Intelligent safety monitoring system based on YOLOv8

Abstract

The invention discloses an intelligent safety monitoring system based on YOLOv8, and the system comprises an RFID work card recognition module which is used for recognizing an RFID work card worn by a person entering a monitoring region, and judging the identity and access authority of the person according to the work card information; the camera equipment detection module is used for detecting whether camera equipment in use exists in the monitoring area or not and marking an abnormal area; the feature precision module is integrated in the YOLOv8 target detection model and is used for enhancing the detection precision of the model on the camera equipment target; the central processing unit is used for receiving detection results of the RFID work card identification module and the camera equipment detection module and triggering a corresponding response mechanism according to the detection results, and the response mechanism comprises movement track recording, alarming and response processing; according to the invention, the accuracy and efficiency of supervision work of workers are effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of security monitoring, and in particular to an intelligent security monitoring system based on YOLOv8. Background Art

[0002] With the continuous development of the information society, the security issues of some places are becoming increasingly prominent. Traditional security measures such as manual patrols and closed-circuit monitoring have problems such as slow response to emergencies, reliance on manual labor, and inability to analyze and judge in real time. Especially when facing outsiders hidden in complex environments, the recognition accuracy of traditional technologies is low, and the risk of false alarms and missed alarms is high. Object detection technology can classify objects in the video according to different categories, but due to the wide variety of camera equipment currently on the market, relying solely on video-based object detection to monitor the camera equipment being used in the venue will result in a very high missed detection rate and cannot guarantee the security of venue information. At the same time, the current object detection technology can only detect all categories of personnel, and cannot identify the identity of personnel, so it is impossible to determine whether the personnel appearing in the venue have relevant permissions. Traditional permission management is only applicable to fixed areas such as access control. For the periphery of a larger area, security personnel are mainly used for supervision, which requires a lot of human resources and has great security monitoring loopholes. Summary of the invention

[0003] Purpose of the invention: The purpose of the present invention is to provide an intelligent security monitoring system based on YOLOv8, which solves the problems existing in the background technology by comprehensively utilizing deep learning, target detection, sensors and radio frequency technology.

[0004] Technical solution: The intelligent safety monitoring system based on YOLOv8 described in the present invention includes: RFID badge recognition module, used to identify the RFID badge worn by personnel entering the monitoring area, and determine the identity and access rights of personnel based on the badge information; The camera device detection module is used to detect whether there are any camera devices in use in the monitoring area and mark abnormal areas; Feature precision module, integrated into the YOLOv8 target detection model, is used to enhance the model's detection accuracy for camera device targets; The central processing unit is used to receive the detection results of the RFID work badge recognition module and the camera detection module, and trigger the corresponding response mechanism according to the detection results, including activity track recording, alarm and response processing; Furthermore, the RFID work badge recognition module includes an RFID work badge, a reader, a camera and a buzzer. The reader is used to read the unique electronic product code EPC in the work badge and match it with the identity information in the identity recognition database to determine the identity and access rights of the personnel; the camera equipment detection module includes a FLIR Lepton thermal imaging infrared sensor and a HackRF One software-defined radio frequency module, which are used to detect high-temperature target areas and confirm whether they emit wireless signals; the feature precision module processes feature maps through the deeply separable spatial attention block DS-SAM and the dual-pooling squeeze excitation attention block DP-SE.

[0005] Furthermore, the RFID work badge uses a passive tag with an operating frequency band of ultra-high frequency 860-960MHz and an identification distance set to 10 meters; when the reader detects an unauthorized person, it triggers the corresponding buzzer to issue a warning and records the person's activity trajectory; when the reader detects an authorized person, it starts the corresponding camera to perform device target detection.

[0006] Furthermore, the camera device detection module screens high-temperature target areas by setting a temperature threshold, and confirms whether the target area transmits wireless signals in combination with radio frequency signal detection; if a wireless signal is detected, the area is marked as an abnormal area and its location information is recorded.

[0007] Furthermore, the feature precision module weights the feature map by spatial dimension through the deep separable spatial attention block DS-SAM, and enhances the interaction between channels through channel shuffling; the feature precision module fuses the global spatial information of the feature map through the double pooling squeeze excitation attention block DP-SE, generates channel weights, and multiplies them bit by bit with the original feature map to obtain the final feature map.

[0008] Furthermore, the feature precision module weights the feature map by spatial dimension through the deep separable spatial attention block DS-SAM, and enhances the interaction between channels through channel shuffling; the feature precision module fuses the global spatial information of the feature map through the double pooling squeeze excitation attention block DP-SE, generates channel weights, and multiplies them bit by bit with the original feature map to obtain the final feature map.

[0009] Furthermore, the specific process of the feature precision module is as follows: Set the input i batch feature maps, where C is the number of input channels, H is the height, W is the width, and after DS-SAM and DP-SE, the depth feature maps are obtained respectively. And pooling feature map , then and Add each element together to get the final discriminant feature. The formula is as follows: .

[0010] Furthermore, DS-SAM consists of two parts: spatial attention transformation and channel shuffling; the spatial attention transformation transforms the input i The batch feature maps are passed through two depth-wise separable convolutional layers to weight the importance of the spatial dimensions of the feature maps. The first depth-wise separable convolutional layer reduces the number of channels to , the second depth-wise separable convolutional layer increases the number of channels back to the original number of channels , after batch normalization and activation function operation, the feature map space importance representation is obtained, the formula is as follows: ; in, , They represent the feature maps of the input feature map after the first and second depth-separable convolutions, respectively. Indicates that all negative values ​​in the output results are set to zero, the sigmoid function The operation limits the output to the range [0,1]. At the same time, DS-SAM replaces the standard convolution with a depth-separable convolution; the feature map after spatial attention transformation Enter the channel shuffling part, divide the number of channels into 4 groups, and rearrange the channels of the 4 groups to obtain the feature map , and finally the original feature map And the feature map after channel shuffle Multiply bit by bit to get the final feature map.

[0011] Furthermore, the specific process of DP-SE is as follows: i Batch feature map The global spatial information of is compressed into and , fuse the double pooling results and generate channel weights through the fully connected layer Finally, the original feature map and the channel weight feature map are multiplied bit by bit to obtain the final feature map , the formula is as follows: ; in, Represents a fully connected layer operation.

[0012] Furthermore, the central processing unit triggers different levels of response mechanisms according to the detection results, including level one response, level two response and level three response; level one response is used to deal with personnel who are not wearing work badges and have video cameras detected around them, level two response is used to deal with unauthorized personnel, and level three response is used to deal with situations where authorized personnel have video cameras detected around them.

[0013] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages: the present invention performs preliminary detection of the camera equipment by integrating the FLIRLepton thermal imaging infrared sensor and the HackRF One software-defined radio frequency module, and confirms the identity of the personnel in the area through the RFID work badge recognition module, and then uses the YOLOv8 target detection model with the feature precision module added to perform device target detection, and conducts relevant alarms and responses, and designs an intelligent safety monitoring method and system to identify the camera equipment and personnel identities in use in real-time scenes, thereby realizing collaborative monitoring of abnormal conditions in places across multiple scenes; the present invention surpasses the traditional method of relying on manual supervision of places, effectively improves the accuracy and efficiency of the staff's supervision work, and greatly guarantees the information security of the place. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 It is a schematic diagram of the system flow of the present invention; Figure 2 is a flow chart of a camera device detection module of the present invention; Figure 3 It is a schematic diagram of a characteristic precision module of the present invention; Figure 4 It is a schematic diagram of the RFID work badge recognition module of the present invention. DETAILED DESCRIPTION

[0015] The technical solution of the present invention is further described below in conjunction with the accompanying drawings.

[0016] like Figure 1 As shown, an embodiment of the present invention provides an intelligent safety monitoring system based on YOLOv8, including: like Figure 4As shown in the figure, the RFID badge recognition module is used to identify the RFID badge worn by the personnel entering the monitoring area, and judge the identity and access rights of the personnel based on the badge information; specifically, the RFID (radio frequency identification) badge recognition module includes four parts: RFID badge, reader, camera and buzzer. The RFID badge is embedded with an RFID tag, which contains a unique electronic product code (EPC) corresponding to the person to whom the badge belongs. A passive tag is used, the operating frequency band is UHF 860-960MHz, and the recognition distance is set to 10 meters. The reader includes two parts: an antenna and a decoding module, which are used to read the tag data of the badge and transmit the decoded corresponding ID information to the system.

[0017] When a person enters the monitoring range of the reader, if the badge information is detected, the subsequent link will be directly entered for equipment target detection. If the badge information is detected, the RFID badge recognition function will be started. When the badge is close to the reader antenna, the RFID tag will receive the radio frequency signal emitted by the antenna. The passive tag obtains energy from the signal. The tag will transmit the unique EPC stored in the tag to the decoding module for decoding by wireless means. The decoded information will be transmitted to the system and matched with the identity information in the identity recognition database. The identity and access rights of the person are determined based on the matching results, thereby completing the identity authentication. If the badge wearer is identified as an unauthorized person, the buzzer corresponding to the reader will be turned on to warn the person to leave the area; if the badge wearer is identified as an authorized person, the camera corresponding to the reader will be turned on to perform equipment target detection in the area and further record the security status of the area.

[0018] At the same time, each reader is set with its corresponding number and location information when deployed. When an abnormal situation is detected, the activity trajectory of the person can be drawn through its location and time information, which is convenient for staff to verify and trace.

[0019] The camera device detection module is used to detect whether there are any cameras in use in the monitoring area and mark abnormal areas. The camera device detection module includes a FLIR Lepton thermal imaging infrared sensor and a HackRF One software-defined radio frequency module, which are used to detect high-temperature target areas and confirm whether they emit wireless signals. The camera device detection module screens high-temperature target areas by setting temperature thresholds, and confirms whether the target area emits wireless signals in combination with radio frequency signal detection. If a wireless signal is detected, the area is marked as an abnormal area and its location information is recorded.

[0020] The details are as follows: Figure 2As shown, the FLIR Lepton thermal imaging infrared sensor detects the real-time scene area, obtains thermal map data through the SPI interface, uses OpenCV to process thermal images, and screens out high-temperature target areas according to the set temperature threshold and marks them; the HackRF One software-defined radio RF module uses the Fast Fourier Transform (FFT) through the Python interface to perform spectrum analysis on the RF signal in the marked area of ​​the infrared sensor, detects strong signals in the 2.4GHz frequency band, and confirms whether the target area transmits wireless signals. If a wireless signal is detected, the area is marked as an abnormal area and its location information is recorded. The details are as follows: The feature precision module is integrated into the YOLOv8 target detection model to enhance the model's detection accuracy for camera device targets; the feature precision module processes feature maps through the deeply separable spatial attention block DS-SAM and the dual-pooling squeeze excitation attention block DP-SE; The feature precision module weights the spatial dimension of the feature map through the deep separable spatial attention block DS-SAM, and enhances the interaction between channels through channel shuffling; the feature precision module fuses the global spatial information of the feature map through the double pooling squeeze excitation attention block DP-SE, generates channel weights, and multiplies them bit by bit with the original feature map to obtain the final feature map.

[0021] The specific process of the feature precision module is as follows: Set the input i Batch feature map ,in C is the number of input channels, H is the height, W is the width, and after DS-SAM and DP-SE, the depth feature maps are obtained respectively. And pooling feature map , then and Add element by element to get the final discriminant feature , the formula is as follows: .

[0022] DS-SAM consists of two parts: spatial attention transformation and channel shuffling; the spatial attention transformation transforms the input i The batch feature maps are passed through two depth-wise separable convolutional layers to weight the importance of the spatial dimensions of the feature maps. The first depth-wise separable convolutional layer reduces the number of channels to , the second depth-wise separable convolutional layer increases the number of channels back to the original number of channels , after batch normalization and activation function operation, the feature map space importance representation is obtained, the formula is as follows: ; in, , They represent the feature maps of the input feature map after the first and second depth-separable convolutions, respectively. Indicates that all negative values ​​in the output results are set to zero, the sigmoid function The operation limits the output to the range [0,1]. At the same time, DS-SAM replaces the standard convolution with a depth-separable convolution; the feature map after spatial attention transformation Enter the channel shuffling part, divide the number of channels into 4 groups, and rearrange the channels of the 4 groups to obtain the feature map , and finally the original feature map And the feature map after channel shuffle Multiply bit by bit to get the final feature map.

[0023] The specific process of DP-SE is as follows: i Batch feature map The global spatial information of is compressed into and , fuse the double pooling results and generate channel weights through the fully connected layer Finally, the original feature map and the channel weight feature map are multiplied bit by bit to obtain the final feature map , the formula is as follows: ; in, Represents a fully connected layer operation.

[0024] The central processing unit is used to receive the detection results of the RFID work badge recognition module and the camera device detection module, and trigger the corresponding response mechanism according to the detection results, including activity trajectory recording, alarm and response processing; the central processing unit triggers different levels of response mechanisms according to the detection results, including primary response, secondary response and tertiary response; the primary response is used to deal with personnel who are not wearing work badges and have camera devices detected around them, the secondary response is used to deal with unauthorized personnel, and the tertiary response is used to deal with authorized personnel but camera devices are detected around them. Specifically: if a device is detected around a person who is not wearing a work badge, the primary response is immediately triggered; if a device is detected around a person wearing an unauthorized work badge, the secondary response is immediately triggered. If a device is detected around a person wearing an authorized work badge, the tertiary response is immediately triggered. The identity of the person who triggers the second and third level responses is locked, the location information of the work badge is dynamically recorded between multiple readers, and the time of entering and leaving a specific area is analyzed using timestamps to generate activity trajectories. When the first-level response is triggered, an alarm signal is immediately issued, and relevant security personnel are notified to go to the alarm area for investigation through mobile phones, text messages or voice broadcasts; when the second-level response is triggered, relevant information is sent to the corresponding superior of the person to verify the person's movement trajectory. If the person has three or more abnormal situations, the intermediate leader will be notified through mobile phones, text messages, etc. to verify; when the third-level response is triggered, the person needs to take the initiative to go to the relevant security department to verify the shooting information. The abnormal video frames of the system's responses at all levels are temporarily saved, and the frames are reassembled after the response disappears to generate abnormal videos, which are saved in the system database according to the time and camera location information for easy reference and call later.

Claims

1. An intelligent security monitoring system based on YOLOv8, characterized in that: include: The RFID badge recognition module is used to identify the RFID badges worn by people entering the monitoring area, and determine the identity and access rights of the people based on the badge information; the camera detection module is used to detect whether there are any cameras in use in the monitoring area and mark abnormal areas; the feature precision module is integrated into the YOLOv8 target detection model to enhance the model's detection accuracy for camera targets; the central processing unit is used to receive the detection results of the RFID badge recognition module and the camera detection module, and trigger the corresponding response mechanism based on the detection results, including activity trajectory recording, alarm and response processing.

2. According to claim 1, a YOLOv8-based intelligent security monitoring system is characterized in that: The RFID work badge recognition module includes an RFID work badge, a reader, a camera and a buzzer. The reader is used to read the unique electronic product code EPC in the work badge and match it with the identity information in the identity recognition database to determine the identity and access rights of the personnel; the camera equipment detection module includes a FLIR Lepton thermal imaging infrared sensor and a HackRF One software-defined radio frequency module, which are used to detect high-temperature target areas and confirm whether they emit wireless signals; the feature precision module processes feature maps through the deeply separable spatial attention block DS-SAM and the dual-pooling squeeze excitation attention block DP-SE.

3. According to the YOLOv8-based intelligent security monitoring system of claim 1, it is characterized in that: The RFID work badge uses a passive tag with an operating frequency of UHF 860-960MHz and an identification distance of 10 meters. When the reader detects an unauthorized person, it triggers the corresponding buzzer to issue a warning and records the person's activity trajectory. When the reader detects an authorized person, it activates the corresponding camera for device target detection.

4. The intelligent safety monitoring system based on YOLOv8 according to claim 1, characterized in that: The camera device detection module screens high-temperature target areas by setting temperature thresholds, and combines radio frequency signal detection to confirm whether the target area transmits wireless signals; if wireless signals are detected, the area is marked as an abnormal area and its location information is recorded.

5. The intelligent safety monitoring system based on YOLOv8 according to claim 1, characterized in that: The feature precision module weights the spatial dimensions of feature maps through the deep separable spatial attention block DS-SAM and enhances the interaction between channels through channel shuffling; The feature precision module performs global spatial information fusion on the feature map through the double pooling squeeze excitation attention block DP-SE, generates channel weights, and multiplies them bit by bit with the original feature map to obtain the final feature map.

6. The intelligent safety monitoring system based on YOLOv8 according to claim 5, characterized in that: The specific process of the feature precision module is as follows: Set the input i batch feature maps, where C is the number of input channels, H is the height, W is the width, and after DS-SAM and DP-SE, the depth feature maps are obtained respectively. And pooling feature map , then and Add each element together to get the final discriminant feature. The formula is as follows: 。 7. The intelligent safety monitoring system based on YOLOv8 according to claim 6, characterized in that: DS-SAM consists of two parts: spatial attention transformation and channel shuffling; the spatial attention transformation transforms the input i The batch feature maps are passed through two depth-wise separable convolutional layers to weight the importance of the spatial dimensions of the feature maps. The first depth-wise separable convolutional layer reduces the number of channels to , the second depth-wise separable convolutional layer increases the number of channels back to the original number of channels , after batch normalization and activation function operation, the feature map space importance representation is obtained, the formula is as follows: ； in, , They represent the feature maps of the input feature map after the first and second depth-separable convolutions, respectively. Indicates that all negative values ​​in the output results are set to zero, the sigmoid function The operation limits the output to the range [0,1]. At the same time, DS-SAM replaces the standard convolution with a depth-separable convolution; the feature map after spatial attention transformation Enter the channel shuffling part, divide the number of channels into 4 groups, and rearrange the channels of the 4 groups to obtain the feature map , and finally the original feature map And the feature map after channel shuffle Multiply bit by bit to get the final feature map.

8. The intelligent safety monitoring system based on YOLOv8 according to claim 6, characterized in that: The specific process of DP-SE is as follows: i Batch feature map The global spatial information of is compressed into and , fuse the double pooling results and generate channel weights through the fully connected layer Finally, the original feature map and the channel weight feature map are multiplied bit by bit to obtain the final feature map , the formula is as follows: ； in, Represents a fully connected layer operation.

9. The intelligent safety monitoring system based on YOLOv8 according to claim 1, characterized in that: The central processing unit triggers different levels of response mechanisms based on the detection results, including level one, level two and level three responses; level one response is used to deal with people who are not wearing work badges and have cameras detected around them, level two response is used to deal with unauthorized people, and level three response is used to deal with situations where authorized people have cameras detected around them.

Images