Worker positioning system using ai algorithm

Abstract

The present invention relates to a worker positioning system using an AI algorithm. Specifically, the present invention relates to a worker positioning system using an AI algorithm, which may specify a central area (hereinafter, referred to as a "floor point") between a left foot and a right foot at a position where a person stands so as to detect the person on an image or video acquired through a camera installed in a specific space and measure the position of the detected person, thereby reliably performing positioning of a position in an environment where information on the entire human body may not be acquired due to a camera angle of view or obstacles.

Description

Worker location tracking system using AI algorithms

[0001] The present invention relates to a system for determining the location of a worker using an AI algorithm.

[0002] Specifically, the present invention relates to a system for positioning a worker using an AI algorithm, which can reliably perform positioning in an environment where information about the entire body cannot be obtained due to camera angle or obstacles, by specifying the center area between the left and right feet (hereinafter referred to as the 'bottom point') of the position where the person is standing, in order to detect a person in an image or video acquired through a camera installed in a specific space and measure the position of the detected person.

[0003] Recently, due to the rapid development of industrial technology and information and communication technology, research and technology development on computer vision, which implements human visual recognition ability using computers, are actively being conducted.

[0004] Since the core of the above computer vision is to extract information by analyzing images or videos captured by a camera, detecting the position and direction of objects in the captured images is very important. In particular, human pose estimation technology can be viewed as a problem of literally estimating the position of how the joints of a person's body are configured in a photo or video. However, the appearance of a person in a photo is not that all joints are visible, and even in the same pose, it varies depending on the shooting direction, it may be obscured by other objects, the person may be wearing various clothes, and pose estimation may be difficult depending on the intensity or strength of light.

[0005] The traditional method of human pose estimation involves attaching various devices such as sensors to a person, which can precisely detect movements. However, it is expensive and not always possible to wear the devices in real life, so it is only possible in limited areas such as laboratories.

[0006] Accordingly, research on estimating poses from photographs without body-attached equipment has been actively conducted, and as a result, features such as the outline of the human body or outlines that can be used to infer specific body parts from photographs have been extracted, and the patterns of the extracted features can be directly analyzed and applied by people to predict body parts, and efficiency and accuracy are increasing day by day.

[0007] Additionally, deep learning has recently been applied to human pose estimation research, showing remarkable performance improvements.

[0008] The aforementioned deep learning utilizes multilayer artificial neural networks and learns useful features directly from data. This learning-based approach has the advantage of directly identifying useful features from data that humans may not otherwise perceive. As human pose estimation becomes more accurate, it is increasingly being used in areas such as posture correction, action recognition, and safety prevention.

[0009] However, since the conventional human pose estimation research method as described above identifies an object from an image captured by a single camera and estimates the pose of the identified object, it is not easy to continuously track the position of the object through a single camera when the object moves in a specific space.

[0010] Accordingly, the present invention proposes a method for identifying an object from image data captured by multiple cameras installed in a specific space indoors or outdoors, estimating a pattern (e.g., pose) of the movement or behavior of the identified object, and continuously tracking the movement path of the object in a specific space based on the pattern.

[0011] In particular, the present invention proposes a method for recognizing multiple joint parts of an object by using a learning model for skeleton estimation from image data captured by the camera when the object moves in a specific space where a camera is installed, generating a skeleton data set, applying the generated skeleton data set to a learning model for pattern estimation generated in advance, estimating a pattern for the movement or action of the object, and tracking the location of the object based on the estimated pattern.

[0012] Furthermore, we would like to propose a method to deal with cases where proper detection is difficult due to obstacles other than the object existing in a specific space where the camera is installed.

[0013] Patent Publication No. 10-15-17359 describes an object tracking method and an object tracking device using the same.

[0014] The above technology relates to an object tracking method and an object tracking device using the same, which can track an object more accurately with a small amount of computation by tracking the object according to a particle filter method that is easy to track the object when there is a lot of object movement when tracking the object fails due to a sudden movement of the object when tracking the object in parallel using an optical flow method and a CAMshift method that are easy to track the object when there is little object movement.

[0015] Additionally, Patent Publication No. 10-1243294 describes a method and device for extracting and tracking moving objects.

[0016] The above technology relates to a method and device for extracting and tracking a moving object by extracting an object using a difference image between adjacent images and continuously transmitting the location information of the extracted object, thereby tracking the moving object based on the accurate location information of at least one moving object.

[0017] That is, the above prior art describes a method and device capable of extracting at least one moving object from a complex real-life environment, tracking and analyzing the moving object by identifying the location of the moving object and acquiring location information.

[0018] However, if we cannot even extract at least one moving object in a complex environment with obstacles, there is no solution.

[0019]

[0020] The purpose of the present invention is to provide a system for positioning a worker using an AI algorithm, which can reliably perform positioning in an environment where information about the entire body cannot be obtained due to camera angle or obstacles, by specifying the center area between the left and right feet (hereinafter referred to as the 'bottom point') of the position where the person is standing, in order to detect a person in an image or video acquired through a camera installed in a specific space and measure the position of the detected person.

[0021] In order to achieve the above-described purpose, a system for positioning a worker using an AI algorithm according to the present invention comprises: at least one camera installed in a specific space; a platform for receiving an image from the camera and tracking a displacement of a position of a detected person based on a distance from a bottom point of the person detected through image analysis to the camera; and, when the detection of the person is not complete due to a plurality of obstacles provided in the specific space, the platform estimates a physical pose of the detected person through an AI learning model to specify the bottom point.

[0022] At this time, the platform includes: an image receiving unit that receives image information or image information from the camera; a floor plan information obtaining unit that secures floor plan information based on the information received from the image receiving unit and maps the camera information to the floor plan information based on the actual installation location of the camera; a person detection unit that detects a person on the information received from the image receiving unit; a skeleton analysis unit that analyzes the left and right feet of the person detected through the person detection unit; a floor point specifying unit that specifies the center area of ​​the left and right feet of the person detected by the skeleton analysis unit as a floor point; and a distance measuring unit that measures the distance from the floor point specified by the floor point specifying unit to the camera.

[0023] The above floor point specific unit is characterized by mapping the floor point to the plan information based on the distance from the floor point to the camera measured through the distance measuring unit.

[0024] In addition, the platform is characterized in that, when there are multiple cameras, the overlapping shooting ranges of each camera are set as intersection areas in the plan information.

[0025] In addition, the platform is characterized by further including a floor point location tracking unit that tracks the movement of a person who is the cause of the floor point by tracking the location of the floor point based on the floor point that is continuously mapped to the floor plan information by the floor point specific unit.

[0026] In addition, the platform is characterized in that, if information on the right and left feet of the person detected by the skeleton analysis unit is not analyzed, it determines that an obstacle exists, and the floor point specifying unit estimates and specifies the floor point by comparing it with pre-learned information having a pose similar to that of the person detected in the image in which the obstacle is determined to exist.

[0027] According to the system for positioning a worker using an AI algorithm according to the present invention, in order to detect a person in an image or video acquired through a camera installed in a specific space and measure the position of the detected person, the system has the advantage of being able to reliably perform positioning in an environment where information about the entire body cannot be acquired due to camera angle or obstacles by specifying the central area between the left and right feet (hereinafter referred to as the 'bottom point') of the position where the person is standing.

[0028] Figure 1 illustrates a worker positioning system using an AI algorithm according to the present invention.

[0029] Figure 2 is a flowchart showing the process of a worker positioning system using an AI algorithm according to the present invention positioning and tracking a person.

[0030] Figure 3 illustrates an example of a worker positioning system using an AI algorithm according to the present invention obtaining floor plan information and mapping camera information.

[0031] Figure 4 illustrates an example of a worker positioning system using an AI algorithm according to the present invention for detecting a person, identifying a floor point, and tracking a location.

[0032] FIG. 5 illustrates an example of a case where an obstacle exists when detecting a person, specifying a floor point, and tracking a location using an AI algorithm of a worker's location positioning system according to the present invention.

[0033] Terms or words used in this specification and claims should not be interpreted as limited to their usual or dictionary meanings, but should be interpreted as meanings and concepts that conform to the technical spirit of the present invention, based on the principle that the inventor can appropriately define the concept of the term to best explain his or her own invention.

[0034] Therefore, the embodiments described in this specification and the configurations illustrated in the drawings are only the most preferred embodiments of the present invention and do not represent all of the technical ideas of the present invention. Therefore, it should be understood that there may be various equivalents and modified examples that can replace them at the time of filing this application.

[0035] Hereinafter, before explaining with reference to the drawings, it is to be noted that matters that are not necessary to reveal the gist of the present invention, that is, known configurations that can be obviously added by a person skilled in the art with ordinary knowledge, are not illustrated or specifically described.

[0036] The invention relates to a system for determining the location of a worker using an AI algorithm.

[0037] Specifically, the present invention relates to a system for positioning a worker using an AI algorithm, which can reliably perform positioning in an environment where information about the entire body cannot be obtained due to camera angle or obstacles, by specifying the center area between the left and right feet (hereinafter referred to as the 'bottom point') of the position where the person is standing, in order to detect a person in an image or video acquired through a camera installed in a specific space and measure the position of the detected person.

[0038] Figure 1 illustrates a worker positioning system using an AI algorithm according to the present invention.

[0039] The system for positioning a worker using an AI algorithm of the present invention according to Fig. 1 of the attached drawing is configured to include an image receiving unit that receives an image from a camera on a platform configured as an application (App) or a physical / virtual server to which a terminal connects to output and share information.

[0040] The above cameras are installed in numbers of one or more within a specific space being filmed.

[0041] In addition, the above image receiving unit obtains floor plan information for a specific space based on the received camera video or image information.

[0042] For this purpose, the platform may be configured to further include a floor plan information acquisition unit.

[0043] In addition, the floor plan information acquisition unit can map information about the installation location of the camera on the secured floor plan information, and it is preferable that the mapping of the camera be identical to the location where it is actually installed in a specific space.

[0044] A description of the acquisition and mapping of such plan information is shown in Figure 3 of the attached drawing.

[0045] Figure 3 illustrates an example of a worker positioning system using an AI algorithm according to the present invention obtaining floor plan information and mapping camera information.

[0046] Based on Figure 3 of the attached drawing, the mapping of camera information is based on actual location information, and in case there are multiple cameras, an intersection area for the shooting range of each camera can be set.

[0047] Additionally, the mapping of camera information is based on actual GPS information, and it is desirable to apply GPS information to the floor plan information.

[0048] Additionally, the platform includes a person detection unit that detects a person in the video or image information received from the video receiver. At this time, the person detection can utilize conventional techniques such as pixel analysis.

[0049] In addition, when a person is detected in video or image information through the person detection unit, the platform applies a skeleton analysis technique to analyze the positions of the person's left and right feet.

[0050] Once the analysis is completed through the skeleton analysis unit, the platform identifies the floor point, which is the central area between the left and right feet of the person detected through the floor point feature unit.

[0051] In this way, when the bottom point of a person detected through the bottom point specific unit is specified, the distance from the bottom point to the camera is measured. This is performed through the distance measuring unit.

[0052] At this time, the location of the bottom point may be a camera capable of measuring distances, and the technology for measuring distances using such cameras is commonly known, so a detailed description will be omitted.

[0053] Alternatively, the distance can be measured using a triangulation technique using two cameras and the distance to the floor point of the detected person.

[0054] In addition, when the distance measurement of the floor point is performed through the distance measurement unit, the floor point specification unit maps the floor point to the plan information.

[0055] Afterwards, by continuously specifying the bottom point of a person detected through a camera and measuring the distance, the change in the location of the bottom point can be measured, and the direction of movement of the detected person can be tracked through this. This is possible through a bottom point location tracking unit.

[0056] As an example, identifying a floor point and tracking it by measuring its location is shown in Fig. 4 of the attached drawing.

[0057] Figure 4 illustrates an example of a worker positioning system using an AI algorithm according to the present invention for detecting a person, specifying a floor point, and tracking a location.

[0058] This series of processes is further explained in Figure 2 of the attached drawing.

[0059] Figure 2 is a flowchart showing the process of a worker positioning system using an AI algorithm according to the present invention positioning and tracking a person.

[0060] First, after receiving video or image information, a person is detected from the information. Then, as described above, the floor point is specified and the distance from the floor point to the camera is measured to track the location. However, for example, the presence of various obstacles within a specific space may prevent the person from being completely detected. In this case, it may be difficult to specify the floor point because the right and left feet of the detected person cannot be detected due to the limitations of skeleton analysis. However, the platform of the present invention is linked to a model capable of AI-based learning (deep learning), and can learn the entire process of performing skeletal analysis of the detected person from the received video or image information to specify the floor point.

[0061] Accordingly, the bottom point-specification unit determines that an obstacle exists if an error occurs in the skeleton analysis in the skeleton analysis unit, and specifies the bottom point based on information with a high matching rate from previous learning information in conjunction with the learning model.

[0062] This is shown in Figure 5 of the attached drawing.

[0063] FIG. 5 illustrates an example of a case where an obstacle exists when detecting a person, specifying a floor point, and tracking a location using an AI algorithm of a worker's location positioning system according to the present invention.

[0064] That is, in cases where only a portion of a person is detected due to the presence of an obstacle, as shown in Figure 5 of the attached drawing, and thus skeletal analysis is not satisfactory, information regarding similar body poses is retrieved from among the already perfectly analyzed skeletal analysis information to estimate analysis in the absence of an obstacle. This is performed by the floor point specification unit if an error occurs in the skeleton analysis unit.

[0065] At this time, a similar body pose is determined by matching the image in which a detected person is judged to have an obstacle and the skeleton value with the highest matching rate among the perfectly analyzed skeleton analysis information that has been learned, and defining the body pose according to the skeleton value as a similar body pose.

[0066] Referring to Figure 2 of the attached drawing again, when a person is detected on the received video or image information and an error occurs in the skeleton analysis while analyzing it through the skeleton analysis unit, it is determined that an obstacle exists, and after obtaining information on a similar body pose by comparing it with previously learned information, the bottom point is estimated by the bottom point specification unit and then specified.

[0067] This has the advantage of preventing obstacles from making it difficult to detect people or reducing the reliability of location information acquisition.

[0068] Meanwhile, as described above, when there are multiple cameras and an intersection area is set, all information acquired from multiple cameras can be used to detect a person, specify a floor point, and track positional displacement.

[0069] At this time, by pre-setting location information based on the x and y axes on the floor plan information, the location information on the floor plan information can be classified, and the floor points mapped to exist within a certain range within the pre-set intersection area can be controlled to be displayed as one.

[0070] The description using the drawings above only describes the main aspects of the present invention, and it is obvious that the present invention is not limited to the configuration of the drawings, as various designs are possible within the technical scope.

Claims

1. At least one camera installed within a specific space; A platform that receives images from the above camera and tracks the displacement of the detected person's position based on the distance from the bottom point of the person detected through image analysis to the camera; including; A system for determining the position of a worker using an AI algorithm, characterized in that when detection of a person is not complete due to a number of obstacles provided in the specific space, the platform estimates the physical pose of the detected person through an AI learning model to determine the floor point.

2. In claim 1, The above platform, An image receiving unit that receives image information or video information from the camera; A floor plan information acquisition unit that obtains floor plan information based on information received from the above image receiving unit and maps camera information to floor plan information based on the actual installation location of the camera; A person detection unit that detects a person on the information received from the above image receiving unit; A skeleton analysis unit that analyzes the left and right feet of a person detected through the above human detection unit; A bottom point specifying unit that specifies the center area of ​​the left and right feet of the person detected and analyzed by the above skeleton analysis unit as the bottom point; A distance measuring unit that measures the distance from the floor point specified by the floor point specific unit to the camera; including; The above floor point specific part is, A worker positioning system using an AI algorithm, characterized in that the floor point is mapped to the floor plan information based on the distance from the camera to the floor point measured through the above distance measuring unit.

3. In claim 2, The above platform, A worker positioning system using an AI algorithm, characterized in that, when there are multiple cameras, the overlapping shooting ranges of each camera are set as intersection areas in the above-mentioned flat map information.

4. In claim 2, The above platform, By tracking the location of the floor point based on the floor point that is continuously mapped to the plan information by the above floor point specific unit, A worker location positioning system using an AI algorithm, characterized in that it further includes a floor point location tracking unit that tracks the movement of a person causing the floor point.

5. In claim 2, The above platform, Information about the right and left feet of the person detected by the above skeleton analysis unit If it is not analyzed, it is assumed that an obstacle exists, The above floor point specific part is, A system for determining the position of a worker using an AI algorithm, characterized in that the bottom point is estimated and specified by comparing it with pre-learned information having a pose similar to that of a person detected in an image where an obstacle is determined to exist.

Images