Method for detecting intrusion of personnel in a hazardous area

Abstract

The application discloses a dangerous area personnel intrusion detection method, comprising the following steps: acquiring image data and preprocessing video; detecting the position and confidence of personnel; continuously tracking the personnel target; comprehensively analyzing the effectiveness of the target trajectory; and judging personnel intrusion to trigger an alarm mechanism. The application realizes real-time personnel intrusion detection of the workshop automation station by combining the YOLOv8 personnel detection algorithm, the Kalman filter and the target tracking technology of the Hungarian algorithm, and the target effectiveness judgment mechanism, constructs a real-time and continuous monitoring process, ensures real-time monitoring and personnel intrusion detection of the workshop automation station, effectively improves the detection precision, reduces the false alarm rate, guarantees production safety, has high detection precision and low false alarm rate in a complex scene, can effectively guarantee the production safety of the workshop, and can also adjust the algorithm parameters and threshold according to the actual field demand, and adapt to different detection frame rate requirements.

Description

Technical Field

[0001] This invention relates to the field of personnel intrusion detection technology, specifically to a method for detecting personnel intrusion in dangerous areas. Background Technology

[0002] In modern industrial production, the safety of automated workstations in workshops is of paramount importance. With the increasing level of automation, these areas often contain high-speed machinery, high-temperature equipment, or hazardous materials. Accidental entry by personnel may lead to serious safety accidents. Therefore, real-time personnel intrusion detection of automated workstations is of great significance. Existing personnel detection technologies mainly rely on traditional image recognition and multi-target tracking technologies. Image recognition technology detects personnel by analyzing features in images, while multi-target tracking technology continuously tracks targets after detection. The closest existing technology is a personnel detection and tracking system based on deep learning, which uses convolutional neural networks to extract features from images and detect personnel, and then combines algorithms such as Kalman filtering for target tracking.

[0003] Current image recognition technology accuracy is limited by image quality and interference from complex backgrounds. Existing multi-target tracking technology is prone to losing targets and generating false tracking when targets are occluded or lighting changes. In practical applications, deep learning-based personnel detection and tracking systems are not adaptable enough to cope with complex scenes, cannot effectively distinguish between real people and ghosts, and are prone to creating dummy images under light reflection, resulting in a high false alarm rate. Summary of the Invention

[0004] This invention provides a method for detecting intrusion into dangerous areas, which can effectively solve the problems mentioned in the background art. The accuracy of current image recognition technology is limited by image quality and interference from complex backgrounds. Existing multi-target tracking technology is prone to losing targets and generating false tracking when targets are occluded or lighting changes. In practical applications, deep learning-based personnel detection and tracking systems are not adaptable enough to cope with complex scenes, cannot effectively distinguish between real people and ghosts, and are prone to creating dummy images under light reflection, resulting in a high false alarm rate.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for detecting personnel intrusion in dangerous areas, which improves the accuracy and reliability of personnel intrusion detection and reduces the false alarm rate through in-depth analysis of spatiotemporal relationships and in-depth judgment of target effectiveness using detection and tracking algorithms, including the following steps:

[0006] S1. Acquire image data and preprocess the video;

[0007] S2, Location and confidence level of the testing personnel;

[0008] S3. Continuously track personnel targets;

[0009] S4. Comprehensively analyze the effectiveness of the target trajectory;

[0010] S5. Determine if personnel intrusion is detected and trigger the alarm mechanism;

[0011] S1 involves acquiring a high-quality video stream to provide clear image data for subsequent personnel detection and tracking, and preprocessing the acquired video stream.

[0012] S2 uses the YOLOv8 algorithm to detect the location and confidence level of people in the image;

[0013] S3 involves continuously tracking the detected personnel targets and maintaining their trajectories.

[0014] S4 analyzes the effectiveness of the target trajectory and distinguishes between real people and ghosts;

[0015] S5 triggers an alarm mechanism when it determines that a real person has entered a dangerous area, and notifies relevant personnel to take measures through an alarm signal.

[0016] According to the above technical solution, step S1 specifically includes the following steps:

[0017] S101, Data Acquisition;

[0018] S102, Image preprocessing;

[0019] S101 involves installing a high-definition camera at a suitable location in the automated workstation of the workshop to ensure coverage of the hazardous area that needs to be monitored. The resolution of the high-definition camera is not less than 1920×1080 pixels, and the camera parameters, including exposure time, gain and focal length, also need to be adjusted.

[0020] The exposure time is adjusted according to the actual lighting conditions, between 10 and 100 microseconds, to ensure moderate image brightness. The gain is adjusted according to the image brightness, between 0 and 20 dB, to avoid the image being too bright or too dark. The focal length is adjusted according to the size and distance of the monitoring area to ensure image clarity.

[0021] During the data acquisition process, it is also necessary to record the lighting conditions and camera position when the high-definition camera is shooting, so as to facilitate subsequent processing. The current frame image is extracted from the video stream, and the acquired video stream is converted into an image sequence for preprocessing.

[0022] According to the above technical solution, in step S102, image preprocessing includes grayscale conversion and noise reduction.

[0023] Grayscale conversion is the process of converting a color image to a grayscale image, reducing the amount of data and computational complexity. The formula is:

[0024] Igray = 0.299 × R + 0.587 × G + 0.114 × B;

[0025] Where R, G, and B are the pixel values ​​of the red, green, and blue channels of the color image, respectively;

[0026] Denoising is achieved by using Gaussian filtering to remove noise from an image. The Gaussian filtering formula is:

[0027]

[0028] Where G(i,j) is the Gaussian kernel function and k is the filtering radius;

[0029] The preprocessed grayscale image sequence is denoted as I. pre .

[0030] According to the above technical solution, step S2 specifically includes the following steps:

[0031] S201. Input the preprocessed image;

[0032] S202, Network Reasoning;

[0033] S203, Spatiotemporal Relationship Analysis;

[0034] In step S201, the input preprocessed image is the preprocessed grayscale image sequence I. pre Input into the trained YOLOv8 network;

[0035] In S202, the YOLOv8 network performs feature extraction and target detection on images through convolutional layers, pooling layers, and fully connected layers;

[0036] Output the detection box and its corresponding confidence score. The detection box is represented as a rectangle: (x, y, w, h).

[0037] Where (x,y) is the coordinate of the top left corner, w is the width, h is the height, and the confidence level represents the probability of detecting a person.

[0038] According to the above technical solution, in step S203, the motion trend and spatial position changes of the target are analyzed by combining the spatiotemporal relationship between consecutive frames to improve detection accuracy. For the detection results of frame t and frame t+1, the displacement and velocity changes between the targets are calculated using the following formula:

[0039] Δx=x t+1 -x t Δy=y t+1 -y t ,

[0040] Where Δt is the time interval between two frames, denoted as D after the detected personnel location information and confidence level. t ={(x i ,y i ,w i ,h i ,c i )}, where c i , where is the confidence level.

[0041] According to the above technical solution, step S3 specifically includes the following steps:

[0042] S301. Initialize the Kalman filter;

[0043] S302, Target Prediction and Matching;

[0044] S303, Track Maintenance and Update;

[0045] In step S301, for each detection box (x, y, w, h) in the first frame, the state vector of the Kalman filter is initialized as x = [x, y, w, h, v]. x ,v y ] T ;

[0046] Among them, v x and v y The initial velocity is set to 0.

[0047] The state transition matrix F and the measurement matrix H are respectively:

[0048]

[0049] In step S302, the Kalman filter is used to predict the target's position in subsequent frames, as follows:

[0050] x pred =F×x prev ;

[0051] The distance cost matrix between predicted and detected bounding boxes is calculated using the Hungarian algorithm for matching, minimizing the total cost, and is calculated by the following formula:

[0052]

[0053] Where, x pred,i y pred,i Let x be the coordinates of the predicted bounding box. det,j y det,j These are the coordinates of the detection box.

[0054] According to the above technical solution, in step S303, for a successfully matched target, the state of the Kalman filter is updated by the following formula:

[0055] x new =x pred +K×(zH×x pred );

[0056] Where K is the Kalman gain and z is the measured value;

[0057] For unmatched targets, they are placed in the waiting pool and survive for 3 frames. If a target in the waiting pool can be matched with another target in the waiting pool within 3 frames, it is considered meaningful; otherwise, it is deleted. When a trajectory fails to match a newly detected target for 3 consecutive frames, the trajectory is considered to have ended its life.

[0058] The maintained target trajectory information, including position, velocity, and direction, is denoted as T. t ={x1,x2,...,x n}

[0059] According to the above technical solution, step S4 specifically includes the following steps:

[0060] S401, Trajectory Continuity Analysis;

[0061] S402, Motion Characteristic Analysis;

[0062] S403, Appearance Feature Analysis;

[0063] S404, Confidence analysis;

[0064] In step S401, the displacement and velocity changes of the target in consecutive frames are calculated to determine whether the trajectory is continuous and smooth. For trajectory T, the displacement difference between adjacent frames is calculated using the following formula:

[0065]

[0066] Δd i If the trajectory is within a reasonable range and the change is stable, it is considered continuous.

[0067] In step S402, the target's speed and acceleration are analyzed to determine whether they conform to human motion characteristics.

[0068] The formula for calculating speed is:

[0069] The formula for calculating acceleration is:

[0070] If the speed and acceleration are within the normal range of human movement, then the movement characteristics are considered normal.

[0071] According to the above technical solution, in step S403, the shape, size and color appearance features of the target are analyzed to determine whether they conform to human characteristics. The aspect ratio of a human body is generally between 0.3 and 0.5, and the color distribution conforms to the range of human skin color.

[0072] In step S404, the average confidence level of the trajectory containing the target is calculated using the following formula:

[0073]

[0074] c avg If the value is higher than the set threshold, the target confidence level is considered high.

[0075] In S4, the following judgment results are obtained:

[0076] If the target is a valid real person, then mark it as valid=1;

[0077] If the target is a ghost, then mark it as valid=0.

[0078] According to the above technical solution, step S5 specifically includes the following steps:

[0079] S501, Hazardous Area Assessment;

[0080] S502, Alarm triggered;

[0081] In step S501, the coordinate range (x) of the danger zone is preset. min ,y min ,x max ,y max For each valid target, determine whether its location is within the danger zone using the following formula:

[0082]

[0083] In step S502, when in_danger = 1, an alarm mechanism is triggered, including audible and visual alarms, SMS notifications, and email reminders.

[0084] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0085] 1. By using a high-precision personnel detection algorithm based on spatiotemporal relationships, the spatiotemporal features of the target in continuous frames are analyzed, which improves the adaptability to complex scenes. The spatiotemporal relationship analysis enables the detection algorithm to better understand the changes of the target in different times and spaces, thereby more accurately detecting real people. Even in complex backgrounds, it can maintain high detection accuracy. The high-precision personnel detection algorithm based on spatiotemporal relationships can better adapt to complex scenes and improve the accuracy of personnel detection.

[0086] By introducing a target validity judgment mechanism and combining it with trajectory analysis methods, the system accurately distinguishes between real people and ghosts, filtering out ghost targets. The target validity judgment mechanism utilizes features such as trajectory validity to screen tracked targets, effectively removing ghosts caused by light reflection and reducing false alarms. Furthermore, the mechanism effectively filters out ghosts and reduces false alarms. It also introduces deeper spatiotemporal analysis and target validity judgment in the process of personnel detection and tracking, thus maintaining stable detection performance even in complex environments.

[0087] 2. By combining the YOLOv8 personnel detection algorithm, Kalman filtering and Hungarian algorithm target tracking technology, as well as the target validity judgment mechanism, real-time personnel intrusion detection of automated workstations in the workshop is realized, constructing a real-time and continuous monitoring process to ensure real-time monitoring and personnel intrusion detection of automated workstations in the workshop, effectively improving detection accuracy, reducing false alarm rate, and ensuring the safety of workshop production;

[0088] Furthermore, this detection process has high detection accuracy and low false alarm rate in complex scenarios, which can effectively ensure the safety of workshop production. In addition, the algorithm parameters and thresholds can be adjusted according to the actual needs of the site to adapt to different detection frame rate requirements. Attached Figure Description

[0089] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof.

[0090] In the attached diagram:

[0091] Figure 1 This is an overall flowchart of the detection method of the present invention;

[0092] Figure 2 This is a flowchart of the target tracking process of the present invention;

[0093] Figure 3 This is a flowchart for determining the effectiveness of the objective of this invention. Detailed Implementation

[0094] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0095] Example: Figure 1-3 As shown, this invention provides a technical solution: a method for detecting personnel intrusion in dangerous areas. Through in-depth analysis of spatiotemporal relationships using detection and tracking algorithms, and by making in-depth judgments on the effectiveness of targets, the method improves the accuracy and reliability of personnel intrusion detection and reduces the false alarm rate. The method includes the following steps:

[0096] S1. Acquire image data and preprocess the video;

[0097] S2, Location and confidence level of the testing personnel;

[0098] S3. Continuously track personnel targets;

[0099] S4. Comprehensively analyze the effectiveness of the target trajectory;

[0100] S5. Determine if personnel intrusion is detected and trigger the alarm mechanism;

[0101] S1 acquires high-quality video streams to provide clear image data for subsequent personnel detection and tracking, and preprocesses the acquired video streams.

[0102] S2, using the YOLOv8 algorithm to detect the location and confidence level of people in an image;

[0103] S3 continuously tracks detected personnel targets and maintains their trajectories;

[0104] S4, analyze the effectiveness of the target trajectory and distinguish between real people and ghosts;

[0105] S5 triggers an alarm mechanism when it determines that a real person has entered a dangerous area, and notifies relevant personnel to take measures through an alarm signal.

[0106] Based on the above technical solution, S1 specifically includes the following steps:

[0107] S101, Data Acquisition;

[0108] S102, Image preprocessing;

[0109] S101 uses a high-definition camera installed in a suitable location at the automated workstation in the workshop to ensure coverage of the hazardous areas that need to be monitored. The resolution of the high-definition camera is no less than 1920×1080 pixels, and the camera parameters, including exposure time, gain and focal length, also need to be adjusted.

[0110] The exposure time is adjusted according to the actual lighting conditions, 60 microseconds, to ensure moderate image brightness. The gain is adjusted according to the image brightness, 10dB, to avoid the image being too bright or too dark. The focal length is adjusted according to the size and distance of the monitoring area to ensure image clarity.

[0111] During the data acquisition process, it is also necessary to record the lighting conditions and camera position when the high-definition camera is shooting, so as to facilitate subsequent processing. The current frame image is extracted from the video stream, and the acquired video stream is converted into an image sequence for preprocessing.

[0112] Based on the above technical solution, in step S102, image preprocessing includes grayscale conversion and noise reduction.

[0113] Grayscale conversion is the process of converting a color image to a grayscale image, reducing the amount of data and computational complexity. The formula is:

[0114] I gray = 0.299 × R + 0.587 × G + 0.114 × B;

[0115] Where R, G, and B are the pixel values ​​of the red, green, and blue channels of the color image, respectively;

[0116] Denoising is achieved by using Gaussian filtering to remove noise from an image. The Gaussian filtering formula is:

[0117]

[0118] Where G(i,j) is the Gaussian kernel function and k is the filtering radius;

[0119] The preprocessed grayscale image sequence is denoted as I. pre .

[0120] Based on the above technical solution, S2 specifically includes the following steps:

[0121] S201. Input the preprocessed image;

[0122] S202, Network Reasoning;

[0123] S203, Spatiotemporal Relationship Analysis;

[0124] S201, The input preprocessed image is the preprocessed grayscale image sequence I. pre Input into the trained YOLOv8 network;

[0125] S202, the YOLOv8 network performs feature extraction and object detection on images through convolutional layers, pooling layers and fully connected layers;

[0126] Output the detection box and its corresponding confidence score. The detection box is represented as a rectangle: (x, y, w, h).

[0127] Where (x,y) is the coordinate of the top left corner, w is the width, h is the height, and the confidence level represents the probability of detecting a person.

[0128] Based on the above technical solution, in step S203, the motion trend and spatial position changes of the target are analyzed by combining the spatiotemporal relationship between consecutive frames to improve detection accuracy. For the detection results of frame t and frame t+1, the displacement and velocity changes between the targets are calculated using the following formula:

[0129] Δx=x t+1 -x t ,Δy=yt+1 -y t

[0130] Where Δt is the time interval between two frames, denoted as D after the detected personnel location information and confidence level. t ={(x i ,y i ,w i ,h i ,c i )}, where c i , where is the confidence level.

[0131] Based on the above technical solution, S3 specifically includes the following steps:

[0132] S301. Initialize the Kalman filter;

[0133] S302, Target Prediction and Matching;

[0134] S303, Track Maintenance and Update;

[0135] S301, for each detection box (x,y,w,h) in the first frame, initialize the state vector of the Kalman filter x = [x,y,w,h,v]. x ,v y ] T ;

[0136] Among them, v x and v y The initial velocity is set to 0.

[0137] The state transition matrix F and the measurement matrix H are respectively:

[0138]

[0139] S302, In subsequent frames, the Kalman filter is used to predict the target's position, as follows:

[0140] x pred =F×x prev ;

[0141] The distance cost matrix between predicted and detected bounding boxes is calculated using the Hungarian algorithm for matching, minimizing the total cost, and is calculated by the following formula:

[0142]

[0143] Where, x pred,i y pred,i Let x be the coordinates of the predicted bounding box. det,j y det,j These are the coordinates of the detection box.

[0144] Based on the above technical solution, in step S303, for a successfully matched target, the state of the Kalman filter is updated using the following formula:

[0145] x new =x pred +K×(zH×x pred );

[0146] Where K is the Kalman gain and z is the measured value;

[0147] For unmatched targets, they are placed in the waiting pool and survive for 3 frames. If a target in the waiting pool can be matched with another target in the waiting pool within 3 frames, it is considered meaningful; otherwise, it is deleted. When a trajectory fails to match a newly detected target for 3 consecutive frames, the trajectory is considered to have ended its life.

[0148] The maintained target trajectory information, including position, velocity, and direction, is denoted as T. t ={x1,x2,...,x n}

[0149] Based on the above technical solution, S4 specifically includes the following steps:

[0150] S401, Trajectory Continuity Analysis;

[0151] S402, Motion Characteristic Analysis;

[0152] S403, Appearance Feature Analysis;

[0153] S404, Confidence analysis;

[0154] S401, calculate the displacement and velocity changes of the target in consecutive frames to determine the continuity and smoothness of the trajectory. For trajectory T, calculate the displacement difference between adjacent frames using the following formula:

[0155]

[0156] Δd i If the trajectory is within a reasonable range and the change is stable, it is considered continuous.

[0157] S402, Analyze the target's velocity and acceleration to determine whether it conforms to human motion characteristics;

[0158] The formula for calculating speed is:

[0159] The formula for calculating acceleration is:

[0160] If the speed and acceleration are within the normal range of human movement, then the movement characteristics are considered normal.

[0161] Based on the above technical solution, S403, analyze the appearance characteristics of the target's shape, size, and color to determine whether it conforms to human characteristics. The aspect ratio of a human body is generally 0.4, and the color distribution conforms to the range of human skin color.

[0162] S404, calculate the average confidence level of the trajectory containing the target, using the following formula:

[0163]

[0164] c avg If the value exceeds a set threshold of 0.8, the target confidence level is considered high.

[0165] In S4, the following judgment results are obtained:

[0166] If the target is a valid real person, then mark it as valid=1;

[0167] If the target is a ghost, then mark it as valid=0.

[0168] Based on the above technical solution, S5 specifically includes the following steps:

[0169] S501, Hazardous Area Assessment;

[0170] S502, Alarm triggered;

[0171] S501, Pre-set the coordinate range of the hazardous area (x min ,y min ,x max ,y max For each valid target, determine whether its location is within the danger zone using the following formula:

[0172]

[0173] S502, when in_danger=1, trigger the alarm mechanism, including audible and visual alarms, SMS notifications, and email reminders.

[0174] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for detecting intrusion of personnel into a hazardous area, characterized in that: By employing in-depth analysis of spatiotemporal relationships through detection and tracking algorithms, and by making thorough judgments on the effectiveness of targets, the accuracy and reliability of personnel intrusion detection can be improved, and the false alarm rate reduced. This includes the following steps: S1. Acquire image data and preprocess the video; S2, Location and confidence level of the testing personnel; S3. Continuously track personnel targets; S4. Comprehensively analyze the effectiveness of the target trajectory; S5. Determine if personnel intrusion is detected and trigger the alarm mechanism; S1 involves acquiring a high-quality video stream to provide clear image data for subsequent personnel detection and tracking, and preprocessing the acquired video stream. S2 uses the YOLOv8 algorithm to detect the location and confidence level of people in the image; S3 involves continuously tracking the detected personnel targets and maintaining their trajectories. S4 analyzes the effectiveness of the target trajectory and distinguishes between real people and ghosts; S5, when it is determined that a real person has entered the dangerous area, triggers an alarm mechanism and notifies relevant personnel to take measures through an alarm signal; S4 specifically includes the following steps: S401, Trajectory Continuity Analysis; S402, Motion Characteristic Analysis; S403, Appearance Feature Analysis; S404, Confidence analysis; In step S401, the displacement and velocity changes of the target in consecutive frames are calculated, and it is determined whether the trajectory is continuous and smooth. For the trajectory... The displacement difference between adjacent frames is calculated using the following formula: ； If the trajectory is within a reasonable range and the change is stable, it is considered continuous. In step S402, the target's speed and acceleration are analyzed to determine whether they conform to human motion characteristics. The formula for calculating speed is: ; The formula for calculating acceleration is: ; If the speed and acceleration are within the normal range of human motion, then the motion characteristics are considered normal. In step S403, the shape, size, and color appearance features of the target are analyzed to determine whether they conform to human characteristics. The aspect ratio of the human body is between 0.3 and 0.5, and the color distribution conforms to the range of human skin color. In step S404, the average confidence level of the trajectory containing the target is calculated using the following formula: ； If the value is higher than the set threshold, the target confidence level is considered high. In S4, the following judgment results are obtained: If the target is a valid real person, then mark it as... ; If the target is a ghost, then mark it as... .

2. The method for detecting personnel intrusion into dangerous areas according to claim 1, characterized in that: S1 specifically includes the following steps: S101, Data Acquisition; S102, Image preprocessing; S101 involves installing a high-definition camera at a suitable location in the automated workstation of the workshop to ensure coverage of the hazardous area that needs to be monitored. The resolution of the high-definition camera is not less than 1920×1080 pixels, and the camera parameters, including exposure time, gain and focal length, also need to be adjusted. The exposure time is adjusted according to the actual lighting conditions, between 10 and 100 microseconds, to ensure moderate image brightness. The gain is adjusted according to the image brightness, between 0 and 20 dB, to avoid the image being too bright or too dark. The focal length is adjusted according to the size and distance of the monitoring area to ensure image clarity. During the data acquisition process, it is also necessary to record the lighting conditions and camera position when the high-definition camera is shooting, so as to facilitate subsequent processing. The current frame image is extracted from the video stream, and the acquired video stream is converted into an image sequence for preprocessing.

3. The method for detecting personnel intrusion into dangerous areas according to claim 2, characterized in that: S102, the image preprocessing includes grayscale processing and noise reduction processing; Grayscale conversion is the process of converting a color image to a grayscale image, reducing the amount of data and computational complexity. The formula is: ； in, These are the pixel values ​​of the red, green, and blue channels of a color image, respectively. Denoising is achieved by using Gaussian filtering to remove noise from an image. The Gaussian filtering formula is: ； in, For Gaussian kernel function, The filter radius; The preprocessed grayscale image sequence is denoted as .

4. The method for detecting personnel intrusion into dangerous areas according to claim 3, characterized in that: S2 specifically includes the following steps: S201. Input the preprocessed image; S202, Network Reasoning; S203, Spatiotemporal Relationship Analysis; In step S201, the input preprocessed image is the preprocessed grayscale image sequence. Input into the trained YOLOv8 network; In S202, the YOLOv8 network performs feature extraction and target detection on images through convolutional layers, pooling layers, and fully connected layers; Output the detection box and its corresponding confidence score. The detection box is represented as a rectangle: ; in, The coordinates of the top left corner For width, The confidence level represents the probability of detecting a person.

5. The method for detecting personnel intrusion into dangerous areas according to claim 4, characterized in that: S203, by combining the spatiotemporal relationship between consecutive frames, analyzes the target's motion trend and spatial position changes to improve detection accuracy. For the first... Frame and the Based on the frame detection results, the displacement and velocity changes between targets are calculated using the following formula: ； in, The time interval between two frames, after detecting the personnel location information and confidence level, is denoted as... ,in, , where is the confidence level.

6. The method for detecting personnel intrusion into a dangerous area according to claim 5, characterized in that: S3 specifically includes the following steps: S301. Initialize the Kalman filter; S302, Target Prediction and Matching; S303, Track Maintenance and Update; In step S301, for each detection box in the first frame... Initialize the state vector of the Kalman filter. ; in, and The initial velocity is set to 0. State transition matrix and measurement matrix They are respectively: ； ； In step S302, the Kalman filter is used to predict the target's position in subsequent frames, as follows: ； The distance cost matrix between predicted and detected bounding boxes is calculated using the Hungarian algorithm for matching, minimizing the total cost, and is calculated by the following formula: ； in, The coordinates of the prediction box, These are the coordinates of the detection box.

7. The method for detecting personnel intrusion into a dangerous area according to claim 6, characterized in that: In step S303, for a successfully matched target, the state of the Kalman filter is updated using the following formula: ； in, For Kalman gain, These are measured values; For unmatched targets, they are placed in the waiting pool and survive for 3 frames. If a target in the waiting pool can be matched with another target in the waiting pool within 3 frames, it is considered meaningful; otherwise, it is deleted. When a trajectory fails to match a newly detected target for 3 consecutive frames, the trajectory is considered to have ended its life. The maintained target trajectory information, including position, velocity, and direction, is denoted as... .

8. The method for detecting personnel intrusion into a dangerous area according to claim 1, characterized in that: S5 specifically includes the following steps: S501, Hazardous Area Assessment; S502, Alarm triggered; S501, the coordinate range of the danger zone is preset. For each valid target, determine whether its location is within the danger zone using the following formula: ； In S502, when When the alarm mechanism is triggered, it includes audible and visual alarms, SMS notifications, and email alerts.

Images