System and method for detecting and monitoring personnel within hazardous areas of industrial sites

Abstract

To provide a personnel detection monitoring system and method for quickly detecting the presence of a person in a dangerous area of ​​an industrial site and controlling the drive of a work machine at the industrial site to ensure the safety of the person. [Solution] The system comprises a camera 100 mounted on a mobile device or fixed equipment that uses deep learning technology to detect whether a photographed and detected object is a person; a detected object coordinate extraction unit 200 that extracts the box pixel coordinate matrix of the detected object if it is a person; an object information providing unit 300 that assigns the person's 3D coordinate values, converts them into a 3D coordinate matrix, and saves the object information; a rotation and translation matrix calculation unit 400 that calculates rotation and translation matrices using the camera's internal parameters, the box pixel coordinate matrix, and the person's 3D coordinate matrix provided by the object information providing unit; and a real-time distance measurement unit 500 that measures the distance between the camera center and the detected object in real time using the calculated matrix values.

Description

[Technical Field]

[0001] The present invention relates to a system for detecting people in danger zones at industrial sites, and more specifically to a system for detecting and monitoring people in danger zones at industrial sites, which involves attaching a camera to a mobile or fixed facility at the industrial site, monitoring the surrounding area, and, if it is determined that a person is in the danger zone, maintaining the safety of that person. [Background technology]

[0002] Korean Patent Publication No. 10-2629019 (registered on January 19, 2024), which is a prior art related to personnel detection in dangerous areas, discloses technology related to a dangerous area personnel detection and machine control system that detects personnel entering each of four areas and generates and transmits different notification messages. The system includes multiple video cameras that capture images of the inside of a workplace in real time and transmit the captured images to the outside, a personnel detection device that receives multiple images captured by the video cameras, detects personnel in the dangerous area based on the received images, and transmits the detection results to the outside, a machine control device that receives the personnel detection results in the workplace from the personnel detection device and controls the operation of machines in the workplace, and an internal management device.

[0003] In addition, Korean Patent Publication No. 10-2024-0039618 (registered on March 27, 2024) discloses technology related to an intelligent integrated monitoring system, including a detection device that detects objects such as people, cars, and drones within a security area, and detects the occurrence of events such as fires, and transmits detection location information; a video capture device that automatically adjusts the shooting position and direction based on the detection location information transmitted from the detection device, captures video, and transmits the captured video; a video analysis device that analyzes the video transmitted from the video capture device using an artificial intelligence algorithm and transmits the analysis results; and an integrated monitoring server that performs monitoring operations based on the video transmitted from the video capture device and generates and outputs an alarm based on the video analysis results transmitted from the video analysis device.

[0004] In addition, Korean Patent Publication No. 10-2019-0035186 (published on April 3, 2019) discloses technology related to an intelligent unmanned security system that utilizes deep learning techniques, including a physical sensor located in a security area that detects images, sounds, movements, heat, gas, vibrations, and impacts of objects; at least one image acquisition unit that is linked to the physical sensor and acquires images of events in the security area; a video signal transmission unit that transmits the video signal acquired from the image acquisition unit through a communication network in response to detection by the physical sensor; and a deep learning analysis unit that receives the video information acquired from the video signal transmission unit and determines whether a security issue has occurred using deep learning techniques.

[0005] In addition, Korean Patent Publication No. 10-1378071 (registered on March 19, 2014) discloses technology relating to a system and method for monitoring aquaculture farms and artificial reefs to prevent theft of fish and shellfish from the farms and artificial reefs by using a combination of radar, video and thermal imaging to detect, identify and track intruders who invade the farms and artificial reefs above and below the water. [Prior art documents] [Patent documents]

[0006] [Patent Document 1] Korean Patent Registration No. 10-2629019 [Patent Document 2] Korean Patent Publication No. 10-2024-0039618 [Patent Document 3] Korean Patent Publication No. 10-2019-0035186 [Patent Document 4] Korean Patent Registration No. 10-1378071 Summary of the Invention [Problem to be solved by the invention]

[0007] To provide a system for quickly detecting a person in a dangerous area of ​​an industrial site and controlling the operation of a work machine at the industrial site when the person approaches within a certain distance from the work machine.

[0008] Another object of the present invention is to attach a camera to a mobile means such as a work machine, monitor the surroundings, quickly detect the presence of a person near the mobile means, and control the operation of the mobile means at the industrial site when the person approaches within a certain distance of the work machine.

[0009] Another object of the present invention is to control the operation of a work machine on the assumption that an object moving into a danger zone in an industrial site is a human when the object is identified.

[0010] Another object of the present invention is to manage the range of danger zones in industrial sites by accurate distance measurement.

[0011] Another object of the present invention is to not only quickly send a notification message to a worker when there is a person in a dangerous area of ​​an industrial site, but also to control the drive of the work machine. [Means for solving the problem]

[0012] The present invention aims to solve the above-mentioned problems, and provides: [1] a camera (100) mounted on a mobile means or fixed equipment, which detects whether a detected object photographed and detected by deep learning technology is a person; a detected object coordinate extraction unit (200) which extracts a coordinate matrix of a box pixel of the detected object (person) if the detected object photographed and detected is a person; an object information providing unit (300) which assigns 3D coordinate values ​​of the person, converts it into a 3D coordinate matrix, and stores object information of the 3D coordinate matrix; and an object information providing unit (300) which stores intrinsic parameters of the camera (100) and the box pixel provided by the detected object coordinate extraction unit (200). The present invention relates to a personnel detection and monitoring system for detecting and monitoring personnel in dangerous areas of industrial sites, characterized by comprising: a rotation and translation matrix calculation unit (400) that calculates rotation and translation matrices based on the pixel coordinate matrix and the person's 3D coordinate matrix provided by the object information providing unit; and a real-time distance measurement unit (500) that measures the distance between the camera center and the detected object (person) in real time using the matrix values ​​calculated by the rotation and translation matrix calculation unit (400).

[0013] The present invention also relates to a system for detecting and monitoring personnel in dangerous areas in industrial sites, characterized in that in the above [1], the rotation and translation matrix calculation unit (400) calculates the rotation matrix and the translation matrix using the solvePnP technique.

[0014] The present invention also relates to a personnel detection and monitoring system in a dangerous area of ​​an industrial site, characterized in that in the above [2], the real-time distance measurement unit (500) measures the distance between the object, i.e., a person, and the center of the camera (100) in real time using the L2-norm method.

[0015] The present invention also relates to [4] a personnel detection and monitoring system in a dangerous area of ​​an industrial site, characterized in that, in any of [1] to [3] above, when the distance measured by the real-time distance measurement unit (500) becomes equal to or less than a predetermined value, a notification means is activated in a danger signal notification unit (not shown).

[0016] The present invention also relates to [5] a personnel detection and monitoring system for a dangerous area in an industrial site, characterized in that in [1] above, the mobile means is one of an automobile, bulldozer, excavator, loader, dump truck, grader, or crane, and the camera is mounted in at least one of the front, rear, left, and right.

[0017] The present invention also relates to [6] a personnel detection and monitoring system in a dangerous area of ​​an industrial site, characterized in that in [1] above, the fixed facility is an indoor or outdoor work area, and the camera is mounted within the fixed facility near the dangerous area.

[0018] [7] The present invention also provides a method for detecting and monitoring personnel in a dangerous area of ​​an industrial site using the system for detecting personnel in a dangerous area of ​​an industrial site according to any one of [1] to [3], comprising: a step (S-1) of detecting an object by photographing it with a camera (100) mounted on a mobile means or a fixed facility, and determining whether the detected object is a person; a step (S-2) of extracting a coordinate matrix of box pixels of the detected object (person) photographed by the camera (100) when the detected object detected in the step (S-1) is determined to be a person; a step (S-3) of assigning 3D coordinate values ​​of the person to an object information providing unit (300), converting the 3D coordinate values ​​into a 3D coordinate matrix, and storing the object information; and a step (S-4) of extracting the coordinate matrix of box pixels of the detected object (person) extracted in the step (S-2), the 3D coordinate matrix stored in the step (S-3), and the intrinsic parameters (intrinsic parameters) of the camera (100). The present invention relates to a method for detecting and monitoring personnel in a dangerous area of ​​an industrial site, comprising: a step (S-4) of calculating a rotation and translation matrix based on the provided parameter; and a step (S-5) of measuring the distance between the center of the camera (100) and the detected object (person) in real time using the matrix value calculated in the step (S-4). [Effects of the Invention]

[0019] The present invention, which has the above-described configuration, can quickly detect the presence of a person in a dangerous area of ​​an industrial site and control the operation of the industrial site's work equipment to ensure the safety of the person.

[0020] In addition, the present invention can attach a camera to a mobile device at an industrial site to monitor the surroundings, and if there is a person near the mobile device, it can quickly detect this and control the operation of the work equipment to ensure the safety of the person.

[0021] Furthermore, in determining an object moving into a danger zone in an industrial site, the present invention can control the operation of a work machine on the assumption that the object is a person.

[0022] Furthermore, the present invention can manage the danger zone range in an industrial site by accurate distance measurement, so that the safety management of workers can be efficiently carried out.

[0023] Furthermore, when a person is in a dangerous area at an industrial site, the present invention can not only quickly send a notification message to the worker, but also control the operation of the work machine. [Brief explanation of the drawings]

[0024] [Figure 1] 1 is a conceptual diagram of the present invention. [Figure 2] 1 is a conceptual diagram illustrating a coordinate matrix of a detected object (person). [Figure 3] FIG. 10 is a conceptual diagram illustrating a coordinate matrix of object information. [Figure 4] FIG. 1 is a conceptual diagram illustrating the concepts of intrinsic parameters and extrinsic parameters of a camera according to the present invention. [Figure 5] FIG. 1 is a conceptual diagram illustrating a camera coordinate system and a world coordinate system. [Figure 6] This is a conceptual diagram showing the world coordinate values ​​(X1, X2, X3) and the camera orientation (X0). [Figure 7] This is a conceptual diagram showing the angles (α, β, γ) between vectors. [Figure 8] This is a conceptual diagram of sides (s1, s2, s3) with angles (α, β, γ) between vectors. DETAILED DESCRIPTION OF THE INVENTION

[0025] The present invention provides a camera (100) mounted on a mobile device or a fixed facility, which detects whether a detected object photographed and detected by deep learning technology is a person; a detected object coordinate extraction unit (200) which extracts a coordinate matrix of a box pixel of the detected object (person) if the detected object photographed and detected is a person; an object information providing unit (300) which assigns 3D coordinate values ​​of the person, converts it into a 3D coordinate matrix, and stores object information of the 3D coordinate matrix; and an object information providing unit (300) which stores intrinsic parameters of the camera (100) and the box pixel provided by the detected object coordinate extraction unit (200). The system comprises a rotation and translation matrix calculation unit (400) that calculates rotation and translation matrices based on the pixel coordinate matrix and the person's 3D coordinate matrix provided by the object information providing unit, and a real-time distance measurement unit (500) that measures the distance between the camera center and the detected object (person) in real time using the matrix values ​​calculated by the rotation and translation matrix calculation unit (400). Each of the components will be described in detail below.

[0026] [Camera (100)] The camera (100) of the present invention is attached to fixed equipment or mobile means in an industrial site, and is used as a means of constantly capturing images within a dangerous area and capturing images of moving objects when they are discovered in the industrial site.

[0027] The camera 100 detects the danger zone in real time and continuously captures moving objects if any.

[0028] The camera 100 is preferably installed at various locations in the industrial site so that it can detect all places where people can approach.

[0029] As shown in FIG. 1, the camera 100 determines whether an object image captured and detected in an industrial site is a human or not using a deep learning algorithm.

[0030] As shown in FIG. 1, intrinsic parameter values ​​are calculated for the camera (100).

[0031] The camera intrinsic parameters are parameters that indicate the internal characteristics of the camera and explain the process by which points in the 3D world are projected onto a 2D image plane in an image captured by the camera. These intrinsic parameters can be estimated through camera calibration and consist of the following elements:

[0032] First, the focal length (fx, fy) represents the focal distance in the X-axis and Y-axis directions, respectively. The unit is Pixel, which is the actual focal distance (mm) divided by the pixel size (mm / Pixel).

[0033] Next, the Principal Point (Cx, Cy) refers to the point where the optical center of the lens meets the image plane. It is located near the center of the image, but its exact location can vary from camera to camera.

[0034] The camera matrix (K) of the intrinsic parameters expresses the parameters in one matrix, as follows:

[0035]

number

[0036] This matrix is ​​used to convert 3D world coordinates to 2D image coordinates. The intrinsic parameters are determined during the camera construction process, but in practice they are accurately estimated through the camera calibration process. Using the intrinsic parameters obtained through this process allows for more precise 3D reconstruction, image measurement, and other tasks.

[0037] The intrinsic parameters of the camera calculated as described above are provided for rotation and translation matrix calculation in the rotation and translation matrix calculation unit (400) described below.

[0038] In addition, in the present invention, the detected object (person) coordinate extraction unit (200), object information providing unit (300), rotation / translation matrix calculation unit (400), real-time distance measurement unit (500) and danger signal notification unit (not shown) described below can all be installed together on one side of the camera (100), or only the detected object (person) coordinate extraction unit (200) configuration can be installed together, and the remaining configurations can be configured and installed together in a separate control unit.

[0039] [Detected object (person) coordinate extraction unit (200)] In the present invention, when the object photographed and detected by the camera 100 is a person, the detected object (person) coordinate extracting unit 200 extracts a coordinate matrix of the box pixel of the detected object (person).

[0040] The detected object (person) coordinate extraction unit 300 searches for a bounding box using an object detection algorithm and extracts the coordinates of the bounding box.

[0041] The method for extracting the box pixel coordinate matrix in this invention is performed using common image processing and computer vision techniques, such as OpenCV and PIL (Pillow), which are commonly used tools and libraries, to extract the box pixel coordinate matrix of the bounding box.

[0042] The box pixel coordinate matrix is ​​expressed by the following equation. As can be seen in Figure 2, the top left of the image is 0.0, and the bounding box of the detected object (person) is represented by a blue solid line on the outside. At this time, the top left pixel coordinate matrix and bottom right pixel coordinate matrix of the bounding box are extracted. This is expressed as follows to match the position with the actual 3D object of the object information providing unit (300) described below.

[0043]

number

[0044] [Object information provision department (300)] The present invention includes an object information providing unit 300 that provides human object information in advance to determine whether an object photographed and detected in an industrial site is a human.

[0045] In this case, since the 3D coordinate information of the person as the photographed object cannot be specified as a single coordinate, the 3D coordinate information is arbitrarily set using various body information of the person, such as the height and shoulder width of an adult male or female.

[0046] When generating 3D coordinates based on the person shown in Figure 3, assuming the center of the person's body is (0,0,0), the Y value increases as the height approaches the head, the X value increases as the width approaches the right, and the Z value is 0, generating the body coordinates of an adult (shoulder width / 2, height / 2,0).

[0047] At this time, the coordinates of the human object information are expressed as follows:

[0048]

number

[0049] [Rotation / Translation Matrix Calculation Unit (400)] The present invention has a rotation and translation matrix calculation unit (400) that calculates rotation and translation matrices for real-time distance measurement using the coordinate matrix of the object (person) extracted from the detected object coordinate extraction unit (200), the 3D coordinate matrix of the object (person) provided from the object information providing unit (300), and the intrinsic parameter values ​​of the camera (100).

[0050] In this invention, the rotation and translation matrices are calculated using the solvePnP technique, which converts between image coordinates and world coordinates. The solvePnP technique is an algorithm for solving the problem of estimating the 3D coordinates of an object when the 2D position of a specific point of the object in the camera image is given.

[0051] Below, a method for calculating a rotation matrix and a translation matrix (movement vector) using the solvePnP method will be specifically explained using the P3P (Perspective-3-points) algorithm.

[0052] The P3P (Perspective-3-points) algorithm is a method for calculating the pose information of the camera currently viewing in a 3D world coordinate system when given four pairs of 2D feature points and 3D landmarks in the world coordinate system.

[0053] Figure 5 shows the camera coordinate system and the world coordinate system.

[0054] The mathematically abstracted structure shown in Figure 5 is similar to the tetrahedron diagram in Figure 6. The coordinate values ​​(X 1, X 2, X3) and the direction vector in the camera coordinate system ( k x'1, k x'2, k x'3) is given, calculate the camera pose (X0) in 3D space.

[0055] At this time, the stage of acquiring the camera pose using the P3P algorithm consists of the following two stages.

[0056] Phase 1: Using the information we have, we construct a linear equation to calculate the relative "ray length" from the camera to each 3D point.

[0057] Second stage: This is the stage where "point cloud matching" between 3D points on the world coordinate system is performed using the relative distances to each 3D point obtained in the previous stage.

[0058] Then, after "performing point cloud matching", a step of calculating the rotation and translation matrices will follow.

[0059] Third stage: This is the stage where the rotation matrix is ​​calculated based on the result of the positive group matching in the second stage.

[0060] Fourth stage: This is the stage where the translation matrix (movement vector) is obtained based on the rotation matrix from the third stage.

[0061] Now, we will explain how to calculate the "length of the projection ray" in the first step. To calculate the length of the projection ray (s1, s2, s3), three steps must be performed in order: (1) calculate the angle between the three vectors that form the sides of the tetrahedron, (2) construct an equation using the cosine law, and (3) find the solution of the equation that corresponds to the length of the projection ray (s1, s2, s3) that we are trying to obtain.

[0062] The calculation of the angles between the three vectors in (1) above will be explained with reference to FIG.

[0063] First, the three vectors that make up the edges of the tetrahedron ( k x'1, k x'2, k To calculate the angle between x'3), use the arccos function to calculate the angles α, β, and γ. At this time, the direction vector ( k x' i ) is the feature point (x' i) and the inverse matrix of the camera internal parameters matrix (K -1 ) to calculate the direction vector in the camera coordinate system.

[0064]

number

[0065] Next, the method of forming an equation using the cosine law (2) above will be explained with reference to FIG.

[0066] As mentioned above, after determining the angles α, β, and γ, equations for the distances a, b, and c between the points are constructed using the cosine law.

[0067] First, when two sides (s1, s2) with an angle γ between them are given, an equation can be constructed for the remaining side (c) using the cosine law, as shown in Figure 8.

[0068] At this time, the magnitude of the angle γ and the length of the remaining side (c) are already known information, and the lengths of the two sides (s1, s2) are unknown quantities to be found.

[0069] Similarly, equations can be constructed for the angle α and the angle β using the cosine law, and when all equations are listed, the following formula is obtained.

[0070]

number

[0071] In this case, the angles α, β, and γ and the lengths a, b, and c of the sides between the three points are known information, and the lengths of the projection rays (s1, s2, s3) are unknown quantities that must be found from the equation.

[0072] Next, a method for finding the solution of the equation rearranged for (s1, s2, s3) in (3) above will be explained.

[0073] The three simultaneous equations obtained above are rearranged for (s1, s2, s3), and the first of the three simultaneous equations is first rearranged.

[0074]

number

[0075] Substituting and rearranging using new variables (u=s2 / s1, v=s3 / s1), we get the following equation:

[0076]

number

[0077] If we summarize this formula again with an equation for (s1), we get the following:

[0078]

number

[0079] Applying this process to the remaining two equations, we obtain three formulas for (s1).

[0080]

number

[0081] By collecting and rearranging the above three equations, we obtain the following fourth-order equation for the unknown (v):

[0082]

number

[0083] By solving the quartic equation for the unknown (v) obtained in this way and finding the coefficients A4, A3, A2, A1, and A0, we can rearrange them as follows:

[0084] What is noteworthy here is that the variables that make up the coefficients A4, A3, A2, A1, and A0 are composed of the angles α, β, and γ, which are already known information, and the lengths a, b, and c of the sides between the three points.

[0085]

number

[0086] Now, by using the calculated coefficients to find the variable v, we can finally calculate the desired variable, "length of the projection ray (s1, s2, s3)".

[0087]

number

[0088] However, since the polynomial for v mentioned above is a quartic equation, there is no unique solution; there are four algebraically possible solutions for (s1, s2, s3).

[0089] To find the actual solution among the four possible solutions, a geometric verification step can be performed, or an additional 3D point can be used to perform P3P several times using three randomly selected points among the four 3D points to finally calculate one satisfactory solution.

[0090] Next, the two-step method of "performing point cloud matching" will be described below.

[0091] In the first step, we calculated the "length of the projection ray" (s1, s2, s3), and through this, we can calculate the 3D point in the camera coordinate system. k x'1, k x'2, k The position of x'3 can be calculated.

[0092]

number

[0093] 3D point in the camera coordinate system k x'1, k x'2, k Since we know x'3 and the positions of all the 3D points (X1, X2, X3) in the world coordinate system, we can match the two point clouds and predict the camera pose (rotation matrix, translation matrix (= movement vector)).

[0094] First, assume that there are two different sets of points (X,Y) for which we know the correspondence.

[0095]

number

[0096] At this time, the desired information is calculated as the rotation matrix (R) and translation matrix (movement vector, t) that minimizes the sum of the Euclidean distances between the two point groups.

[0097]

number

[0098] In the present invention, when the correspondence relationship is known as prior information, a point cloud matching method is used that can quickly obtain a solution with a single calculation, rather than a nonlinear optimization method that calculates while iteratively updating the solution.

[0099] First, calculate the midpoint x0, y0 of each point group using the following formula.

[0100]

number

[0101] Next, the residual vector (a) is calculated from the original point cloud position by removing the midpoint position value. i ,b i ) is calculated.

[0102]

number

[0103] Next, the calculated residual vector is used to calculate the cross covariance matrix (H).

[0104]

number

[0105] Next, the point cloud alignment is calculated, and the calculated cross-covariance matrix (H) is decomposed using singular value decomposition (SVD).

[0106] Singular value decomposition is an operation that decomposes any matrix into three distinct matrices consisting of singular values ​​and singular vectors.

[0107]

number

[0108] At this time, among U, D, and V obtained by decomposing the cross-covariance matrix, the rotation matrix (R) can be calculated as follows using matrices U and V composed of singular vectors.

[0109]

number

[0110] Now that we have calculated the rotation matrix (R), we can calculate the translation matrix (movement vector, t) as follows:

[0111]

number

[0112] [Real-time distance measurement unit (500)] The present invention has a real-time distance measurement unit (500) that measures the distance between the center of the camera (100) and the object in real time using the translation matrix (movement vector, t) calculated by the rotation translation matrix calculation unit (400).

[0113] In the present invention, real-time distance measurement can be performed using the L2-norm method. At this time, the camera translation matrix can be expressed as follows:

[0114]

number

[0115] Here, t1, t2, and t3 respectively mean the distance traveled by x, t2 the distance traveled by y, and t3 the distance traveled by z.

[0116] In this case, the distance (d) between the two points is calculated as follows:

[0117]

number

[0118] That is, the distance is calculated by squaring the distance traveled along each of the x, y, and z axes, adding them all together, and then taking the square root of the sum. For example, to calculate the distance in two-dimensional space, if a translation matrix (t1=3, t2=4) is calculated, first the distance traveled along each axis is squared.

[0119] 3 2 =9,4 2 =16 Then find the sum of each. 9+16=25

[0120] Next, find the square root of that sum.

[0121]

number

[0122] Therefore, the final actual distance for the translation matrix (t1=3, t2=4) is 5.

[0123] The present invention makes it possible to measure distances in real time using the above method.

[0124] In the above, we have explained that the real-time distance measurement unit (500) measures the distance between the camera center and the object (person), but the present invention can further install a danger signal notification unit (not shown) that notifies of danger when the distance measured by the real-time distance measurement unit (500) is deemed to be a certain distance, i.e., a distance that may pose a danger to the object (person).

[0125] For example, if the distance between the center of the camera and the object (person) is less than 10 meters, the danger signal notification unit can be designed to activate. Also, the object (person) is not necessarily limited to a person, but can be any other object with a uniform range of movement.

[0126] It is also desirable that the danger signal notification unit be able to display an alarm, light, display, or the like. [Explanation of symbols]

[0127] 100: Camera 200: Detection object coordinate extraction part 300: Object information provision department 400: Rotation and translation matrix calculation unit 500: Real-time distance measurement unit

Claims

1. A camera (100) mounted on a mobile device or fixed equipment, which detects whether a detected object photographed and detected by deep learning technology is a person; a detection object coordinate extraction unit 200 for extracting a coordinate matrix of a box pixel of the detected object (person) when the detected object is a person; an object information providing unit (300) that assigns 3D coordinate values ​​of a person, converts the 3D coordinate values ​​into a 3D coordinate matrix, and stores object information of the 3D coordinate matrix; a rotation and translation matrix calculation unit (400) for calculating a rotation and translation matrix based on the intrinsic parameters of the camera (100), a coordinate matrix of a box pixel provided from the detected object coordinate extraction unit (200), and a 3D coordinate matrix of a person provided from the object information providing unit (300); and a real-time distance measurement unit (500) that measures the distance between the camera center and the detected object (person) in real time using the matrix value calculated by the rotation and translation matrix calculation unit (400).

2. In claim 1, The rotation and translation matrix calculation unit (400) A personnel detection and monitoring system for detecting personnel in dangerous areas of industrial sites, characterized in that rotation matrices and translation matrices are calculated using the solvePnP technique.

3. In claim 2, The real-time distance measurement unit (500) The system for detecting and monitoring personnel in a dangerous area of ​​an industrial site is characterized in that the distance between the detected object (person) and the center of the camera (100) is measured in real time using an L2-norm method.

4. In any one of claims 1 to 3, A personnel detection and monitoring system for detecting and monitoring personnel in a dangerous area of ​​an industrial site, characterized in that when the distance measured by the real-time distance measurement unit (500) is equal to or less than a predetermined value, a notification means is activated in the danger signal notification unit.

5. In claim 1, the moving means is any one of an automobile, a bulldozer, a shovel, a loader, a dump truck, a grader, a crane, and a forklift; The system for detecting and monitoring personnel in a dangerous area of ​​an industrial site is characterized in that the camera (100) is mounted in at least one of the front, rear, left and right positions.

6. In claim 1, The fixed facility is an indoor workplace or an outdoor workplace, A personnel detection and monitoring system for a dangerous area in an industrial site, characterized in that the camera is mounted in a location close to the dangerous area within the fixed equipment.

7. A method for detecting and monitoring personnel in a dangerous area of ​​an industrial site using the system for detecting and monitoring personnel in a dangerous area of ​​an industrial site according to any one of claims 1 to 3, comprising: A step (S-1) of photographing and detecting an object while the camera (100) is attached to the mobile means or the fixed facility and determining whether the detected object is a person; If the detected object detected in the step (S-1) is determined to be a person, a step (S-2) of extracting a coordinate matrix of a box pixel of the detected object (person) photographed by the camera (100); A step (S-3) of assigning 3D coordinate values ​​of a person to the object information providing unit (300), converting the 3D coordinate values ​​into a 3D coordinate matrix, and storing the object information; Step (S-4) of calculating a rotation and translation matrix based on the coordinate matrix of the box pixel of the detected object (person) extracted in step (S-2), the 3D coordinate matrix stored in step (S-3), and the intrinsic parameters of the camera (100); and (S-5) measuring the distance between the center of the camera (100) and the detected object (person) in real time using the matrix value calculated in the step (S-4).

Images