Mining cage control equipment and control method based on embedded AI board card

Abstract

The invention discloses mining cage control equipment and a control method based on an embedded AI board card, and relates to the technical field of mining equipment, the mining cage control equipment comprises a cage body, and further comprises a mining image processing instrument and an industrial control terminal, and the mining image processing instrument is fixed at a wellhead of the cage body; the mining image processing instrument comprises a camera, an infrared board, an AI card board and an embedded bottom board. The camera and the infrared board are used for monitoring the access condition of personnel at the wellhead in real time and transmitting a video stream to the AI card board; the AI card board is used for analyzing and identifying a video stream and transmitting a signal to the embedded bottom board; the embedded bottom plate is used for receiving the signal transmitted by the AI card board; the problems that in the prior art, a well mouth operator needs to manually open and manually count the number of people in a cage, operation is slow, counting of the number of people is not clear, and overload exists are solved, and meanwhile the problems that a conventional AI product needs to be provided with equipment such as a switch and a server and cannot be linked with industrial control equipment such as a field gate machine door are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of mining equipment, and in particular to a mining cage control device and a control method based on an embedded AI board. Background Art

[0002] Cage is a kind of mine hoisting container, which can generally carry several tons and has similar uses as elevators. It is mainly used for auxiliary shaft hoisting in mines to hoist personnel, ores, equipment, materials, etc. For medium and small mines, cages can also be used for main shaft hoisting to hoist coal. Cage is one of the important equipment in mine hoisting. There are many types of cages. According to the number of hoisting wire ropes, it can be divided into single-rope cage and multi-rope cage. According to the number of cage layers, it can be divided into single-layer, double-layer and multi-layer cages. However, in the process of using metal mine cage hoist system, the following problems still exist:

[0003] At present, the cage doors of mines need to be opened and closed manually by the wellhead operators, and the number of people in the cage needs to be manually counted. This results in slow operation, unclear headcount, overload and other problems. At the same time, conventional AI products need to be configured with switches, servers and other equipment, and cannot be linked with on-site industrial control equipment such as gates. Summary of the invention

[0004] The embodiment of the present application provides a mining cage control device based on an embedded AI board, which solves the problem that the cage door of the mine needs to be manually opened and closed by the wellhead operator and the number of people in the cage needs to be manually counted, resulting in slow operation, unclear headcount, overload and other problems. At the same time, conventional AI products need to be configured with switches, servers and other equipment, and cannot be linked with on-site industrial control equipment such as gate machines.

[0005] The embodiment of the present application provides a mining cage control device based on an embedded AI board, including a cage body, a mining image processor and an industrial control terminal, wherein the mining image processor is fixed at the wellhead of the cage body;

[0006] The mining image processor comprises a camera, an infrared board, an AI card board and an embedded base plate;

[0007] Cameras and infrared boards are used to monitor the entry and exit of personnel at the wellhead in real time and transmit the video stream to the AI ​​card board;

[0008] AI card board, used to analyze and identify video streams and transmit signals to the embedded baseboard;

[0009] The embedded baseboard is used to receive signals from the AI ​​card board and communicate with the industrial control terminal.

[0010] Preferably, the industrial control terminal is electrically connected to the gate.

[0011] Preferably, the embedded baseboard is connected to the industrial control terminal via RS485, RJ45.

[0012] Preferably, the embedded base plate has a switch input detection circuit and a relay output circuit;

[0013] The industrial control terminal transmits the signal after the gate is opened to the embedded base plate through the switch input detection circuit;

[0014] The embedded base plate transmits the signal of personnel entering and leaving the cage to the industrial control terminal through the relay output circuit. The industrial control terminal determines whether to close the gate according to whether the signal is overloaded.

[0015] Preferably, the control method of the mining cage control device based on the embedded AI board is used to realize real-time detection and tracking of pedestrians, count the total number of people appearing in the video, and count the pedestrians crossing the horizontal line. The control method comprises the following steps:

[0016] S1. Collect pedestrian videos in multiple scenes through cameras, extract key frame images of the videos, establish pedestrian datasets, and divide the datasets into training sets, validation sets, and test sets;

[0017] S2, input the images in the training set and the validation set into the YOLOV5 network model for model training, and obtain the optimal weight data of the network model after the training;

[0018] S3, load the optimal weight data into the YOLOV5 network model, input the test set image, and output the target detection result;

[0019] S4. Input the target detection result of YOLOV5 into DeepSort, rely on the Kalman filter to predict the trajectory, use the Hungarian algorithm to match the trajectory data in the frame obtained by the predicted trajectory with the trajectory combination in the current frame, and update the Kalman algorithm filter;

[0020] S5. Set a baseline in the video screen by giving the coordinates of both ends of the line. Use this line as the axis to determine the direction of the pedestrians, determine the number of people entering and leaving in both directions, and create a tracker to start tracking the target detected by YOLOV5.

[0021] S6. The straight line formed by the center points of the two frames before and after the video intersects with the pre-set baseline. If the pedestrian's target position anchor point has exceeded the intersection point at this time, it is determined that the pedestrian has crossed the line. The tangent and inverse tangent principles of the triangle are used, and the degrees method in the math module is used to judge. If the angle between the pedestrian's anchor point and the line connecting the above intersection points is greater than zero, it means that the pedestrian has entered the cage, otherwise it means that the pedestrian has exited the cage.

[0022] S7. Draw bounding boxes and trajectories on the video frames to display the number of pedestrians entering and exiting.

[0023] Preferably, the YOLOV5 network model includes an input end, a Backbone network, a Neck network and a Head output end connected in sequence.

[0024] Preferably, in step S2, the method of inputting the images in the training set and the validation set into the YOLOV5 network model for model training is:

[0025] S2-1, Image preprocessing: When the image enters the input end, the input end preprocesses the image, including Mosaic data enhancement, adaptive anchor box calculation and adaptive image scaling;

[0026] S2-2, Feature extraction: The preprocessed image enters the Backbone network, and after feature extraction, three feature maps of different sizes are obtained;

[0027] S2-3, feature fusion: The three feature maps of different sizes obtained in step S2-2 are input into the Neck network for feature fusion to obtain feature maps of three scales;

[0028] S2-4, prediction output: The feature maps of the three scales obtained in step S2-3 are input into the Head output end to obtain three feature tensors, thereby obtaining the position, category and confidence of the prediction box. In the training phase, the prediction box is matched with the real box to obtain positive and negative samples, and then the weight parameters are adjusted by calculating the loss function. In the verification phase, the prediction box is screened by weighted non-maximum suppression, and the model parameters are continuously optimized by calculating the accuracy and average precision.

[0029] One or more technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages:

[0030] By setting up a mining image processor and an industrial control terminal, the number of people entering and leaving the cage can be identified and analyzed, and the number of people can be quickly counted, which is clear and concise. The gate will close automatically, saving personnel working time and quick to operate. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 The present invention is a schematic diagram of the structure of a mining cage control device based on an embedded AI board. DETAILED DESCRIPTION

[0032] To facilitate the understanding of the present invention, the present application will be described more comprehensively below with reference to the relevant drawings; the drawings show preferred embodiments of the present invention, but the present invention can be implemented in many different forms and is not limited to the embodiments described herein; on the contrary, the purpose of providing these embodiments is to enable a more thorough and comprehensive understanding of the disclosed content of the present invention.

[0033] It should be noted that the terms “vertical”, “horizontal”, “up”, “down”, “left”, “right” and similar expressions used in this document are only for illustrative purposes and do not represent the only implementation method.

[0034] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by technicians in the technical field to which the present invention belongs; the terms used in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention; the term "and / or" used herein includes any and all combinations of one or more related listed items.

[0035] like Figure 1 As shown, the present application provides a mining cage control device based on an embedded AI board, including a cage body, a mining image processor and an industrial control terminal, wherein the mining image processor is fixed at the wellhead of the cage body; the mining image processor includes a camera, an infrared board, an AI card board and an embedded bottom plate; the camera and the infrared board are installed at the wellhead of the cage body; the camera and the infrared board are used to monitor the entry and exit of personnel at the wellhead in real time, and transmit the video stream to the AI ​​card board; the AI ​​card board is used to analyze and identify the video stream and transmit the signal to the embedded bottom plate; wherein the AI ​​card board mainly analyzes the information signals such as the number of personnel entering and exiting the cage at the wellhead of the cage body, the number of people in the area and the status of personnel, and transmits the signal to the embedded bottom plate, and after receiving the signal, the embedded bottom plate transmits the signal to the industrial control terminal. The embedded bottom plate is used to receive the signal transmitted by the AI ​​card board and is connected to the industrial control terminal. The embedded bottom plate receives the signal transmitted by the AI ​​card board and transmits the signal to the industrial control terminal, and the industrial control terminal determines whether to close the gate according to the signal.

[0036] Specifically, the industrial control terminal is electrically connected to the gate door. After receiving the start command from the operator, the industrial control terminal controls the gate door of the employee passage to open, and at the same time, the industrial control terminal transmits the start signal of the gate door to the embedded baseboard.

[0037] Specifically, the embedded baseboard is connected to the industrial control terminal through RS485 and RJ45. RS485 and RJ45 are communication circuits on the embedded baseboard.

[0038] Specifically, the embedded base plate has a switch input detection circuit and a relay output circuit;

[0039] The industrial control terminal transmits the signal after the gate is opened to the embedded base plate through the switch input detection circuit;

[0040] The embedded base plate transmits the signal of personnel entering and leaving the cage to the industrial control terminal through the relay output circuit, and the industrial control terminal determines whether to close the gate door according to whether the signal is overloaded. After the industrial control terminal controls the gate door to open, it synchronously transmits the switch signal to the embedded base plate through the switch detection circuit; the AI ​​card board transmits the personnel signal at the wellhead to the embedded base plate, and the embedded base plate transmits the overload signal to the industrial control terminal through the relay output circuit, and the industrial control terminal determines whether to close the gate door according to whether the signal is overloaded.

[0041] According to another aspect of an embodiment of the present invention, a control method for a mining cage control device based on an embedded AI board is provided, wherein the control method is used to realize real-time detection and tracking of pedestrians, count the total number of people appearing in the video, and count pedestrians crossing the horizontal line. The control method comprises the following steps:

[0042] S1. Collect pedestrian videos in multiple scenes through cameras, extract key frame images of the videos, establish pedestrian datasets, and divide the datasets into training sets, validation sets, and test sets;

[0043] S2, input the images in the training set and the validation set into the YOLOV5 network model for model training, and obtain the optimal weight data of the network model after the training;

[0044] S3, load the optimal weight data into the YOLOV5 network model, input the test set image, and output the target detection result;

[0045] S4. Input the target detection result of YOLOV5 into DeepSort, rely on the Kalman filter to predict the trajectory, use the Hungarian algorithm to match the trajectory data in the frame obtained by the predicted trajectory with the trajectory combination in the current frame, and update the Kalman algorithm filter;

[0046] S5. Set a baseline in the video screen by giving the coordinates of both ends of the line. Use this line as the axis to determine the direction of the pedestrians, determine the number of people entering and leaving in both directions, and create a tracker to start tracking the target detected by YOLOV5.

[0047] S6. The straight line formed by the center points of the two frames before and after the video intersects with the pre-set baseline. If the pedestrian's target position anchor point has exceeded the intersection point at this time, it is determined that the pedestrian has crossed the line. The tangent and inverse tangent principles of the triangle are used, and the degrees method in the math module is used to judge. If the angle between the pedestrian's anchor point and the line connecting the above intersection points is greater than zero, it means that the pedestrian has entered the cage, otherwise it means that the pedestrian has exited the cage.

[0048] S7. Draw bounding boxes and trajectories on the video frames to display the number of pedestrians entering and exiting.

[0049] Specifically, the YOLOV5 network model includes an input end, a Backbone network, a Neck network and a Head output end connected in sequence.

[0050] Specifically, in step S2, the method of inputting the images in the training set and the validation set into the YOLOV5 network model for model training is:

[0051] S2-1, Image preprocessing: When the image enters the input end, the input end preprocesses the image, including Mosaic data enhancement, adaptive anchor box calculation and adaptive image scaling;

[0052] S2-2, Feature extraction: The preprocessed image enters the Backbone network, and after feature extraction, three feature maps of different sizes are obtained;

[0053] S2-3, feature fusion: The three feature maps of different sizes obtained in step S2-2 are input into the Neck network for feature fusion to obtain feature maps of three scales;

[0054] S2-4, prediction output: The feature maps of the three scales obtained in step S2-3 are input into the Head output end to obtain three feature tensors, thereby obtaining the position, category and confidence of the prediction box. In the training phase, the prediction box is matched with the real box to obtain positive and negative samples, and then the weight parameters are adjusted by calculating the loss function. In the verification phase, the prediction box is screened by weighted non-maximum suppression, and the model parameters are continuously optimized by calculating the accuracy and average precision.

[0055] Working principle: After receiving the start command from the operator, the industrial control terminal controls the gate door to open, and simultaneously transmits the switch signal to the embedded bottom board through the switch detection circuit on the embedded bottom board; in addition, the camera and infrared board obtain the current monitoring video stream in real time and transmit the signal to the AI ​​card board. The AI ​​card board mainly analyzes the information signals such as the number of people entering and leaving the cage at the wellhead of the cage body, the number of people in the area and the status of people, and transmits the analysis signal to the embedded bottom board. After receiving the signal, the embedded bottom board transmits the signal to the industrial control terminal through the relay output circuit. The industrial control terminal determines whether to close the gate door based on whether the signal is overloaded; when the number of people entering the cage reaches the set maximum number of people in the cage, the embedded bottom board outputs an action instruction through the on-board relay. After receiving the instruction, the external industrial control terminal controls the gate door of the employee channel to close, prohibiting employees from entering the cage, thereby achieving the purpose of automatic closing of the gate door, saving personnel working time and quick operation, and the number of people entering the cage is clear.

[0056] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A mining cage control device based on an embedded AI board, comprising a cage body, characterized in that: It also includes a mining image processor and an industrial control terminal, wherein the mining image processor is fixed at the wellhead of the cage body; The mining image processor comprises a camera, an infrared board, an AI card board and an embedded base plate; Cameras and infrared boards are used to monitor the entry and exit of personnel at the wellhead in real time and transmit the video stream to the AI ​​card board; AI card board, used to analyze and identify video streams and transmit signals to the embedded baseboard; The embedded baseboard is used to receive signals from the AI ​​card board and communicate with the industrial control terminal.

2. The mining cage control device based on embedded AI board as claimed in claim 1, characterized in that: The industrial control terminal is electrically connected to the gate.

3. The mining cage control device based on embedded AI board as claimed in claim 1, characterized in that: The embedded baseboard is connected to the industrial control terminal through RS485 and RJ45 communication.

4. The mining cage control device based on embedded AI board as claimed in claim 1, characterized in that: The embedded bottom plate is provided with a switch quantity input detection circuit and a relay output circuit; The industrial control terminal transmits the signal after the gate is opened to the embedded base plate through the switch input detection circuit; The embedded base plate transmits the signal of personnel entering and leaving the cage to the industrial control terminal through the relay output circuit. The industrial control terminal determines whether to close the gate according to whether the signal is overloaded.

5. A control method for a mining cage control device based on an embedded AI board as described in claims 1-4, characterized in that: The control method is used to realize real-time detection and tracking of pedestrians, count the total number of people appearing in the video, and count the pedestrians crossing the horizontal line. The control method includes the following steps: S1. Collect pedestrian videos in multiple scenes through cameras, extract key frame images of the videos, establish a pedestrian dataset, and divide the dataset into training set, verification set and test set; S2, input the images in the training set and the validation set into the YOLOV5 network model for model training, and obtain the optimal weight data of the network model after the training; S3, load the optimal weight data into the YOLOV5 network model, input the test set image, and output the target detection result; S4. Input the target detection result of YOLOV5 into DeepSort, rely on the Kalman filter to predict the trajectory, use the Hungarian algorithm to match the trajectory data in the frame obtained by the predicted trajectory with the trajectory combination in the current frame, and update the Kalman algorithm filter; S5. Set a baseline in the video screen by giving the coordinates of both ends of the line. Use this line as the axis to determine the direction of the pedestrians, determine the number of people entering and leaving in both directions, and create a tracker to start tracking the target detected by YOLOV5. S6. The straight line formed by the center points of the two frames before and after the video intersects with the pre-set baseline. If the pedestrian's target position anchor point has exceeded the intersection point at this time, it is determined that the pedestrian has crossed the line. The tangent and inverse tangent principles of the triangle are used, and the degrees method in the math module is used to judge. If the angle between the pedestrian's anchor point and the line connecting the above intersection points is greater than zero, it means that the pedestrian has entered the cage, otherwise it means that the pedestrian has exited the cage. S7. Draw bounding boxes and trajectories on the video frames to display the number of pedestrians entering and exiting.

6. The control method of the mining cage control device based on the embedded AI board as claimed in claim 5, characterized in that: The YOLOV5 network model includes an input end, a Backbone network, a Neck network and a Head output end which are connected in sequence.

7. The control method of the mining cage control device based on the embedded AI board as claimed in claim 5, characterized in that: In the step S2, the method of inputting the images in the training set and the validation set into the YOLOV5 network model for model training is: S2-1, Image preprocessing: When the image enters the input end, the input end preprocesses the image, including Mosaic data enhancement, adaptive anchor box calculation and adaptive image scaling; S2-2, Feature extraction: The preprocessed image enters the Backbone network, and after feature extraction, three feature maps of different sizes are obtained; S2-3, feature fusion: The three feature maps of different sizes obtained in step S2-2 are input into the Neck network for feature fusion to obtain feature maps of three scales; S2-4, prediction output: The feature maps of the three scales obtained in step S2-3 are input into the Head output end to obtain three feature tensors, thereby obtaining the position, category and confidence of the prediction box. In the training phase, the prediction box is matched with the real box to obtain positive and negative samples, and then the weight parameters are adjusted by calculating the loss function. In the verification phase, the prediction box is screened by weighted non-maximum suppression, and the model parameters are continuously optimized by calculating the accuracy and average precision.

Images