AI self-rescuer test system and identification method thereof

Abstract

The invention discloses an AI self-rescuer test system and an identification method thereof, the test system comprises a self-rescuer body and an AI identification workbench, and the identification method comprises the steps of collecting self-rescuer wearing data through an image identification module, and transmitting the collected data to a remote server for preprocessing; and a deep neural network is utilized to learn specific image features of the preprocessed data, an input image is predicted and mapped to a corresponding input result, a prediction result of the image is analyzed, and a logic processing module performs comprehensive judgment on the prediction result of the video frame according to the model. The self-rescuer is simple and reasonable in structure and easy and convenient to operate, an operator can effectively and correctly use the self-rescuer, and the self-rescuer has high practical value.

Description

Technical Field

[0001] The present invention relates to the technical field of self-rescuer assessment, and in particular to an AI self-rescuer assessment and training system and an identification method thereof. Background Art

[0002] Coal mine safety is an important part of protecting the lives and property of miners. Due to the particularity of the coal mining environment, there are many potential safety hazards such as gas outbursts, fires, and water seepage. Once an accident occurs, it often causes serious casualties and property losses. Therefore, improving the safety awareness and self-rescue ability of miners is of great significance to reducing coal mine accidents and protecting the lives of miners.

[0003] Mining self-rescuers are widely used in coal mines and other mining fields. Safety issues are of vital importance when working in mines. Self-rescuers are essential protective equipment for workers working underground in coal mines and are the primary equipment for effective escape when disasters occur underground in coal mines.

[0004] Therefore, how underground workers in coal mines use self-rescuers correctly plays a vital role in the field of mine safety. Therefore, it is very important to provide an efficient self-rescuer training system to improve the operating accuracy of underground workers in coal mines through the training system. Summary of the invention

[0005] The present invention provides an AI self-rescuer testing and training system and an identification method thereof.

[0006] The specific technical solution of the present invention is: an AI self-rescuer training system, including a self-rescuer body, an AI recognition workbench, the AI ​​recognition workbench includes a main shell, a movable base is installed under the main shell, an embedded soft light is installed on the main shell, an image recognition module is installed under the embedded soft light, a touch screen is installed under the image recognition module, a power amplifier is installed under the touch screen, a status indicator light is installed under the power amplifier, and an identity recognition module is installed under the status indicator light.

[0007] Furthermore, the image recognition module includes an AI camera that recognizes the examinee's movements.

[0008] Furthermore, the embedded soft light is used to provide lighting, the touch screen is used to display the test and training interface system, and the identity recognition module is used to identify the candidate information.

[0009] Furthermore, the present invention also provides an identification method for an AI self-rescuer test and training system, comprising: S1. Collect the self-rescuer wearing data through the image recognition module and transmit the collected data to the remote server for preprocessing; S2. Use the deep neural network to learn specific image features from the preprocessed data, predict the input image, map it to the corresponding input result, and analyze the predicted result of the image. The system can quickly and in real time determine whether the operator correctly performs the specific action of the self-rescuer in actual operation; S3. The logic processing module makes a comprehensive judgment on the prediction results of the video frame based on the model.

[0010] Furthermore, the self-rescuer wearing data collected in the S1 step is a video of different operators operating the self-rescuer. The recorded video is transmitted to a remote server via wireless communication, and the self-rescuer data is preprocessed. The preprocessing includes video frame extraction, data cleaning, and data enhancement. For the data containing the self-rescuer nose clip, it is also necessary to obtain the face position through the human key point detection model built into the MediaPipe computer vision library, and the face position of the source image is cropped to obtain a face image data set containing the self-rescuer nose clip.

[0011] Furthermore, the specific learning process of using a deep neural network to learn specific image features in the S2 step is to annotate the image features and divide them into a training set and a verification set. The image features of the self-rescuer include a strap, a shell, a mouthpiece, an airbag and a nose clip. A deep neural network is selected as a model for training and verification data, and the video frames of the camera are input into the deep neural network model to obtain prediction results of the video frames. The prediction results of the image are analyzed to obtain recognition schemes for the operations of opening the self-rescuer cover, deploying the self-rescuer airbag and wearing the self-rescuer nose clip.

[0012] Furthermore, the process of analyzing the prediction results of the image in the S2 step is to import the neural network model framework and load the trained model weights, use the model with loaded weights to infer the input video frame in the GPU environment, and the obtained inference results include the category and location information of the target. If there is a specific target category in the image, this category will be recorded. After noise reduction processing, the result will be passed to the self-rescuer training interface system for display.

[0013] Furthermore, the neural network in step S2 adopts a single-stage target detection algorithm, specifically the YOLO series, which is known for its high efficiency and real-time performance. It can simultaneously predict the bounding box and category of the target in a single forward propagation, thereby achieving rapid target recognition. Target detection is used to identify images or video frames to determine the objects described in the image or stream, and each object has its category and location information.

[0014] Furthermore, the comprehensive judgment process in the S3 step is to use the deep learning model and the human key point detection model to predict the video frame respectively. When the result of the deep learning model predicting that the video frame contains a predetermined target category, the detection frame position coordinates of the predetermined target category are obtained, and the position coordinates of the artificially selected human key points are compared with the detection frame position coordinates or the derived frame generated by the detection frame position coordinates. If they are within the detection frame position area, it is considered that the operator has completed the established action, and an identification scheme for the established action is obtained.

[0015] The present invention can accurately identify human body movements, analyze the prediction results of the image for the self-rescuer wearing data, input the video frame into the deep neural network model, obtain the prediction results of the video frame, and obtain the operations of opening the self-rescuer cover, unfolding the self-rescuer airbag and wearing the self-rescuer nose clip. The comprehensive judgment process is to use the deep learning model and the human key point detection model to predict the video frame respectively. When the result of the deep learning model predicting the video frame contains a predetermined target category, the detection frame position coordinates of the predetermined target category are obtained, and the position coordinates of the human key point are compared with the detection frame position coordinates or the derivative frame generated by the detection frame position coordinates. If it is within the detection frame position area, it is considered that the operator has completed the established action. The system can quickly and in real time determine whether the operator has correctly performed the specific action of the self-rescuer in actual operation, thereby improving the efficiency and accuracy of the test and training system. The present invention has a simple and reasonable structure, is simple and convenient to operate, can effectively realize the correct use of the self-rescuer by the operator, and has high practical value. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A three-dimensional image of the AI ​​recognition workbench of the present invention; In the figure: 1. Embedded soft light; 2. Image recognition module; 3. Touch screen; 4. Power amplifier; 5. Status indicator light; 6. Identity recognition module; 7. Main shell; 8. Base. DETAILED DESCRIPTION

[0017] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0018] like Figure 1An AI self-rescuer examination and training system is shown, including a self-rescuer body and an AI recognition workbench, wherein the AI ​​recognition workbench includes a main shell 7, a movable base 8 is installed below the main shell 7, an embedded soft light 1 is installed on the main shell 7, the embedded soft light 1 is used to provide lighting, an image recognition module 2 is installed below the embedded soft light 1, the image recognition module 2 includes an AI camera for identifying the movements of the examinee, a touch screen 3 is installed below the image recognition module 2, the touch screen 3 is used to display the examination and training interface system, a power amplifier 4 is installed below the touch screen 3, a status indicator light 5 is installed below the power amplifier 4, an identity recognition module 6 is installed below the status indicator light 5, and the identity recognition module 6 is used to identify the examinee information.

[0019] An identification method for an AI self-rescuer testing and training system, comprising: S1. Collect the self-rescuer wearing data through the image recognition module (1), and transmit the collected data to the remote server for preprocessing. The collected self-rescuer wearing data is a video of different operators operating the self-rescuer. The recorded video is transmitted to the remote server through wireless communication, and the self-rescuer data is preprocessed. The preprocessing includes video frame extraction, data cleaning, and data enhancement. For the data containing the self-rescuer nose clip, it is also necessary to obtain the face position through the human body key point detection model built into the MediaPipe computer vision library, and the face position of the source image is cropped to obtain a face image dataset containing the self-rescuer nose clip.

[0020] S2. Use a deep neural network to learn specific image features from the preprocessed data, predict the input image, map it to the corresponding input result (such as target category, target position, etc.), analyze the predicted results of the image, and the system can quickly and in real time determine whether the operator has correctly performed the specific actions of the self-rescuer in actual operation, such as opening the self-rescuer cover, deploying the self-rescuer airbag, and wearing the self-rescuer nose clip. The specific learning process of using a deep neural network to learn specific image features in the S2 step is to annotate the image features and divide them into a training set and a verification set. The image features of the self-rescuer include straps, shells, mouthpieces, airbags, and nose clips. A deep neural network is selected as a model for training and verification data, and the video frames of the camera are input into the deep neural network model to obtain the predicted results of the video frames. The predicted results of the image are analyzed to obtain the recognition schemes for opening the self-rescuer cover, deploying the self-rescuer airbag, and wearing the self-rescuer nose clip. The process of analyzing the prediction results of the image in step S2 is to import the neural network model framework and load the trained model weights, use the model with loaded weights to infer the input video frame in the GPU environment, and the obtained inference results include the category and location information of the target. If there is a specific target category in the image, this category will be recorded. After noise reduction processing, the result will be passed to the self-rescuer test interface system for display. The neural network in step S2 adopts a single-stage target detection algorithm, specifically the YOLO series. This algorithm is known for its high efficiency and real-time performance. It can simultaneously predict the bounding box and category of the target in a single forward propagation, thereby achieving rapid target recognition. Target detection is used to identify images or video frames to determine the objects described in the image or stream. Each object has its category and location information. S3. The logic processing module makes a comprehensive judgment on the prediction results of the video frame according to the model, determines whether the operator has performed the operation of putting the strap on the neck, biting the mouthpiece into the mouth, opening the oxygen valve of the self-rescuer and pressing the compressed oxygen replenishment plate, and obtains the recognition scheme of putting the strap on the neck, biting the mouthpiece into the mouth, opening the oxygen valve of the self-rescuer and pressing the compressed oxygen replenishment plate in the self-rescuer operation. The process of comprehensive judgment in the S3 step is to use the deep learning model and the human key point detection model to predict the video frame respectively. When the result of the deep learning model predicting the video frame contains a predetermined target category, the detection frame position coordinates of the predetermined target category are obtained, and the position coordinates of the human key points selected by the human body are compared with the detection frame position coordinates or the derived frame generated by the detection frame position coordinates. If they are within the detection frame position area, it is considered that the operator has completed the predetermined action, and the recognition scheme of the predetermined action is obtained.

[0021] The self-rescuer identification scheme is transmitted to the self-rescuer testing and training system. The self-rescuer testing and training system identifies and judges the read video stream image through the camera according to the self-rescuer identification scheme; the real-time judgment result of the video stream will be synchronized to the display interface of the self-rescuer testing and training system. When it is detected that the operator has executed the current step, the self-rescuer system will automatically jump to the identification of the next step, and so on, until the entire self-rescuer operation process is completed.

[0022] The above is a schematic description of the present invention and its implementation methods, which is not restrictive. The accompanying drawings show only one implementation method of the present invention, and the actual structure is not limited thereto. Therefore, if a person skilled in the art is inspired by it and, without departing from the purpose of the invention, designs a structural method and an embodiment similar to the technical solution without creativity, they all belong to the protection scope of the present invention.

Claims

1. An AI self-rescuer training system, including a self-rescuer body, characterized in that: It also comprises an AI recognition workbench, the AI ​​recognition workbench comprising a main shell (7), a movable base (8) is installed below the main shell (7), an embedded soft light (1) is installed on the main shell (7), an image recognition module (2) is installed below the embedded soft light (1), a touch screen (3) is installed below the image recognition module (2), a power amplifier (4) is installed below the touch screen (3), a status indicator light (5) is installed below the power amplifier (4), and an identity recognition module (6) is installed below the status indicator light (5).

2. The AI ​​self-rescuer training system according to claim 1, characterized in that: The image recognition module (2) includes an AI camera for recognizing the examinee's movements.

3. The AI ​​self-rescuer training system according to claim 1, characterized in that: The embedded soft light (1) is used to provide lighting, the touch screen (3) is used to display the test-taking and training interface system, and the identity recognition module (6) is used to recognize the examinee information.

4. An identification method for an AI self-rescuer test and training system, characterized in that: include S1. Collecting self-rescuer wearing data through the image recognition module (1), and transmitting the collected data to the remote server for preprocessing; S2. Use the deep neural network to learn specific image features from the preprocessed data, predict the input image, map it to the corresponding input result, and analyze the predicted result of the image. The system can quickly and in real time determine whether the operator correctly performs the specific action of the self-rescuer in actual operation; S3. The logic processing module makes a comprehensive judgment on the prediction results of the video frame based on the model.

5. The identification method of the AI ​​self-rescuer training system according to claim 4 is characterized by: The self-rescuer wearing data collected in the S1 step is a video of different operators operating the self-rescuer. The recorded video is transmitted to a remote server via wireless communication, and the self-rescuer data is preprocessed. The preprocessing includes video frame extraction, data cleaning, and data enhancement. For the data containing the self-rescuer nose clip, it is also necessary to obtain the face position through the human key point detection model built into the MediaPipe computer vision library, and the face position of the source image is cropped to obtain a face image dataset containing the self-rescuer nose clip.

6. The identification method of the AI ​​self-rescuer training system according to claim 4 is characterized by: The specific learning process of using the deep neural network to learn specific image features in the S2 step is to annotate the image features and divide them into a training set and a verification set. The image features of the self-rescuer include a strap, a shell, a mouthpiece, an airbag and a nose clip. A deep neural network is selected as a model for training and verification data, and the video frames of the camera are input into the deep neural network model to obtain prediction results of the video frames. The prediction results of the image are analyzed to obtain recognition schemes for the operations of opening the self-rescuer cover, deploying the self-rescuer airbag and wearing the self-rescuer nose clip.

7. The identification method of the AI ​​self-rescuer training system according to claim 4 is characterized by: The process of analyzing the prediction results of the image in the S2 step is to import the neural network model framework and load the trained model weights, use the model with loaded weights to infer the input video frame in the GPU environment, and the obtained inference results include the category and location information of the target. If there is a specific target category in the image, this category will be recorded. After noise reduction processing, the result will be passed to the self-rescuer training interface system for display.

8. The identification method of the AI ​​self-rescuer training system according to claim 4 is characterized by: The neural network in step S2 adopts a single-stage target detection algorithm, specifically the YOLO series, which is known for its high efficiency and real-time performance. It can simultaneously predict the bounding box and category of the target in a single forward propagation, thereby achieving rapid target recognition. Target detection is used to identify images or video frames to determine the objects described in the image or stream, and each object has its category and location information.

9. The identification method of the AI ​​self-rescuer training system according to claim 4 is characterized by: The comprehensive judgment process in the S3 step is to use the deep learning model and the human key point detection model to predict the video frame respectively. When the result of the deep learning model predicting that the video frame contains a predetermined target category, the detection frame position coordinates of the predetermined target category are obtained, and the position coordinates of the artificially selected human key points are compared with the detection frame position coordinates or the derived frame generated by the detection frame position coordinates. If they are within the detection frame position area, it is considered that the operator has completed the predetermined action, and the recognition scheme of the predetermined action is obtained.

Images