System and method for detecting and monitoring personnel in dangerous area of industrial site

Abstract

The invention relates to a system for detecting and monitoring personnel in a dangerous area of an industrial site, which is used for detecting personnel in the dangerous area of the industrial site, and comprises a camera which is arranged on a mobile unit or fixed equipment and is used for detecting whether a detection object shot and detected by a deep learning technology is a human or not; a detection object coordinate extraction unit for extracting a matrix of frame pixel coordinates of a detection object when the detection object to be photographed and detected is a human being; an object information provision unit that allocates an average three-dimensional coordinate value of a human being, converts the average three-dimensional coordinate value into a three-dimensional coordinate matrix, and stores object information; a rotation and translation matrix calculation unit that calculates a rotation and translation matrix from the internal parameters of the camera, the coordinate matrix of the frame pixels provided by the detection object coordinate extraction unit, and the average three-dimensional coordinate matrix of the human being provided by the object information provision unit; and a real-time distance measurement unit for measuring the distance between the center of the camera and the detection object in real time by using the matrix value calculated by the rotation and translation matrix calculation unit.

Description

Technical Field

[0001] This invention relates to a personnel detection and monitoring system for hazardous areas in industrial sites. Specifically, it involves attaching cameras to mobile units or fixed equipment in the industrial site. When the surrounding environment is monitored and it is determined that personnel are present in the hazardous area, the system ensures the safety of the personnel. Background Technology

[0002] As existing technology related to personnel detection in hazardous areas, Korean Patent Publication No. 10-2629019 (granted on January 19, 2024) discloses the following technology related to personnel detection and machinery control systems in hazardous areas. The system includes: multiple image capturing devices that capture images of the interior of a workshop in real time and transmit the captured images to the outside; a personnel detection device that receives multiple images captured by the image capturing devices, detects personnel in hazardous areas based on the received images, and transmits the detection results to the outside; and a machinery control management device that receives the personnel detection results from the personnel detection device to control the operation of machinery in the workshop. The personnel detection device divides the interior of the workshop into four areas based on an artificial intelligence model, detects personnel entering the four areas respectively, and generates different reminder messages for transmission.

[0003] Furthermore, Korean Patent Publication No. 10-2024-0039618 (granted on March 27, 2024) discloses the following technology related to an intelligent integrated monitoring system, which includes: a detection device that performs detection of objects such as humans, vehicles, and drones, as well as events such as fires, within a safe area, and transmits detection location information upon completion of the detection; an image capturing device that automatically adjusts the capturing position and direction based on the detection location information received from the detection device and captures images, and transmits the captured images; an image analysis device that analyzes the images received from the image capturing device using artificial intelligence algorithms and transmits the analysis results; and an integrated monitoring server that performs monitoring work based on the images received from the image capturing device and generates and outputs alerts based on the image analysis results received from the image analysis device.

[0004] Furthermore, Korean Patent Publication No. 10-2019-0035186 (published on April 3, 2019) discloses a technology related to an intelligent unmanned safety system using deep learning methods. This system is characterized by comprising: a physical detection unit located in a safety area, used to detect images, sounds, movement, heat, gas, vibration, and impacts of objects; at least one image acquisition unit linked to the physical detection unit to acquire event images of the safety area; an image signal transmission unit that, based on the detection by the physical detection unit, transmits the image signal acquired from the image acquisition unit via a communication network; and a deep learning analysis unit that receives the image information acquired from the image signal transmission unit and uses deep learning methods to determine whether a safety issue has occurred.

[0005] Furthermore, Korean Patent Publication No. 10-1378071 (granted on March 19, 2014) discloses the following technology related to a monitoring system and method for preventing theft of fish and shellfish from aquaculture farms and artificial reefs, which combines the use of radar, images and thermal imaging to detect, identify and track intruders who enter aquaculture farms and artificial reefs above and below water to prevent theft of fish and shellfish from aquaculture farms and artificial reefs.

[0006] Existing technical documents

[0007] Patent documents

[0008] Patent Document 1: Korean Patent Publication No. 10-2629019 (Granted on January 19, 2024)

[0009] Patent Document 2: Korean Patent Publication No. 10-2024-0039618 (published on March 27, 2024)

[0010] Patent Document 3: Korean Patent Publication No. 10-2019-0035186 (published on April 3, 2019)

[0011] Patent Document 4: Korean Patent Publication No. 10-1378071 (Granted on March 19, 2014) Summary of the Invention

[0012] Technical issues

[0013] The purpose of this invention is to quickly detect the presence of humans in a hazardous area of ​​an industrial site, and to control the operation of the working equipment when the humans approach within a specified distance of the equipment.

[0014] Furthermore, the purpose of this invention is to install cameras on mobile units such as work equipment to monitor the surrounding environment, quickly detect the presence of humans around the mobile unit, and control the drive of the mobile unit in the industrial site when the humans approach within a specified distance of the work equipment.

[0015] Furthermore, the purpose of this invention is to control the drive of the operating equipment based on the premise that the object is a human being when determining the movement of an object in a hazardous area of ​​an industrial site.

[0016] Furthermore, the purpose of this invention is to manage the scope of hazardous areas in industrial sites by accurately measuring distances.

[0017] Furthermore, the purpose of this invention is to quickly send alert messages to workers when humans are present in hazardous areas of an industrial site, and to simultaneously control the operation of the work equipment.

[0018] Technical solution

[0019] The present invention is used to solve the aforementioned problem. The present invention relates to a personnel detection and monitoring system in hazardous areas of industrial sites, characterized in that, [1] it includes: a camera 100, installed on a mobile unit or fixed device, for detecting whether the detected object captured by deep learning technology is a human; a detection object (human) coordinate extraction unit 200, which extracts a matrix of box pixel coordinates of the detected object (human) when the detected object is a human; an object information providing unit 300, which allocates the average three-dimensional coordinate values ​​of the human and converts them into a three-dimensional coordinate matrix to store the object information; a rotation and translation matrix calculation unit 400, which calculates the rotation and translation matrix using the intrinsic parameters of the camera 100, the coordinate matrix of the box pixels provided by the detection object (human) coordinate extraction unit 200, and the average three-dimensional coordinate matrix of the human provided by the object information providing unit; and a real-time distance measurement unit 500, which uses the matrix value calculated by the rotation and translation matrix calculation unit 400 to measure the distance between the camera center and the detected object (human) in real time.

[0020] Furthermore, the present invention relates to a personnel detection and monitoring system in hazardous areas of industrial sites, characterized in that, [2] in the above [1], the rotation and translation matrix calculation unit 400 uses the solvePnP method to calculate the rotation matrix and translation matrix.

[0021] Furthermore, the present invention relates to a personnel detection and monitoring system in hazardous areas of industrial sites, characterized in that, [3] in the above [2], the real-time distance measuring unit 500 measures the distance between the detection object (human) and the center of the camera 100 in real time using the L2-norm method.

[0022] Furthermore, the present invention relates to a personnel detection and monitoring system in hazardous areas of industrial sites, characterized in that, [4] in any of the [1] to [3], when the distance measured by the real-time distance measuring unit 500 is below a preset value, an alert unit is activated in the danger signal alert unit (not shown).

[0023] Furthermore, the present invention relates to a personnel detection and monitoring system in hazardous areas of industrial sites, characterized in that, [5] in the above [1], the mobile unit is one of a car, bulldozer, excavator, loader, dump truck, grader, crane, or forklift, and in this case, the camera is installed in at least one of the following positions: front, back, left, right, or right.

[0024] Furthermore, the present invention relates to a personnel detection and monitoring system in hazardous areas of industrial sites, characterized in that, [6] in the above [1], the fixed equipment is an indoor or outdoor workshop, and in this case, the camera is installed in the fixed equipment at a position close to the hazardous area.

[0025] Furthermore, the present invention relates to a method for detecting and monitoring personnel in hazardous areas of industrial sites, [7] being a method for detecting and monitoring personnel in hazardous areas of industrial sites using any one of [1] to [3], characterized in that it includes: step S-1, in the state where the camera 100 is installed on a mobile unit or a fixed device, determining whether the detected object is a human by shooting the object; step S-2, when it is determined that the detected object detected in step S-1 is a human, extracting the coordinate matrix of the frame pixels of the detected object (human) captured by the camera 100; step S-3, assigning the three-dimensional coordinate values ​​of the human to the object information providing unit 300, and converting them into a three-dimensional coordinate matrix to store the object information; step S-4, receiving the coordinate matrix of the frame pixels of the detected object (human) extracted in step S-2, the three-dimensional object information coordinate matrix stored in step S-3, and the internal parameter values ​​of the camera to calculate the rotation and translation matrix; and step S-5, using the matrix values ​​calculated in step S-4 to determine the distance between the center of the camera 100 and the detected object (human) in real time.

[0026] The effects of the invention

[0027] The present invention, through the aforementioned structure, can quickly detect the presence of humans in hazardous areas of an industrial site and safeguard human safety by controlling the operation of industrial site equipment.

[0028] Furthermore, in this invention, cameras are installed on mobile units in industrial settings to monitor the surrounding environment. When humans are present around the mobile unit, the system can quickly detect the situation and ensure human safety by controlling the operation of the equipment.

[0029] Furthermore, in this invention, when determining an object moving within a hazardous area of ​​an industrial site, the drive of the operating equipment can be controlled based on the premise that the object is a human.

[0030] Furthermore, in this invention, the range of hazardous areas in industrial sites can be managed by accurately measuring distances, thus effectively managing the safety of workers.

[0031] Furthermore, in this invention, when humans are present in a hazardous area of ​​an industrial site, a warning message can be quickly sent to the workers, and the operation of the equipment can also be controlled simultaneously. Attached Figure Description

[0032] Figure 1 This is a schematic diagram of the present invention.

[0033] Figure 2 A schematic diagram illustrating the coordinates of the object being detected.

[0034] Figure 3 A schematic diagram illustrating the coordinates of an object.

[0035] Figure 4 A schematic diagram illustrating the concepts of internal parameters and external parameters of the camera of the present invention.

[0036] Figure 5 This is a schematic diagram of the camera coordinate system and the world coordinate system.

[0037] Figure 6 This is a schematic diagram showing the world coordinates (X1, X2, X3) and the camera's pose (X0).

[0038] Figure 7 This is a schematic diagram illustrating the included angle (α, β, γ) between vectors.

[0039] Figure 8 This is a schematic diagram of the sides (s1, s2, s3) with vector angles (α, β, γ).

[0040] Explanation of reference numerals in the attached figures

[0041] 100: Camera

[0042] 200: Detection Object Coordinate Extraction Unit

[0043] 300: Object Information Provider

[0044] 400: Rotation and Translation Matrix Calculation Section

[0045] 500: Real-time distance measurement unit Detailed Implementation

[0046] This invention includes: a camera 100, mounted on a mobile unit or fixed device, for detecting whether a detected object captured by deep learning technology is a human; a detected object (human) coordinate extraction unit 200, which extracts a matrix of frame pixel coordinates of the detected object (human) when the captured detected object is a human; an object information providing unit 300, which assigns average three-dimensional coordinate values ​​of the human and converts them into a three-dimensional coordinate matrix to store object information; a rotation and translation matrix calculation unit 400, which calculates a rotation and translation matrix using the internal parameters of the camera 100, the frame pixel coordinate matrix provided by the detected object (human) coordinate extraction unit 200, and the average three-dimensional coordinate matrix of the human provided by the object information providing unit; and a real-time distance measurement unit 500, which uses the matrix values ​​calculated by the rotation and translation matrix calculation unit 400 to measure the distance between the camera center and the detected object (human) in real time. Each structure will be described in detail below.

[0047] [Camera 100]

[0048] The camera 100 of the present invention is installed on a fixed device or mobile unit in an industrial site. The camera 100 is a unit that continuously captures images of a moving detection object in a hazardous area when such an object is detected in the industrial site.

[0049] The camera 100 detects dangerous areas in real time and will continuously film any moving objects within the detection area.

[0050] Preferably, the camera 100 is installed in multiple locations on the industrial site, and can detect all locations that humans can access.

[0051] like Figure 1 As shown, the camera 100 uses a deep learning algorithm to determine whether the image of the detected object captured in the industrial site is a human.

[0052] And, as Figure 1 As shown, the internal parameter values ​​of the camera 100 have been calculated.

[0053] The camera's intrinsic parameters, representing its internal characteristics, refer to the process of projecting points in the three-dimensional world of the image captured by the camera into a two-dimensional image plane. These intrinsic parameters can be estimated through camera calibration and consist of the following elements.

[0054] Focal length (fx, fy) represents the focal length along the x-axis and y-axis, respectively. In this case, the unit is pixels, which is the actual focal length (mm) divided by the pixel size (mm / pixel).

[0055] The principal point (Cx, Cy) is the point where the optical center of the lens contacts the image plane. Although located near the center of the image, its exact position may differ for each camera.

[0056] The camera matrix (K) of the internal parameters represents the multiple parameters through a single matrix, as follows.

[0057]

[0058] The matrix is ​​used to convert three-dimensional world coordinates into two-dimensional image coordinates. Although the intrinsic parameters are determined during camera setup, they are actually accurately estimated through the camera calibration process. Using the intrinsic parameters obtained through this process allows for more precise and accurate 3D reconstruction, image measurement, and other operations.

[0059] The camera's intrinsic parameters calculated above are used to calculate the rotation and translation matrices in the rotation and translation matrix calculation unit 400 described below.

[0060] Furthermore, in this invention, the object coordinate extraction unit 200, object information providing unit 300, rotation and translation matrix calculation unit 400, real-time distance measuring unit 500 and danger signal warning unit (not shown) described below can be provided together on one side of the camera 100. Alternatively, only the object coordinate extraction unit 200 structure can be provided together, and the remaining structures can be configured together and set in a separate control unit.

[0061] [Detection Target (Human) Coordinate Extraction Unit 200]

[0062] In this invention, when the object detected by the camera 100 is a human, the coordinate matrix of the frame pixels of the object (human) detected by the object coordinate extraction unit 200 will be extracted.

[0063] The object coordinate extraction unit 200 uses an object detection algorithm to find the bounding box and extract the coordinate values ​​of the bounding box.

[0064] In this invention, the method for extracting the coordinate matrix of the bounding box pixels is implemented using general image processing and computer vision techniques. For example, the bounding box pixel coordinate matrix is ​​extracted using commonly used tools and libraries such as OpenCV and PIL (Pillow).

[0065] The pixel coordinates of the bounding box are represented by the following formula. For example... Figure 2 As shown, the top left edge of the image is 0,0, and the bounding box of the detected object (human) is represented by a solid blue line extending outwards. In this case, the coordinate matrices of the top left and bottom right pixels of the bounding box are extracted. To ensure that their positions are consistent with the actual 3D object position of the object information providing unit 300 described below, they are represented as follows.

[0066]

[0067] [Object Information Provision Unit 300]

[0068] The present invention includes an object information providing unit 300, which pre-sets and provides object information related to humans in order to determine whether the object being photographed and detected in an industrial site is a human.

[0069] In this situation, the three-dimensional coordinate information of the human being being photographed cannot be specified as a single entity. Therefore, the three-dimensional coordinate information is arbitrarily set using body information from multiple humans. This body information is set using various parameters such as height and shoulder width of an adult male or female.

[0070] If we take Figure 3 Using the human body as a reference to generate three-dimensional coordinates, if the center of the human body is assumed to be (0, 0, 0), and the Y-value is positive when the height is towards the head, the X-value is positive when the width is to the right, and the Z-value is 0, then the adult body coordinates (shoulder width / 2, height / 2, 0) will be generated. In this case, the adult body coordinates (coodinate) are expressed as follows.

[0071]

[0072] [Rotation / Translation Matrix Calculation Unit 400]

[0073] The present invention includes a rotation and translation matrix calculation unit 400 that calculates a rotation and translation matrix for determining real-time distance using the coordinate matrix of the detected object (human) extracted from the detected object coordinate extraction unit 200, the three-dimensional coordinate matrix of the object (human) provided by the object information providing unit 300, and the intrinsic parameter values ​​of the camera 100.

[0074] In this invention, the solvePnP method, which converts images into image coordinates and world coordinates, can be used to calculate the rotation and translation matrix. The solvePnP method is an algorithm used to estimate the three-dimensional coordinates of an object when its specific (two-dimensional) position in a camera image is known.

[0075] The following section provides a detailed explanation of the P3P (Perspective-3-points) algorithm for calculating rotation and translation matrices (movement vectors) using the solvePnP method.

[0076] The P3P (Perspective-3-points) algorithm is a method for calculating the pose information of the camera observed from the current viewpoint in the three-dimensional world coordinate system when the two-dimensional feature points and the three-dimensional landmark pairs in the world coordinate system are known.

[0077] Figure 5 The camera coordinate system and the world coordinate system are shown.

[0078] If Figure 5 If the structure shown is mathematically abstracted and presented using symbols, then as follows: Figure 6 The tetrahedral diagram. Given the coordinates X1, X2, X3 in the world coordinate system relating to the 3D point and the direction vector in the camera coordinate system. At that time, calculate the camera pose X existing in three-dimensional space. c .

[0079] In this case, the steps to obtain the camera pose using the P3P algorithm consist of the following two steps.

[0080] Step 1: Use the known information to construct a linear equation and calculate the relative (projected ray length) from the camera to each 3D point.

[0081] Step 2: Use the relative distances to the acquired 3D points to perform point cloud registration between the 3D points in the world coordinate system.

[0082] Next, after (performing point cloud registration), the next step is to calculate the rotation and translation matrices.

[0083] Step 3: Calculate the rotation matrix based on the results of (performing point cloud registration) in Step 2.

[0084] Step 4: Obtain the translation matrix (movement vector) based on the rotation matrix in Step 3.

[0085] The method for calculating the "distance of the projected ray" in step 1 is explained. In order to calculate the length (s1, s2, s3) of the projected ray, the following steps are required: step (1), calculate the angle between the three vectors that form the edge of the tetrahedron; step (2), use the law of cosines to construct an equation; step (3), for the length (s1, s2, s3) of the projected ray to be obtained, find the solution of the prepared equation.

[0086] Reference Figure 7 This describes the method for calculating the angle between the three vectors in step (1).

[0087] First, in order to calculate the three direction vectors of the edges that form the tetrahedron The angles between them are calculated using the arccosine function, specifically the angles α, β, and γ. In this case, the direction vector... Feature points (x') placed on a two-dimensional image plane i Multiply by the inverse of the intrinsic parameter matrix (K) -1 This is used to calculate the direction vector in the camera coordinate system.

[0088]

[0089] Next, refer to Figure 8 Explain the method of using the cosine law described in (2) to construct the equation.

[0090] As mentioned above, after finding the included angles α, β, and γ, the law of cosines is used to construct equations related to the distances a, b, and c between each point.

[0091] First, given the two sides s1 and s2 with γ as the included angle, if we use the law of cosines to construct an equation for the remaining side c, then... Figure 8 It is formed in the manner shown.

[0092] In this case, the size of the included angle γ and the length of the remaining side c are known information, while the lengths of the two sides s1 and s2 are unknowns that need to be determined.

[0093] Similarly, we can use the law of cosines to construct equations for both included angles α and β. If all equations are listed, the mathematical expressions are as follows.

[0094]

[0095] In this case, the included angles α, β, γ and the side lengths a, b, c between the three points are known information, while the lengths of the projected rays (s1, s2, s3) are unknowns that need to be obtained from the equations.

[0096] Next, the method for finding the solution to the rearranged equation (s1, s2, s3) for (3) will be explained.

[0097] Regarding the three simultaneous equations obtained by arranging (s1, s2, s3), firstly, the first equation in the three simultaneous equations is arranged.

[0098]

[0099] If new variables are used Substituting and rearranging, the mathematical expression is as follows.

[0100]

[0101] The mathematical expression can be rearranged to express the expression for s1 as follows.

[0102]

[0103] Applying the above process to the remaining two equations yields three mathematical expressions for s1.

[0104]

[0105] By combining and rearranging the three mathematical expressions, we can obtain a fourth-order equation for the following unknown υ.

[0106] A4v 4 +A3v 3 +A2v 2 +A1v 1 +A0=0

[0107] If the coefficients are obtained by solving the fourth-order equation for the unknown υ, the result can be simplified as follows.

[0108] In this case, it should be noted that the multiple variables constituting the coefficients A4, A3, A2, A1, and A0 are composed of the included angles α, β, and γ, and the side lengths a, b, and c between the three points, which are known information.

[0109]

[0110] Then, if the calculated coefficients are used to determine the variable υ, the lengths s1, s2, and s3 of the projected rays of the final determined variable can be calculated.

[0111]

[0112] However, the polynomial for υ is a fourth-order equation, therefore, there is no unique solution, and s1, s2, and s3 have four algebraically possible solutions.

[0113] In order to obtain the actual solution from the four algebraically possible solutions, a geometric verification step is required. An additional three-dimensional point is used, and P3P is performed multiple times using three points randomly obtained from the four three-dimensional points. Finally, a solution that satisfies the condition can be calculated.

[0114] Next, the method for performing point cloud registration in the two steps will be explained.

[0115] In step 1, the lengths of the projected rays s1, s2, and s3 were calculated; therefore, three-dimensional points in the camera coordinate system can be calculated. The location.

[0116]

[0117] All know the three-dimensional points in the camera coordinate system. Given the positions of 3D points X1, X2, and X3 in the world coordinate system, registration between the two point clouds can be performed to predict the camera pose (rotation matrix, translation matrix (= translation vector)).

[0118] First, assume there exist two distinct point cloud sets X and Y with known correspondences.

[0119] X={x1,x2,x3},Y={y1,y2,y3}

[0120] In this case, the required information is the rotation matrix R and the translation matrix (translation vector, t) that minimizes the sum of the Euclidean distances between the two point clouds.

[0121]

[0122] In this invention, when the correspondence is known in advance, instead of using a nonlinear optimization method that repeatedly updates the solution while performing calculations, a point cloud registration method that can directly obtain the solution through a single calculation is used.

[0123] First, calculate the centers x0 and y0 of each point cloud using the following mathematical formula.

[0124]

[0125] Next, calculate the residual vector a from the original point cloud location after removing the center point position value. i b i .

[0126] a i =(y i -y0), b i =(x i -x0)

[0127] Next, the calculated residual vector is used to calculate the cross covariance matrix H.

[0128]

[0129] Next, the covariance matrix (H) obtained by computational point cloud registration is decomposed using singular value decomposition (SVD).

[0130] Singular value decomposition is an operation that decomposes any matrix into three distinct matrices composed of specific values ​​and singular vectors.

[0131] svd(H)=UDV T

[0132] In this case, the rotation matrix R is calculated using matrices U and V, which are composed of specific vectors, obtained by decomposing the cross-covariance matrix.

[0133] R = VU T

[0134] The translation matrix (translation vector, t) can be calculated by calculating the rotation matrix R as follows.

[0135] t = y0 - Rx0

[0136] [Real-time distance measuring unit 500]

[0137] The present invention includes a real-time distance measuring unit 500 that uses a translation matrix (movement vector, t) calculated by the rotation-translation matrix calculation unit 400 to measure the distance between the center of the camera 100 and the object in real time. In the present invention, the real-time distance measurement can be performed using the L2.norm method.

[0138] In this case, the translation matrix of the camera can be represented as follows.

[0139] t = y0 - Rx0 = (t1, t2, t3)

[0140] In t1, t2, and t3, t1 refers to the distance moved by x, t2 refers to the distance moved by y, and t3 refers to the distance moved by z.

[0141] In this case, the distance d between the two points is calculated as follows.

[0142]

[0143] That is, the distance moved relative to each x-axis, y-axis, and z-axis is squared, and these squares are added together. The square root of the sum is then used to calculate the distance. For example, in two-dimensional space, the translation matrix is ​​calculated as (t1 = 3, t2 = 4). To calculate this distance, the distance moved along each axis is first squared.

[0144] 3 2 =9,4 2 =16

[0145] Furthermore, calculate their respective sums.

[0146] 9 + 16 = 25

[0147] Next, find the square root of the sum.

[0148]

[0149] Therefore, the final actual distance of the translation matrix (t1 = 3, t2 = 4) is 5. This invention allows for real-time distance measurement using the described method.

[0150] As described above, the case where the distance between the camera center and the object (human) is measured by the real-time distance measuring unit 500 has been explained. In this invention, when the distance measured by the real-time distance measuring unit 500 is a predetermined distance, that is, a distance that may pose a danger to the object (human), a danger signal warning unit (not shown) can also be provided to indicate the danger. For example, it can be designed so that the danger signal warning unit can operate when the distance between the camera center and the object (human) is 10 meters or less. Furthermore, the object (human) is not limited to humans, but can also be other objects with a similar range of movement.

[0151] In this case, preferably, the danger signal alert can be presented through an alarm, light, display, or the like.

Claims

1. A personnel detection and monitoring system for hazardous areas in industrial sites, characterized in that, include: A camera (100), installed on a mobile unit or fixed device, detects whether the detected object captured by deep learning technology is a human; The detection object coordinate extraction unit (200) extracts a matrix of frame pixel coordinates of the detection object when the detected object is a human being. The detection object refers to a human being. The object information providing unit (300) assigns the average three-dimensional coordinate values ​​of humans and converts them into a three-dimensional coordinate matrix to store the object information; The rotation and translation matrix calculation unit (400) calculates the rotation and translation matrix using the internal parameters of the camera (100), the coordinate matrix of the frame pixels provided by the object detection coordinate extraction unit (200), and the average three-dimensional coordinate matrix of the human provided by the object information providing unit, wherein the object detection refers to the human. as well as The real-time distance measuring unit (500) uses the matrix value calculated by the rotation and translation matrix calculation unit (400) to measure the distance between the center of the camera and the target object in real time, wherein the target object refers to a human.

2. The personnel detection and monitoring system in hazardous areas of industrial sites according to claim 1, characterized in that, The rotation and translation matrix calculation unit (400) uses the solvePnP method to calculate the rotation and translation matrices.

3. The personnel detection and monitoring system in hazardous areas of industrial sites according to claim 2, characterized in that, The real-time distance measuring unit (500) measures the distance between the detected object and the center of the camera (100) in real time using the L2-norm method. The detected object refers to a human being.

4. The personnel detection and monitoring system in hazardous areas of industrial sites according to any one of claims 1 to 3, characterized in that, When the distance measured by the real-time distance measuring unit (500) is below a preset value, the warning unit is activated in the danger signal warning unit (not shown).

5. The personnel detection and monitoring system in hazardous areas of industrial sites according to claim 1, characterized in that, The mobile unit is one of the following: automobile, bulldozer, excavator, loader, dump truck, grader, crane, or forklift. In this case, the camera is installed in at least one of the following positions: front, back, left, or right.

6. The personnel detection and monitoring system in hazardous areas of industrial sites according to claim 1, characterized in that, The fixed equipment can be an indoor workshop or an outdoor workshop. In this case, the camera is installed in a fixed device at a location close to the danger zone.

7. A method for personnel detection and monitoring in a hazardous area of ​​an industrial site, comprising a personnel detection and monitoring system for hazardous areas of an industrial site according to any one of claims 1 to 3, characterized in that, include: Step (S-1): With the camera (100) installed on a mobile unit or a fixed device, determine whether the target is a human by taking a picture of the target. Step (S-2): When it is determined that the detected object in step (S-1) is a human, the coordinate matrix of the frame pixels of the detected object captured by the camera (100) is extracted, wherein the detected object refers to a human. Step (S-3): Assign the three-dimensional coordinate values ​​of the human to the object information providing unit (300) and convert them into a three-dimensional coordinate matrix to store the object information; Step (S-4): Receive the coordinate matrix of the bounding pixel of the target object extracted in step (S-2), the coordinate matrix of the three-dimensional object information stored in step (S-3), and the internal parameter values ​​of the camera to calculate the rotation and translation matrices. The target object refers to a human. as well as Step (S-5) uses the matrix value calculated in step (S-4) to determine in real time the distance between the center of the camera (100) and the target being detected, the target being referred to as a human.

Images