method and system for monitoring confined space operations using AI vision, in order to solve the technical problems that existing technologies are unable to achieve real-time detection of all areas of the work site and cannot detect specific areas, resulting in low detection efficiency.

[0005] In view of the above problems, this application provides a method and system for monitoring confined space operations with the assistance of AI vision.

[0006] Firstly, this application provides a confined space operation supervision method assisted by AI vision, which is implemented through a confined space operation supervision system assisted by AI vision. The method includes: acquiring high-definition operation image data of a target area in real time using the image acquisition device; compressing and transcoding the high-definition operation image data using the video server to generate packaged data; inputting the packaged data into an industrial control computer, which includes a cropping unit and a detection unit; the cropping unit cropping the packaged data according to the operation area to generate operation area image data; detecting the operation area image data using an operation detection model deployed in the detection unit to obtain operation detection results, including a first detection result, a second detection result, and a third detection result; performing a hazard assessment based on preset rules using the first detection result, the second detection result, and the third detection result to obtain a hazard assessment result; generating control instructions based on the hazard assessment result; and controlling the alarm according to the control instructions.

[0007] Secondly, this application also provides a confined space operation monitoring system assisted by AI vision, used to execute the confined space operation monitoring method assisted by AI vision as described in the first aspect. The confined space operation monitoring system assisted by AI vision includes: a data acquisition module for real-time acquisition of high-definition operation image data of a target area using an image acquisition device; an image processing module for compressing and transcoding the high-definition operation image data via a video server to generate packaged data; an image cropping module for inputting the packaged data into an industrial control computer, the industrial control computer including a cropping unit and a detection unit, the cropping unit performing ROI cropping on the packaged data according to the operation area to generate operation area image data; an image detection module for detecting the operation area image data using an operation detection model deployed in the detection unit to obtain operation detection results, including a first detection result, a second detection result, and a third detection result; a hazard assessment module for performing a hazard assessment based on preset rules using the first detection result, the second detection result, and the third detection result to obtain a hazard assessment result; and a control command generation module for generating control commands based on the hazard assessment results and controlling an alarm according to the control commands.

[0008] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0009] By utilizing the image acquisition device to acquire high-definition operational image data of the target area in real time; compressing and transcoding the high-definition operational image data through the video server to generate packaged data; inputting the packaged data into the industrial control computer, which includes a cropping unit and a detection unit, the cropping unit performs ROI cropping on the packaged data according to the operational area to generate operational area image data; detecting the operational area image data through an operational detection model deployed in the detection unit to obtain operational detection results, including a first detection result, a second detection result, and a third detection result; combining the first detection result, the second detection result, and the third detection result with preset rules to perform a hazard assessment to obtain a hazard assessment result; generating control commands based on the hazard assessment results, and controlling the alarm according to the control commands, this effectively solves the technical problems of existing technologies that make it difficult to achieve real-time detection of all areas of the operational site, cannot detect specific areas, resulting in low detection efficiency, and only using the operational area for detection. This improves detection accuracy, reduces false alarms, enables intelligent judgment of potential hazards, realizes automatic alarm and intervention at the operational site, and improves operational safety.

[0010] The above description is merely an overview of the technical solution of this application. To better understand the technical means of this application and to facilitate its implementation according to the description, and to make the above and other objects, features, and advantages of this application more apparent, specific embodiments of this application are described below. It should be understood that the content described in this section is not intended to identify key or important features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent through the following description. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0012] Figure 1 This is a flowchart illustrating the AI-assisted vision-based method for monitoring confined space operations as described in this application.

[0013] Figure 2 This is a schematic diagram of the structure of the confined space operation monitoring system that assists AI vision in this application.

[0014] Explanation of reference numerals in the attached figures:

[0015] The system includes a data acquisition module 11, an image processing module 12, an image cropping module 13, an image detection module 14, a hazard assessment module 15, and a control command generation module 16. Detailed Implementation

[0016] This application provides a confined space operation monitoring method and system that assists AI vision. It solves the technical problems of existing technologies, such as the inability to achieve real-time detection of all areas of the work site and the inability to detect specific areas, resulting in low detection efficiency. This allows only the work area to be used for detection, improving detection accuracy, reducing false alarms, enabling intelligent judgment of potential hazards, and realizing automatic alarms and interventions at the work site, thereby improving work safety.

[0017] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. It should be understood that this application is not limited to the exemplary embodiments described herein. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application. It should also be noted that, for ease of description, only the parts related to this application are shown in the accompanying drawings, not all of them.

[0018] Example 1, please refer to the appendix. Figure 1 This application provides a confined space operation supervision method assisted by AI vision. The method is applied to a confined space operation supervision system assisted by AI vision. This system is communicatively connected to a safety compliance detection system. The safety compliance detection system includes an image acquisition device, a video server, an industrial control computer, and an alarm. The specific steps of the AI ​​vision-assisted confined space operation supervision method are as follows:

[0019] S1: Use the image acquisition device to acquire high-definition operational image data of the target area in real time.

[0020] Specifically, select appropriate high-definition cameras based on factors such as the size of the work environment, lighting conditions, and type of work. For example, infrared cameras are needed in dimly lit underground mines. Install cameras in locations that maximize coverage of the work area and minimize blind spots. Multiple cameras may need to be installed at various angles and heights. Adjust camera parameters such as resolution, frame rate, and exposure time to ensure clear images are obtained even under dynamic or changing lighting conditions.

[0021] S2: The high-definition job image data is compressed and transcoded by the video server to generate packaged data.

[0022] Specifically, a video server is a device specifically designed for processing and distributing video data. It typically receives raw video streams from image acquisition devices and processes them. Compression and transcoding parameters, including resolution, frame rate, and bitrate, are configured on the video server to ensure it can handle video streams from multiple cameras. Compression is a technique for reducing data size by removing redundant information, thus reducing required storage space and bandwidth. Video compression can use various encoding standards, such as H.264, H.265, or VP9. Transcoding refers to the process of converting video from one encoding format to another. This is done to adapt to different playback devices or network environments. For example, transcoding high-definition video into a format suitable for playback on mobile devices. Packaging data refers to encapsulating compressed and transcoded video data into a format suitable for transmission. This includes encapsulating video streams into RTSP, RTMP, or other streaming media protocols for easy transmission over the network.

[0023] S3: Input the packaged data into the industrial control computer. The industrial control computer includes a cropping unit and a detection unit. The cropping unit performs ROI cropping on the packaged data according to the work area to generate work area image data.

[0024] Specifically, an industrial PC is an edge computing device designed specifically for industrial environments. It features high stability and resistance to environmental interference, making it suitable for processing data from industrial sites. A cropping unit is a software or hardware module within the industrial PC whose function is to crop specific regions, or Areas of Interest (ROIs), from the original image. An ROI is a region of interest to the user or an area requiring focused analysis. ROI cropping refers to the process of extracting specific regions from the original image data. This reduces the computational load of subsequent image processing, improving processing speed and efficiency. The image data generated after ROI cropping only contains the work area, which will be used for subsequent image analysis and work detection. The cropping unit is configured on the industrial PC to process the input image data according to the defined ROIs. When packaged data is input into the industrial PC, the cropping unit extracts the corresponding ROI from each frame of the image according to preset parameters. The cropped image data is then saved as a new image file or data stream for use by the detection unit.

[0025] S4: The image data of the work area is detected by the work detection model deployed in the detection unit to obtain the work detection results, including the first detection result, the second detection result, and the third detection result.

[0026] Specifically, the detection unit is a key component of the industrial control computer. It is responsible for running the job detection model, processing the input image data, and outputting the detection results. The job detection model is a trained model based on machine learning or deep learning algorithms, capable of identifying specific objects, behaviors, or scenes in an image. The first detection result, second detection result, and third detection result can be different types of analysis and recognition results. Specifically, the first detection result detects whether a person with a blower has entered the work area; the second detection result detects whether a gas monitoring instrument is present in the work area; and the third detection result detects whether a person is monitoring the work area, providing attribute and location information for these three categories of targets.

[0027] S5: Combine the first detection result, the second detection result, and the third detection result to perform a hazard assessment based on preset rules, and obtain the hazard assessment result.

[0028] Furthermore, this application also includes:

[0029] If the first detection result, the second detection result, and the third detection result cannot be obtained simultaneously within a preset number of consecutive frames, a dangerous instruction is generated; at the same time, detection continues, and if the first detection result, the second detection result, and the third detection result can be obtained simultaneously, the dangerous instruction is updated to a safe instruction.

[0030] Specifically, a threshold for the number of consecutive frames is preset by someone skilled in the art, for example, it might be set to 5 or 10 frames. This threshold indicates that a specific object or event is expected to be consistently detected within this number of consecutive frames. The first, second, and third detection results need to be obtained simultaneously, and the relevant object or event, such as personnel, safety equipment, and equipment status, must be successfully detected in each frame. If the three detection results cannot be obtained simultaneously within the preset number of consecutive frames, a hazard instruction is generated. This instruction might activate an alarm, stop work, or notify management to take action. If the system can obtain three detection results simultaneously during continuous detection, it updates the previous hazard instruction to a safety instruction. This means the work environment is now safe, and work can continue or the previous alarm status can be cleared.

[0031] S6: Generate control instructions based on the hazard assessment results, and control the alarm according to the control instructions.

[0032] Specifically, a hazard assessment result is a quantitative evaluation of the safety status of the work site, including the risk level, urgency, or specific warning information. Control commands are specific operational instructions generated based on the hazard assessment results, dictating how the monitoring system should respond. These commands can include activating alarms, sending notifications, suspending work, etc. An alarm is an audio or visual device used to warn on-site personnel when a safety risk is detected. Alarms can be on-site loudspeakers, flashing lights, SMS notification systems, or other communication devices.

[0033] Furthermore, the image acquisition device also includes a high-definition camera, a WIFI transmitter, and a WIFI receiver. The high-definition camera is connected to the WIFI transmitter via PoE; the WIFI receiver is connected to the video server via a first preset interface; the video server is connected to the industrial control computer via a second preset interface; and the industrial control computer is connected to the alarm via a third preset interface.

[0034] Specifically, the high-definition camera is used to collect high-definition image data from the work site. It features high resolution, night vision capabilities, and a wide-angle lens to ensure clear images under various lighting conditions. The high-definition camera connects to the Wi-Fi transmitter via PoE technology. PoE technology allows for simultaneous data and power transmission over an Ethernet cable, simplifying installation and wiring. The Wi-Fi receiver receives image data from the Wi-Fi transmitter. It is a wireless router or wireless access point responsible for converting wireless signals into wired signals for transmission to the video server. The Wi-Fi receiver and video server are connected via a first preset interface. This interface can be an Ethernet interface, a fiber optic interface, or another interface suitable for high-speed data transmission. The video server receives image data from the Wi-Fi receiver, performs compression and transcoding, and then packages the data for transmission to the industrial control computer. The video server and industrial control computer are connected via a second preset interface. This interface also needs to support high-speed data transmission and can be USB, Ethernet, or another specialized interface. The industrial control computer includes a cropping unit and a detection unit, responsible for ROI cropping and work detection of the input image data, and then performing a hazard assessment based on the detection results. The industrial control computer connects to the alarm via a third preset interface. This interface can be a serial port, GPIO, or other suitable interface for controlling the alarm. Based on the control commands generated by the industrial control computer, the alarm will emit sound or light signals to warn on-site personnel and managers to pay attention to safety.

[0035] Furthermore, this application also includes:

[0036] A high-definition camera deployed at the top of the target area collects high-definition operational image data within the target area along a preset acquisition direction, and transmits the collected high-definition operational image data to the WIFI transmitter; the WIFI transmitter transmits the collected high-definition operational image data to the WIFI receiver wirelessly; the WIFI receiver transmits the high-definition operational image data to the video server.

[0037] Specifically, high-definition cameras are deployed at the top of the target area. The camera's position and angle are adjusted as needed, and a preset acquisition direction is used to ensure coverage of all critical parts of the work area. The cameras monitor the target area in real time along the preset acquisition direction, acquiring high-definition image data. A Wi-Fi transmitter receives the image data from the high-definition cameras. Since the cameras are deployed in locations where cabling is difficult, a Wi-Fi transmitter facilitates wireless data transmission. The Wi-Fi transmitter transmits image data to a Wi-Fi receiver via a wireless network. This wireless transmission method reduces the need for cabling and improves installation flexibility. The Wi-Fi receiver receives the high-definition image data transmitted from the Wi-Fi transmitter. The received image data is then transmitted to a video server via the Wi-Fi receiver. This includes converting the wireless signal into a wired signal for processing by the video server. The video server receives the image data from the Wi-Fi receiver and performs compression and transcoding to generate packaged data suitable for transmission and storage.

[0038] Furthermore, step S3 of this application also includes:

[0039] The packaged data is converted into HSV image data; a mask is created based on a preset color detection range according to the HSV image data; the mask is filled using a closing operation, and then an opening operation is performed on the filled data to obtain standard image data; the contour area of ​​the standard image data is calculated, and the largest contour area is taken as the working area; the packaged data is cropped according to the working area to obtain the working area image data.

[0040] Specifically, RGB image data is used for display and storage, but in image analysis, the HSV color space better highlights color information, facilitating image processing based on color features. The RGB image data in the packaged data is converted to HSV format. This can be achieved using functions in OpenCV or other image processing libraries. A mask is created based on a preset color detection range. This range defines the color range of interest; for example, one might only focus on objects of a specific color. The mask is a binary image used to highlight or isolate specific color regions in the image. Closing operations are performed on the mask to fill any small holes or gaps within the color regions. Closing is a morphological operation that first dilates and then erodes, thus connecting adjacent color patches. Opening operations are performed on the filled data to smooth the edges of the color regions and remove small noise points. Opening is the opposite of closing, first eroding and then dilating. The image data processed by closing and opening operations is called standard image data, which contains a clear working area. The area of ​​all contours in the standard image data is calculated, and the largest contour area is found. This largest contour represents the working area. Based on the calculated task area, the original packaged data is cropped to retain only the image data of the task area. Using the contour information of the task area, a Region of Interest (ROI) is created, and that region is extracted from the original image.

[0041] Furthermore, step S4 of this application also includes:

[0042] The image data of the work area is labeled frame by frame using LabelImg; the labeled image data of the work area is divided into a training set and a validation set according to a preset ratio, wherein the training set consists of 1600 frames and the validation set consists of 400 frames; the work detection model is trained based on the training set to obtain detection weights; the performance of the detection weights is evaluated based on the validation set, and the optimal detection weights are selected based on the evaluation results; the performance of the obtained detection weights is evaluated using the validation dataset of a finite work space image dataset to quantify the detection accuracy of the weights and select the optimal detection weights; the work area image data is detected using the TensorRT model, the optimal detection weights, and the work detection model.

[0043] Specifically, LabelImg is a graphical user interface annotation tool that allows users to add and edit bounding boxes to images and assign a label to each box. Open LabelImg, load the job area image data, and then draw bounding boxes for objects in the image frame by frame, labeling the object's category. For example, labeling workers, tools, equipment, etc. After annotation, LabelImg generates an XML or JSON file containing the bounding box and label information for each image. These files will be used to train the job detection model. According to a preset ratio, the labeled job area image data is divided into a training set and a validation set. The training set is used for model learning, while the validation set is used to evaluate the model's performance. In this embodiment, the training set contains 1600 frames, and the validation set contains 400 frames. The training set is used to train the job detection model. This model may be a convolutional neural network or other deep learning model suitable for object detection. During training, the model learns how to identify and locate objects in the image, generating detection weights. These weights are the core of the model and are used for subsequent detection tasks. The validation set is used to evaluate the performance of the trained detection weights. Evaluation metrics include precision, recall, and F1 score. Based on the evaluation results, the optimal detection weights with the best performance are selected. These weights will be used in actual detection tasks. TensorRT is a deep learning inference engine that optimizes deep learning models to improve inference speed and efficiency. The optimal detection weights are integrated with the TensorRT model and the task detection model for real-time detection of image data in the task region.

[0044] Furthermore, step S6 of this application also includes:

[0045] The wearable device uses a built-in hazardous gas sensor to monitor the concentration of hazardous gases in the target area in real time. If the concentration of hazardous gases exceeds a preset concentration, a first alarm command is generated. The device also uses a noise sensor to monitor the noise level in the target area in real time. If the noise level exceeds a preset noise level, a second alarm command is generated. The control command is generated based on the first alarm command, the second alarm command, and the hazard assessment result.

[0046] Specifically, wearable hazardous gas sensors, such as gas detectors or smart wearable devices, are deployed within the target area. These devices can be worn by workers or fixed in specific locations within the work area. The sensors monitor the concentration of hazardous gases, such as hydrogen sulfide, carbon monoxide, and methane, in the target area in real time. The system presets safe concentration ranges for these gases. When the detected concentration exceeds this range, the system triggers an alarm. If the hazardous gas concentration exceeds the preset concentration, the system generates a first alarm command and may take measures such as stopping work, evacuating personnel, and notifying rescue efforts. Noise sensors, such as sound level meters or smart wearable devices, are also deployed within the target area. These devices can be worn by workers or fixed in specific locations within the work area. The sensors monitor noise levels within the target area in real time. The system presets safe noise levels. When the detected noise level exceeds this level, the system triggers an alarm. If the noise level exceeds the preset noise level, the system generates a second alarm command and may take measures such as reducing the noise source, reminding workers to wear earplugs, and notifying management. The system comprehensively assesses the safety status of the work environment based on the first alarm command, the second alarm command, and the hazard assessment results. Based on the comprehensive assessment results, control instructions are generated. These instructions may include stopping operations, adjusting working conditions, and reminding personnel to take protective measures.

[0047] In summary, the AI-assisted vision-based confined space operation monitoring method provided in this application has the following technical effects:

[0048] By utilizing the image acquisition device to acquire high-definition operational image data of the target area in real time; compressing and transcoding the high-definition operational image data through the video server to generate packaged data; inputting the packaged data into the industrial control computer, which includes a cropping unit and a detection unit, the cropping unit performs ROI cropping on the packaged data according to the operational area to generate operational area image data; detecting the operational area image data through an operational detection model deployed in the detection unit to obtain operational detection results, including a first detection result, a second detection result, and a third detection result; combining the first detection result, the second detection result, and the third detection result with preset rules to perform a hazard assessment to obtain a hazard assessment result; generating control commands based on the hazard assessment results, and controlling the alarm according to the control commands, this effectively solves the technical problems of existing technologies that make it difficult to achieve real-time detection of all areas of the operational site and cannot detect specific areas, resulting in low detection efficiency. This ensures that only the operational area is used for detection, improving detection accuracy, reducing false alarms, enabling intelligent judgment of potential hazards, achieving automatic alarm and intervention at the operational site, and improving operational safety.

[0049] Example 2: Based on the same inventive concept as the AI-assisted vision-based confined space operation monitoring method described in the previous examples, this application also provides an AI-assisted vision-based confined space operation monitoring system. Please refer to the appendix. Figure 2 The AI-assisted vision-based confined space operation monitoring system includes:

[0050] The data acquisition module 11 is used to acquire high-definition operational image data of the target area in real time using image acquisition equipment.

[0051] The image processing module 12 is used to compress and transcode the high-definition job image data through the video server to generate packaged data.

[0052] Image cropping module 13 is used to input the packaged data into an industrial control computer. The industrial control computer includes a cropping unit and a detection unit. The cropping unit performs ROI cropping on the packaged data according to the work area to generate work area image data.

[0053] The image detection module 14 is used to detect the image data of the work area through the job detection model deployed in the detection unit, and obtain job detection results, including a first detection result, a second detection result, and a third detection result.

[0054] The hazard assessment module 15 is used to combine the first detection result, the second detection result, and the third detection result to perform a hazard assessment based on preset rules, and obtain a hazard assessment result.

[0055] The control command generation module 16 is used to generate control commands based on the hazard assessment results, and to control the alarm based on the control commands.

[0056] Furthermore, the AI-assisted vision-based confined space operation monitoring system also includes a device connection module for:

[0057] The image acquisition device also includes a high-definition camera, a WIFI transmitter, and a WIFI receiver. The high-definition camera is connected to the WIFI transmitter via PoE; the WIFI receiver is connected to the video server via a first preset interface; the video server is connected to the industrial control computer via a second preset interface; and the industrial control computer is connected to the alarm via a third preset interface.

[0058] Furthermore, the AI-assisted vision-based confined space operation monitoring system also includes a data transmission module for:

[0059] A high-definition camera deployed at the top of the target area collects high-definition operational image data within the target area along a preset acquisition direction, and transmits the collected high-definition operational image data to the WIFI transmitter; the WIFI transmitter transmits the collected high-definition operational image data to the WIFI receiver wirelessly; the WIFI receiver transmits the high-definition operational image data to the video server.

[0060] Furthermore, the image cropping module 13 in the AI-assisted vision-based confined space operation monitoring system is also used for:

[0061] The packaged data is converted into HSV image data; a mask is created based on a preset color detection range according to the HSV image data; the mask is filled using a closing operation, and then an opening operation is performed on the filled data to obtain standard image data; the contour area of ​​the standard image data is calculated, and the largest contour area is taken as the working area; the packaged data is cropped according to the working area to obtain the working area image data.

[0062] Furthermore, the image detection module 14 in the AI-assisted vision-based confined space operation monitoring system is also used for:

[0063] The image data of the work area is labeled frame by frame using LabelImg; the labeled image data of the work area is divided into a training set and a validation set according to a preset ratio, wherein the training set consists of 1600 frames and the validation set consists of 400 frames; the work detection model is trained based on the training set to obtain detection weights; the performance of the detection weights is evaluated based on the validation set, and the optimal detection weights are selected based on the evaluation results; the performance of the obtained detection weights is evaluated using the validation dataset of a finite work space image dataset to quantify the detection accuracy of the weights and select the optimal detection weights; the work area image data is detected using the TensorRT model, the optimal detection weights, and the work detection model.

[0064] Furthermore, the hazard assessment module 15 in the AI-assisted vision-based confined space operation monitoring system is also used for:

[0065] If the first detection result, the second detection result, and the third detection result cannot be obtained simultaneously within a preset number of consecutive frames, a dangerous instruction is generated; at the same time, detection continues, and if the first detection result, the second detection result, and the third detection result can be obtained simultaneously, the dangerous instruction is updated to a safe instruction.

[0066] Furthermore, the control instruction generation module 16 in the AI-assisted vision-based confined space operation monitoring system is also used for:

[0067] The wearable device uses a built-in hazardous gas sensor to monitor the concentration of hazardous gases in the target area in real time. If the concentration of hazardous gases exceeds a preset concentration, a first alarm command is generated. The device also uses a noise sensor to monitor the noise level in the target area in real time. If the noise level exceeds a preset noise level, a second alarm command is generated. The control command is generated based on the first alarm command, the second alarm command, and the hazard assessment result.

[0068] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Figure 1 The AI-assisted vision-based confined space operation monitoring method and specific examples in Embodiment 1 are also applicable to the AI-assisted vision-based confined space operation monitoring system of this embodiment. Through the foregoing detailed description of the AI-assisted vision-based confined space operation monitoring method, those skilled in the art can clearly understand the AI-assisted vision-based confined space operation monitoring system of this embodiment; therefore, for the sake of brevity, it will not be described in detail here. As the system disclosed in the embodiments corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant details can be found in the method section.

[0069] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0070] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application also intends to include such modifications and variations.

Claims

1. A method for monitoring confined space operations using AI-assisted vision, characterized in that, The AI-assisted vision-based confined space operation monitoring method is applied to an AI-assisted vision-based confined space operation monitoring system. This system is communicatively connected to a safety compliance detection system. The safety compliance detection system includes image acquisition equipment, a video server, an industrial control computer, and an alarm. The AI-assisted vision-based confined space operation monitoring method includes: The image acquisition device is used to acquire high-definition operational image data of the target area in real time; The video server compresses and transcodes the high-definition job image data to generate packaged data. The packaged data is input into the industrial control computer, which includes a cropping unit and a detection unit. The cropping unit performs ROI cropping on the packaged data according to the work area to generate work area image data. The image data of the work area is detected by the work detection model deployed in the detection unit to obtain work detection results, including a first detection result, a second detection result, and a third detection result. The first detection result detects whether a person with a blower has entered the work area; the second detection result detects whether a gas monitoring instrument is present in the work area; and the third detection result detects whether a person is monitoring the work area. These three types of targets have attribute and location information. Based on preset rules, a hazard assessment is performed by combining the first, second, and third detection results to obtain the hazard assessment result. Based on the hazard assessment results, control instructions are generated, and the alarm is controlled according to the control instructions. Combining the first detection result, the second detection result, and the third detection result, a hazard assessment is performed based on preset rules to obtain a hazard assessment result, which also includes: If the first detection result, the second detection result, and the third detection result cannot be obtained simultaneously within a preset number of consecutive frames, a dangerous command is generated. Simultaneously, continuous detection is performed. If the first detection result, the second detection result, and the third detection result can be obtained simultaneously, the dangerous instruction is updated to a safe instruction. The cropping unit performs ROI cropping on the packaged data according to the work area to generate work area image data, and also includes: The packaged data is converted into HSV image data; Based on the HSV image data, a mask is created based on a preset color detection range; The mask is filled using a closing operation, and then an opening operation is performed on the filled data to obtain standard image data. Calculate the contour area of ​​the standard image data, and take the largest contour area as the working area; The packaging data is cropped according to the work area to obtain the image data of the work area; The detection of the image data of the work area by the work detection model deployed in the detection unit also includes: The image data of the work area is labeled frame by frame using LabelImg; The labeled image data of the work area is divided into a training set and a validation set according to a preset ratio, wherein the training set consists of 1600 frames and the validation set consists of 400 frames. The job detection model is trained based on the training set to obtain the detection weights; The detection weights are evaluated based on the validation set, and the optimal detection weights are selected based on the evaluation results. The performance of the obtained detection weights is evaluated using a validation dataset of a finite job space image dataset to quantify the detection accuracy of the weights and select the optimal detection weights. The image data of the task area is detected based on the TensorRT model, the optimal detection weights, and the task detection model.

2. The confined space operation monitoring method assisted by AI vision as described in claim 1, characterized in that, The image acquisition device also includes a high-definition camera, a WIFI transmitter, and a WIFI receiver. The high-definition camera is connected to the WIFI transmitter via PoE; the WIFI receiver is connected to the video server via a first preset interface; the video server is connected to the industrial control computer via a second preset interface; and the industrial control computer is connected to the alarm via a third preset interface.

3. The confined space operation monitoring method assisted by AI vision as described in claim 2, characterized in that, Also includes: A high-definition camera deployed at the top of the target area collects high-definition operation image data within the target area along a preset acquisition direction, and transmits the collected high-definition operation image data to the WIFI transmitter. The WIFI transmitter transmits the acquired high-definition operational image data to the WIFI receiver wirelessly. The WIFI receiver transmits the high-definition image data to the video server.

4. The confined space operation monitoring method assisted by AI vision as described in claim 1, characterized in that, The system further includes generating control commands based on the hazard assessment results, controlling the alarm based on the control commands, and also includes: The wearable device uses a built-in hazardous gas sensor to monitor the concentration of hazardous gases in the target area in real time. If the concentration of hazardous gases exceeds a preset concentration, a first alarm command is generated. The noise level in the target area is monitored in real time by a noise sensor. If the noise level exceeds a preset noise level, a second alarm command is generated. The control command is generated based on the first alarm command, the second alarm command, and the hazard assessment result.

5. A confined space operation monitoring system assisted by AI vision, characterized in that, The steps for implementing the confined space operation monitoring method with AI-assisted vision according to any one of claims 1 to 4, wherein the confined space operation monitoring system with AI-assisted vision comprises: The data acquisition module is used to acquire high-definition operational image data of the target area in real time using image acquisition equipment; The image processing module is used to compress and transcode the high-definition job image data through the video server to generate packaged data; An image cropping module is used to input the packaged data into an industrial control computer. The industrial control computer includes a cropping unit and a detection unit. The cropping unit performs ROI cropping on the packaged data according to the work area to generate work area image data. The image detection module is used to detect the image data of the work area through the job detection model deployed in the detection unit, and obtain job detection results, including a first detection result, a second detection result, and a third detection result; The hazard assessment module is used to combine the first detection result, the second detection result, and the third detection result to perform a hazard assessment based on preset rules, and obtain a hazard assessment result. A control command generation module is used to generate control commands based on the hazard assessment results, and to control the alarm based on the control commands. The hazard assessment module is also used for: If the first detection result, the second detection result, and the third detection result cannot be obtained simultaneously within a preset number of consecutive frames, a dangerous instruction is generated; at the same time, detection continues, and if the first detection result, the second detection result, and the third detection result can be obtained simultaneously, the dangerous instruction is updated to a safe instruction.

Images